New in spot 2.14.2  (2025-09-18)

  Library:

  - BuDDy's implementation of bdd_mt_apply1_synthesis() and
    bdd_mt_apply1_synthesis_with_choice() used by ltlfsynt, got some
    small improvements resulting in smaller strategies being
    generated.  (Issue #611.)

  Documentation:

  - https://www.lre.epita.fr/tlsf.html is a new page discussing how
    the --tlsf option of ltlsynt/ltlfsynt/ltlf2dfa uses syfco to
    transform TLSF into LTL/LTLf.

  - https://www.lre.epita.fr/tut60.html shows two examples of
    integration with ppLTLTT to handle some pure-past LTL
    sub-formulas.

  Bug fixes:

  - twadfa_to_mtdfa(aut) was not properly registering the atomic
    proposition of aut into the resulting MTDFA.

  - The part of the HOA parser responsible for dealing with multiple
    initial states (a feature of the HOA format that Spot does not
    support internally) had a bug if the last state of the set of
    initial state without incoming edges also had no outgoing edges.

  - spot::mtdfa_strategy_to_mealy() now has an option to produce a
    strategy that stutter on accepance.  This gets rid of the
    accepting sink, and this is also necessary for the correctness of
    the produced controller in some decomposition cases.  (Issue #610)

New in spot 2.14.1  (2025-07-21)

  Bug fixes:

  - print_hoa() will not use state-based output if the input has been
    tagged as transition-based (i.e. prop_state_acc() == false).  If
    only uses state-based output when prop_state_acc() == true, or if
    prop_state_acc() == maybe and the automaton is found to be
    compatible with a state-based output.

  - ltlfsynt's decomposition optimization (which was implemented in
    the same way as we did in ltlsynt for LTL) isn't always correct
    for LTLf specification.  Until we know and implement a better
    sufficient condition, this version disables decomposition except
    for X-free specifications.  (Issue #610.)

  - Various documentation fixes.

New in spot 2.14  (2025-06-26)

  Build:

  - When Python bindings are enabled, Spot now requires Python 3.8 or
    later.  Python 3.8 is already quite old: it reached end-of-life in
    2024.  Python 3.8 was used for instance in Ubuntu 20.04, where you
    can install this version of Spot.

  Command-line tools:

  - ltlf2dfa is a new tool that converts LTLf to MTDFA, i.e., DFA
    represented with multi-terminal BDDs.   See
    https://spot.lre.epita.fr/ltlf2dfs.html for some examples.

  - ltlfsynt is a new tool that solves the reactive synthesis for LTLf
    specification, using a reachability game interpreted oveer MTDFA.
    See https://spot.lre.epita.fr/ltlfsynt.html

    The output of --realizability has been compared to several other
    tools on benchmarks from the synthesis competition (which all use
    Moore semantics) in addition to some custom specifications (with
    Moore and Mealy semantics) in our test suite.  The production of a
    Mealy machine representing the controller has been implemented,
    but it hasn't received a lot of testing so far, it should be
    regarded as very experimental.

    For more information about ltlf2dfa and ltlfsynt, refer to the
    paper "Engineering an LTLf Synthesis Tool" to be presented at
    CIAA'25.

  - genltl learned four new options: --tv-counter-mealy=N,
    --tv-double-counters-mealy=N, --tv-nim-mealy=N,M,
    --chomp-mealy=N,M.   These are scalable LTLf specifications
    that can be used directly with ltlfsynt.  The first three options
    are taken from the literature; Chomp is an original familly that
    has been submitted as a new model to SyntComp'25.

  - ltlf2tgba learned an --ltlf option.  This currently implements the
    very slow approach of going from LTLf->LTL->DBA->DFA.  ltlf2dfa is
    way faster.

  - Tools that take a "--tlsf FILENAME" option (ltlsynt, ltlfsynt, and
    ltlf2dfa currently) will now accept arguments of the form
    FILENAME/VAR1=VAL1,VAR2=VAL2... to override parameters defined
    in the TLSF file.  E.g.:

        ltlsynt --tlsf arbiter.tlsf/n=3

  Library:

  - spot::mtdfa is a new class implementing a DFA using Multi-Terminal
    BDDs.  The data structure is similar to what can be found in Mona.
    This comes with convenient rendering routines for use in Jupyter
    notebook.

  - spot::ltlf_to_mtdfa() and spot::ltlf_to_mtdfa_compose() are new
    functions that converts a LTLf formula into an MTDFA, using
    either a direct translation, or a compositional one.

  - spot::minimize_mtdfa() is a minimization function for MTDFA.  See
    https://spot.lre.epita.fr/ipynb/ltlf2dfa.html for some illustrations.

  - spot::product_or(), spot::product_and(), spot::product_xor(), etc.
    have been overloaded to implement those operations on MTDFAs as well.

  - spot::backprop_graph is a new class implementing what's needed to
    run a linear backpropagation for solving reachability games.  This
    allows solving the game on-the-fly, as it is constructed.
    See https://spot.lre.epita.fr/ipynb/backprop.html for examples.

  - spot::ltlf_to_mtdfa_for_synthesis() is a new function that
    performs LTLf synthesis.

  - spot::translator learned output_type::Finite to request
    translation to Finite automata.

New in spot 2.13.2  (2025-06-12)

  Bugs fixed:

  - compiling from git with a missing Bison would display
    a confusing error (Issue #607)

  - the Python bindings for iterators generated by Swig are not
    working with libc++ when multiple modules are used.  (Everything
    works with libstdc++.)  This caused issues in Mac OS and OpenBSD.
    This version implements a workaround until Swig is fixed.
    (Issue #603)

  - Fix spurious failure of ltlsynt.test on OpenBSD.

  - difference_word_forq() could return different words on different
    architectures because the code was iterating on hash_map that
    can be implemented in different ways.

  - remove_univ_otf() had a memory-handling bug, that could cause
    use-after-free errors.

New in spot 2.13.1  (2025-05-21)

  Bugs fixed:

  - ltlsynt --print-game did not work with --global-equiv=no --polarity=yes.

  - Work around a couple of compilation issues with GCC 15 and later.

  - Miscelaneous documentation fixes.

New in spot 2.13  (2025-03-24)

  Command-line tools:

  - ltlmix is a new tool that generates formulas by combining existing
    ones. See https://spot.lre.epita.fr/ltlmix.html for examples.

  - ltlf2dfa is a new tool convering LTLf formula to transition-based
    DFAs.  See https://spot.lre.epita.fr/ltlf2dfa.html.

  - randltl can now set the prority of the X[!] operator to generate
    LTLf formulas.   It is disabled by default, but you may
    enable it with  "randltl --ltl-priorities strongX=1 ...".

  - (BACKWARD INCOMPATIBILITY) ltlsynt's CSV output has been largely
    overhauled, and the output columns have been both renamed (for
    consistency) and augmented (with new statistics).  The new CSV
    output should be more useful when the input specification is
    decomposed, in particular, there is a column giving the number of
    sub-specifications obtained, and other statistics columns have
    names starting with "max_" or "sum_" indicating how said
    statistics are updated across sub-specifications.

    See https://spot.lre.epita.fr/ltlsynt.html#csv for an example.

    Additionally, --csv=FILENAME will now save the source filename for
    the processed formulas as the first column of the CSV file, and
    the "formula" column is no longer output by default to save space.
    This "formula" column can be added back with the new option
    --csv-with-formula=FILENAME if really needed.

  - ltlsynt learned a --part-file option, to specify the partition of
    input/output proposition from a *.part file, as used in several
    other tools.

  - ltlfilt learned a --relabel=io mode, that is useful to shorten
    atomic propositions in the context of LTL synthesis.  For instance

    % ltlfilt -f 'G(req->Fack)&G(go->Fgrant)' --relabel=io --ins=req,go
    G(i1 -> Fo0) & G(i0 -> Fo1)

    The resulting formulas are now usable by ltlsynt without having to
    specify which atomic propositions are input or output, as this can
    be inferred from their name.  Instead of using --ins/--outs, a
    --part-file option can be used as well.

  - ltlfilt learned a --save-part-file[=FILENAME] option that can be
    used to create partition files suitable for many synthesis tools
    (including ltlsynt).

  - genltl learned --lily-patterns to generate the example LTL
    synthesis specifications from Lily 1.0.2.  Those come with input
    and output atomic propositions rewriten in the form "iNN" or "oNN",
    so they can be fed to ltlsynt directly, as in

       % genltl --lily-patterns | ltlsynt -q

  - genltl's --pps-arbiter-standard and --pps-arbiter-strict have been
    changed to rename variables {r1,r2,...,g1,g2...} as
    {i1,i2,...,o1,o2,...} so that these formulas can be fed directly to
    ltlsynt.

  - autfilt learned --restrict-dead-end-edges, to restricts labels of
    edges leading to dead-ends.  See the description of
    restrict_dead_end_edges_here() below.

  - autfilt learned --track-formula=F to label states with formulas
    derived from F.  (This is more precise on deterministic automata.)

  - autfilt learned --mcs[=any|scc] to reorder states according to a
    maximum cardinality search.  The argument specifies how to break
    ties.

  - autfilt learned --given-formula, --given-automaton, and
    --given-strategy to simplify automata given some knowledge.
    This is discussed in the PN'25 paper "Simplifying LTL
    Model-Checking Given Prior Knowledge" by Duret-Lutz, Poitrenaud,
    and Thierry-Mieg.

  - ltlfilt learned --pi1, --sigma1, --delta1, --pi2, --sigma2, and
    --delta2 to filter according to classes Π₁,Σ₁,Δ₁,Π₂,Σ₂, and Δ₂.

  - ltlfilt learned --to-delta2 to transform an LTL formula into Δ₂.

  - ltl2tgba learned an experimental --ltlf option that interprets the
    input as an LTLf and produces a Büchi automaton that should be
    interpreted as a finite automaton.
     % ltl2tgba --ltlf FORMULA
    is more than just a shorthand for
     % ltlfilt --from-ltlf -f FORMULA | ltl2tgba | autfilt --to-finite
    because knowing that the input is an LTLf formula allows
    some shortcuts when building the automaton and simplifying it.

  Library:

  - (BACKWARD INCOMPATIBILITY) op::strong_X, the strong next operator
    of LTLf is now defined by default.  Support for this operator was
    added in version 2.9 but its name was not defined publicly by
    default because that would have broken existing code for something
    very niche. Now, since this version is bringing more support for
    LTLf we are removing this special treatement and defining this
    operator publicly.  This is likely to cause compilation warning or
    errors in some user code, for instance when a switch is performed
    over all possible operators: simply add the support for
    op::strong_X.  If you need your code to support older versions of
    Spot, you have two option.

    Option 1: define SPOT_USES_STRONG_X before including formula.hh:

      #define SPOT_USES_STRONG_X 1
      #include <spot/tl/formula.hh>

    This will force any version of Spot since 2.9 to define
    strong_X.  It won't work with older Spot versions, where
    strong_X did not exist.

    Option 2: make any code using strong_X conditional on
    SPOT_HAS_STRONG_X.  Typically, a switch over all possible
    operators would include something like this:

      #if SPOT_HAS_STRONG_X
         case op::strong_X:
            /* do something */
      #endif

    These two options are NOT mutually exclusive.  Setting
    SPOT_USES_STRONG_X will cause SPOT_HAS_STRONG_X to be defined in
    Spot versions ≥ 2.9.

  - restrict_dead_end_edges_here() can reduce non-determinism (but
    not remove it) by restricting the label L of some edge (S)-L->(D)
    going to a state D that does not have other successors than
    itself.  The conditions are detailed in the documentation of
    this function.

  - spot::postprocessor will now call restrict_dead_end_edges_here()
    in its highest setting.  This can be fine-tuned with the "rde"
    extra option; see the spot-x (7) man page for detail.

  - spot::translator and spot::processor now support
       set_type(postprocessor::Finite)
    and
       set_pref(postprocessor::Obligation)
    The first tells that the input should be interpreted as LTLf in order
    to build a Finite automaton.   The second, which is best considered
    as an implementation detail (it will be turned on automatically
    by the translator when necessary), tells the postprocessor that it
    can assumes it is simplifying an obligation property.

  - The formula class now keeps track of membership to the Δ₁, Σ₂,
    Π₂, and Δ₂ syntactic classes.  This can be tested with
    formula::is_delta1(), formula::is_sigma2(), formula::is_pi2(),
    formula::is_delta2().  See doc/tl/tl.pdf from more discussion.

  - spot::to_delta2() implements Δ₂-normalization for LTL formulas,
    following "Efficient Normalization of Linear Temporal Logic" by
    Esparza et al. (J. ACM, 2024).

  - Trivial rewritings (those performed everytime at construction)
    were missing the rule "[*0]|f ≡ f" when f already accepts the
    empty word.  (Issue #545.)

  - spot::match_states(A, B) is a new function that returns a vector
    V such that V[x] contains all states y such that state (x, y) can
    reach an accepting cycle in product(A, B).  In particular, if A
    and B accept the same language, any word accepted by A from state
    x can be accepted in B from some state in V[x].

    That function also has a variant spot::match_states(A, f) where f
    is an LTL formula.  In this case, it returns an array of
    formulas.  If f represents a superset of the language of A, then
    any word accepted by A from state x satisfies V[x].  Related to
    Issue #591.

  - twa_graph::defrag_states(num) no longer requires num[i]≤i; num
    can now describe a permutation of the state numbers.

  - spot::maximum_cardinality_search() and
    spot::maximum_cardinality_search_reorder_here() are new function
    to compute (and apply) an ordering of states based on a maximum
    cardinality search.

  - a new set of functions spot::update_bounds_given(),
    spot::bounds_simplify(), and spot::stutterize_given() implements
    simplification of automata based on apriori knowledge.
    See https://spot.lre.epita.fr/ipynb/given.html for examples.

  Bug fixes:

  - "ltlsynt --aiger -q ..." was still printing the realizability
    status and the AIG circuit; it now does the job silently as
    requested.

  - ltlsynt had a minor memory leak.

New in spot 2.12.2  (2025-01-18)

  Bug fixes:

  - to_finite() was dealing incorrectly with edges that were
    both alive and dead.  (Issue #596.)

  - LaTeX output of the X[!] operator with broken in both
    LaTeX and self-contained LaTeX mode.  (Issue #597)

  - Fix a bug in the AIGER encoding with certain forms of
    global-equivalence.

  - Work around a spurious test failure with Python 3.13.

New in spot 2.12.1  (2024-09-23)

  Bug fixes:

  - Generating random formula without any unary opertors would very
    often create formulas much smaller than asked.

  - The parity game solver, which internally works on "parity max
    odd", but actually accept any type of parity acceptance, could be
    confused by games with "parity min" acceptance using transition
    with several colors (a rather uncommon situation).

  - "ltlsynt ... --print-game --dot=ARGS" was ignoring ARGS.

  - Work around various warnings from g++14.

  - Improved handling of spot-extra/ directory with newer Swig
    versions.  Necessary to recompile Seminator 2 with Swig 4.

New in spot 2.12  (2024-05-16)

  Build:

  - When Python bindings are enabled, Spot now requires Python 3.6 or
    later.  Python 3.6 has reached end-of-life in 2021, but is still
    used on CentOS 7 (which will reach end-of-support later in 2024).

  Documentation:

  - https://spot.lre.epita.fr/tut25.html is a new example showing
    how to print an automaton in the "BA format" (used by Rabbit
    and other tools).

  Command-line tools:

  - In places that accept format strings with '%' sequences, like
    options --stats, --name, or --output, the new '%l' can now be used
    to produce the 0-based serial number of the produced object.  This
    differs from the existing '%L' that is usually related to the line
    number of the input (when that makes sense).  For instance to
    split a file that contains many automata into one file per
    automaton, do

        autfilt input.hoa -o output-%l.hoa

  - For tools that produce automata, using -Hb or --hoa=b will produce
    an HOA file in which aliases are used to form a basis for the
    whole set of labels.  Those aliases are only used when more than
    one atomic proposition is used (otherwise, the atomic proposition
    and its negation is already a basis).  This can help reduce the
    size of large HOA files.

  - autfilt learned --separate-edges, to split the labels of
    the automaton using a basis of disjoint labels.  See
    https://spot.lre.epita.fr/tut25.html for some motivation.

  - autfilt learned --reduce-acceptance-set/--enlarge-acceptance-set
    to heurisitcally remove/add to unnecessary acceptance marks in
    Büchi automata.  (Issue #570)

  - ltlfilt has a new option --relabel-overlapping-bool=abc|pnn that
    will replace boolean subformulas by fresh atomic propositions even
    if those subformulas share atomic propositions.

  - ltlsynt's --ins and --outs options will iterpret any atomic
    proposition surrounded by '/' as a regular expression.
    For intance with

      ltlsynt --ins='/^in/,/env/' --outs=/^out/,/control/' ...

    any atomic proposition that starts with 'in' or contains 'env'
    will be considered as input, and one that starts with 'out'
    or contain 'control' will be considered output.

    By default, if neither --ins nor --outs is given, ltlsynt will
    behave as if --ins='/^[iI]/' and --outs='/^[oO]/ were used.

  - ltlsynt will now check for atomic propositions that always have
    the same polarity in the specification.  When this happens, these
    output APs are replaced by true or false before running the
    synthesis pipeline, and the resulting game, Mealy machine, or
    Aiger circuit is eventually patched to include that constant
    output.  This can be disabled with --polarity=no.

  - ltlsynt will now check for atomic propositions that are specified
    as equivalent.  When this is detected, equivalent atomic
    propositions are replaced by one representative of their class, to
    limit the number of different APs processed by the synthesis
    pipeline.  The resulting game, Mealy machine, or Aiger circuit is
    eventually patched to include the removed APs.  This optimization
    can be disabled with --global-equivalence=no.  As an exception, an
    equivalence between input and output signals (such as G(in<->out))
    will be ignored if ltlsynt is configured to output a game (because
    patching the game a posteriori is cumbersome if the equivalence
    concerns different players).

  - genaut learned two new familes of automata, --cycle-log-nba and
    --cycle-onehot-nba.  They both create a cycle of n^2 states that
    can be reduced to a cycle of n states using reduction based on
    direct simulation.

  Library:

  - split_2step has now multiple ways to split for improved performance.
    The option can be controlled via the synthesis_info::splittype
    - EXPL: explicit splitting of each state as before
    - SEMISYM: The outgoing transition of each state are encoded
      as a bdd; Works better for larger number of input APs
    - FULLYSYM: The automaton is first translated into a
      fully symbolic version, then split.
    - AUTO: Let the heuristic decide what to do.

  - The following new trivial simplifications have been implemented for SEREs:
    - f|[+] = [+] if f rejects [*0]
    - f|[*] = [*] if f accepts [*0]
    - f&&[+] = f if f rejects [*0]
    - b:b[*i..j] = b[*max(i,1)..j]
    - b[*i..j]:b[*k..l] = b[*max(i,1)+max(k,1)-1, j+l-1]

  - The following new trivial simplifications have been implemented
    for PSL operators:
    - {[*]}[]->0 = 0
    - {[*]}<>->1 = 1

  - The HOA parser is a bit smarter when merging multiple initial
    states into a single initial state (Spot's automaton class
    supports only one): it now reuses the edges leaving initial states
    without incoming transitions.

  - The automaton parser has a new option "drop_false_edges" to
    specify whether edges labeled by "false" should be ignored during
    parsing.  It is enabled by default for backward compatibility.

  - spot::bdd_to_cnf_formula() is a new variant of spot::bdd_to_formula()
    that converts a BDD into a CNF instead of a DNF.

  - spot::relabel_bse() has been improved to better deal with more
    cases.  For instance '(a & b & c) U (!c & d & e)' is now
    correctly reduced as '(p0 & p1) U (!p1 & p2)', and
    '((a & !b) | (a->b)) U c' now becomes '1 U c'. (Issue #540.)

  - spot::relabel_overlapping_bse() is a new function that will
    replace boolean subformulas by fresh atomic propositions even if
    those subformulas share atomic propositions.

  - spot::translate() has a new -x option "relabel-overlap=M" that
    augments the existing "relabel-bool=N".  By default, N=4, M=8.
    When the formula to translate has more than N atomic propositions,
    relabel_bse() is first called to attempt to rename non-overlapping
    boolean subexpressions (i.e., no shared atomic proposition) in
    order to reduce the number of atomic proposition, a source of
    exponential explosion in several places of the translation
    pipeline.  This relabel-bool optimization exists since Spot 2.4.
    The new feature is that if, after relabel-bool, the formula still
    has more than M atomic propositions, then spot::translate() now
    attempts to relabel boolean subexpressions even if they have
    overlapping atomic propositions, in an attempt to reduce the
    number of atomic proposition even more.  Doing so has the slightly
    unfortunate side effect of hindering some simplifications (because
    the new atomic propositions hide their interactions), but it
    usually incures a large speedup.  (See Issue #500, Issue #536.)

    For instance on Alexandre's laptop, running
      'ltlsynt --tlsf SPIReadManag.tlsf --aiger'
    with Spot 2.11.6 used to produce an AIG circuit with 48 nodes in
    36 seconds; it now produces an AIG circuit with 53 nodes in only
    0.1 second.

  - spot::contains_forq() is an implementation of the paper "FORQ-Based
    Language Inclusion Formal Testing" (Doveri, Ganty, Mazzocchi;
    CAV'22) contributed by Jonah Romero.

  - spot::contains() still defaults to the complementation-based
    algorithm, however by calling
    spot::containment_select_version("forq") or by setting
    SPOT_CONTAINMENT_CHECK=forq in the environment, the
    spot::contains_forq() implementation will be used instead when
    applicable (inclusion between Büchi automata).

    The above also impacts autfilt's --included-in option.

  - spot::reduce_buchi_acceptance_set_here() and
    spot::enlarge_buchi_acceptance_set_here() will heuristically
    remove/add unnecessary acceptance marks in Büchi automata.
    (Issue #570.)

  - Given a twa_word_ptr W and a twa_ptr A both sharing the same
    alphabet, one can now write W->intersects(A) or A->intersects(W)
    instead of the longuer W->as_automaton()->intersects(A) or
    A->intersects(W->as_automaton()).

  - spot::scc_info has a new option PROCESS_UNREACHABLE_STATES that
    causes it to enumerate even unreachable SCCs.

  - spot::realizability_simplifier is a new class that performs the
    removal of superfluous APs that is now performed by ltlsynt
    (search for --polarity and --global-equivalence above).

  - scc_filter used to reduce automata tagged with the inherently-weak
    property to weak Büchi automata (unless the acceptance was already
    t or co-Büchi).  In cases where the input automaton had no
    rejecting cycle, the Büchi acceptance was overkill: scc_filter
    will now use "t" acceptance.  This change may have unexpected
    consequences in code paths that assume running scc_filter on a
    Büchi automaton will always return a Büchi automaton.  For those,
    a "keep_one_color" option has been added to scc_filter.

  - spot::separate_edges() and spot::edge_separator offers more ways
    to split labels.  See https://spot.lrde.epita.fr/ipynb/split.html

  - ltsmin's interface will now point to README.ltsmin in case an
    error is found while running divine or spins.

  - spot::is_safety_automaton() was generalized to detect any
    automaton for which the acceptance could be changed to "t" without
    changing the language.  In previous versions this function assumed
    weak automata as input, but the documentation did not reflect
    this.

  - spot::remove_alternation() has a new argument to decide whether it
    should raise an exception or return nullptr if it requires more
    acceptance sets than supported.

  - spot::dualize() learned a trick to be faster on states that have
    less outgoing edges than atomic proposition declared on the
    automaton.  spot::remove_alternation(), spot::tgba_powerset(),
    simulation-based reductions, and spot::tgba_determinize() learned
    a similar trick, except it isn't applied at the state level but if
    the entire automaton uses few distinct labels.  These changes may
    speed up the processing of automata with many atomic propositions
    but few distinct labels.  (Issue #566 and issue #568.)

  Backward incompatibilities:

  - spot::dualize() does not call cleanup_acceptance() anymore.  This
    change ensures that the dual of a Büchi automaton will always be a
    co-Büchi automaton.  Previously cleanup_acceptance(), which remove
    unused colors from the acceptance, was sometimes able to simplify
    co-Büchi to "t", causing surprizes.

  - Function minimize_obligation_garanteed_to_work() was renamed to
    minimize_obligation_guaranteed_to_work().  We believe this
    function is only used by Spot currently.

  - Function split_independant_formulas() was renamed to
    split_independent_formulas().  We believe this function is only
    used by Spot currently.

  Python:

  - The spot.automata() and spot.automaton() functions now accept a
    drop_false_edges=False argument to disable the historical behavior
    of ignoring edges labeled by False.

  - Calling aut.highlight_state(s, None) or aut.highlight_edge(e,
    None) may now be used to remove the highlighting color of some
    given state or edge.  Use aut.remove_highlight_states() or
    aut.remove_highlight_edges() to remove all colors. (Issue #554.)

  - Calling aut.get_hight_state(s) or get.highlight_edge(e) will
    return the highlight color of that state/edge or None.

  - Recent version Jupyter Notebook and Jupyter Lab started to render
    SVG elements using <img...> tag to make it easier to copy/paste
    those images.  This breaks several usages, including the
    possibility to have informative tooltips on states and edges (used
    in Spot).  See the following issues for more details.
      https://github.com/jupyter/notebook/issues/7114
      https://github.com/jupyterlab/jupyterlab/issues/10464

    This version of Spot now declares its svg outputs as HTML to
    prevent Jypyter from wrapping them is images.

  Bugs fixed:

  - tgba_determinize()'s use_simulation option would cause it to
    segfault on automata with more than 2^16 SCCs, due to overflows in
    computations of indices in the reachability matrix for SCCs.
    (Issue #541.)  This has been fixed by disabling the use_simulation
    optimization in this case.

  - product_or_susp() and product_susp() would behave incorrectly in
    presence of unsatisifable or universal acceptance conditions that
    were not f or t.  (Part of issue #546.)

  - Several algorithms were incorrectly dealing with the "!weak"
    property.  Because they (rightly) assumed that a weak input would
    produce a weak output, they also wrongly assumed that a non-weak
    output would produce a non-weak output.  Unfortunately removing
    states or edges in a non-weak automaton can lead a weak automaton.
    The incorrect property lead to issue #546.  In addition to fixing
    several algorithms, product() and print_hoa() will now raise an
    exception if automaton with t or f acceptance is declared !weak.
    (This cheap check will not catch all automata incorrectly labeled
    by !weak, but helps detects some issues nonetheless.)

  - The automaton parser forgot to update the list of highlighted
    edges while dropping edges labeled by bddfalse.  (issue #548.)

  - The comparison operators for acceptance condition (==, !=)
    could fail to equate two "t" condition, because we have two ways
    to represent "t": the empty condition, or the empty "Inf({})".

  - Functions complement() and change_parity() could incorrectly read
    or write the unused edge #0.  In the case of complement(), writing
    that edge was usually harmless.  However, in some scenario,
    complement could need to stick a ⓪ acceptance mark on edge #0,
    then the acceptance condition could be simplified to "t", and
    finally change_parity could be confused to find such an accepting
    mark in an automaton that declares no colors, and perform some
    computation on that color that caused it to crash with a "Too many
    acceptance sets used" message.  (issue #552)

  - The detection of terminal automata used did not exactly
    match the definition used in the HOA format.  The definition
    of a terminal automaton is supposed to be:
       1. the automaton is weak
       2. its accepting SCCs are complete
       3. no accepting cycle can reach a rejecting cycle
    However the implementation actually replaced the last point
    by the following variant:
       3'. no accepting edge can reach a rejecting cycle
    This caused issues in automata with t acceptance.  Fixing the code
    by replacing 3' by 3 this also made the third argument of
    is_terminal_automaton(), an optional boolean indicating whether
    transition between SCCs should be ignored when computing 3',
    completely obsolete.  This third argument has been marked as
    depreated.  (Issue #553)

  - twa::defrag_states(), which is called for instance by
    purge_dead_state(), did not update the highlight-edges property.
    (Issue #555.)

  - spot::minimize_obligation will skip attempts to complement very
    weak automata when those would require too many acceptance sets.

  - acd_transform() was not used by spot::postprocessor unless an
    option_map was passed.  This was due to some bad default for the
    "acd" option: it defaulted to true when an option_map was given,
    and to false otherwise.  This had no consequences on
    ltl2tgba/autfilt were some option_map is always passed, but for
    instance the parity automata generated by spot.postprocessor in
    Python were not using ACD by default.

  - Using spot::postprocessor to produce colored parity automata could
    fail to color some transiant edges when the "acd" option was
    activated.

  - The formula printer incorrectly replaced a SERE like "(!a)[*3];a"
    by "a[->]".  The latter should only replace "(!a)[*];a".
    (Issue #559.)

  - The configure script failed to detect the include path for Python 3.12.
    (Issue #577.)

  - Work around many failures caused by incorrect rounding of floating
    point values in the counting of transitions.  (Issue #582)

  - Some incorrectly escaped strings in Python code were causing
    warnings with Python 3.12.

New in spot 2.11.6  (2023-08-01)

  Bug fixes:

  - Running command lines such as "autfilt input.hoa -o output-%L.hoa"
    where thousands of different filenames can be created failed with
    "Too many open files".  (Issue #534)

  - Using --format=... on a tool that output formulas would force
    the output on standard output, even when --output was given.

  - Using "randltl -S" did not correctly go through the SERE printer
    functions.

  - Our copy of BuDDy's bdd_forall() had a 20 year old typo that
    caused cache entries from bdd_exist() and bdd_forall() to be
    mixed.  Spot was safe from this bug because it was only using
    bdd_exist().  (Issue #535)

  - Work around recent Pandas and GCC changes.

New in spot 2.11.5  (2023-04-20)

  Bug fixes:

  - Fix spurious failure of ltlsynt2.test when Python is not installed
    (issue #530).

  - Building from the git repository would fail to report a missing
    emacs (issue #528).

  - Fix exception raised by aut1.intersecting_run(aut2).as_twa()
    because the run did not match transitions present in aut1
    verbatim.  We also changed the behavior of as_twa() to not merge
    identical states.

  - Fix segfaults occuring in determinization of 1-state terminal
    automata.

  - Fix incorrect assertion in game solver when the edge vector
    contains deleted transitions.

New in spot 2.11.4  (2023-02-10)

  Python:

  - spot.acd() no longer depends on jQuery for interactivity.

  Bug fixes:

  - When merging initial states from state-based automata with
    multiple initial states (because Spot supports only one), the HOA
    parser could break state-based acceptance.  (Issue #522.)

  - autfilt --highlight-word refused to work on automata with Fin
    acceptance for historical reasons, however the code has been
    perfectly able to handle this for a while.  (Issue #523.)

  - delay_branching_here(), a new optimization of Spot 2.11 had an
    incorrect handling of states without successors, causing some
    segfaults.  (Issue #524.)

  - Running delay_branching_here() on state-based automata (this was not
    done in Spot so far) may require the output to use transition-based
    acceptance.  (Issue #525.)

  - to_finite(), introduce in 2.11, had a bug that could break the
    completeness of automata and trigger an exception from the HOA
    printer.  (Issue #526.)

New in spot 2.11.3  (2022-12-09)

  Bug fixes:

  - Automata-based implication checks, used to simplify formulas were
    slower than necessary because the translator was configured to
    favor determinism unnecessarily.  (Issue #521.)

  - Automata-based implication checks for f&g and f|g could be
    very slow when those n-ary operator had two many arguments.
    They have been limited to 16 operands, but this value can be changed
    with option -x tls-max-ops=N.  (Issue #521 too.)

  - Running ltl_to_tgba_fm() with an output_aborter (which is done
    during automata-based implication checks) would leak memory on
    abort.

  - configure --with-pythondir should also redefine pyexecdir,
    otherwise, libraries get installed in the wrong place on Debian.
    (Issue #512.)

  - The HOA parser used to silently declare unused and undefined states
    (e.g., when the State: header declare many more states than the body
    of the file).  It now warns about those.

  - 'autfilt -c ...' should display a match count even in presence of
    parse errors.

  - Calling solve_parity_game() multiple times on the same automaton
    used to append the new strategy to the existing one instead of
    overwriting it.

New in spot 2.11.2  (2022-10-26)

  Command-line tools:

  - The --stats specifications %s, %e, %t for printing the number of
    (reachable) states, edges, and transitions, learned to support
    options [r], [u], [a] to indicate if only reachable, unreachable,
    or all elements should be counted.

  Library:

  - spot::reduce_parity() now has a "layered" option to force all
    transition in the same parity layer to receive the same color;
    like acd_transform() would do.

  Bugs fixed:

  - Fix pkg-config files containing @LIBSPOT_PTHREAD@ (issue #520)
  - spot::relabel_bse() was incorrectly relabeling some dependent
    Boolean subexpressions in SERE.  (Note that this had no
    consequence on automata translated from those SERE.)

New in spot 2.11.1  (2022-10-10)

  Bugs fixed:

  - Fix a build issue preventing the update of website (issue #516).
  - Fix a compilation error with clang-14 on FreeBSD (issue #515).

New in spot 2.11  (2022-10-08)

  Build:

  - configure will now diagnose situations where Python bindings will
    be installed in a directory that is not part of Python's search
    path.  A new configure option --with-pythondir can be used to
    modify this installation path.

  - A new configure option --enable-pthread enables the compilation of
    Spot with -pthread, and render available the parallel version of
    some algorithms.  If Spot is compiled with -pthread enabled, any
    user linking with Spot should also link with the pthread library.
    In order to not break existing build setups using Spot, this
    option is currently disabled by default in this release.  We plan
    to turn it on by default in some future release.  Third-party
    project using Spot may want to start linking with -pthread in
    prevision for this change.

  Command-line tools:

  - autfilt has a new options --aliases=drop|keep to specify
    if the HOA printer should attempt to preserve aliases
    present in the HOA input.  This defaults to "keep".

  - autfilt has a new --to-finite option, illustrated on
    https://spot.lrde.epita.fr/tut12.html

  - ltlfilt has a new --sonf option to produce a formula's Suffix
    Operator Normal Form, described in [cimatti.06.fmcad].  The
    associated option --sonf-aps allows listing the newly introduced
    atomic propositions.

  - autcross learned a --language-complemented option to assist in the
    case one is testing tools that complement automata.  (issue #504).

  - ltlsynt has a new option --tlsf that takes the filename of a TLSF
    specification and calls syfco (which must be installed) to convert
    it into an LTL formula.

  - ltlsynt has a new option --from-pgame that takes a parity game in
    extended HOA format, as used in the Synthesis Competition.

  - ltlsynt has a new option --hide-status to hide the REALIZABLE or
    UNREALIZABLE output expected by SYNTCOMP.  (This line is
    superfluous, because the exit status of ltlsynt already indicate
    whether the formula is realizable or not.)

  - ltlsynt has a new option --dot to request GraphViz output instead
    of most output.  This works for displaying Mealy machines, games,
    or AIG circuits.  See https://spot.lrde.epita.fr/ltlsynt.html for
    examples.

  - genaut learned the --cyclist-trace-nba and --cyclist-proof-dba
    options.  Those are used to generate pairs of automata that should
    include each other, and are used to show a regression (in speed)
    present in Spot 2.10.x and fixed in 2.11.

  - genltl learned --eil-gsi to generate a familly a function whose
    translation and simplification used to be very slow.  In particular

       genltl --eil-gsi=23 | ltlfilt --from-ltlf | ltl2tgba

     was reported as taking 9 days.  This is now instantaneous.

  Library:

  - The new function suffix_operator_normal_form() implements
    transformation of formulas to Suffix Operator Normal Form,
    described in [cimatti.06.fmcad].

  - "original-classes" is a new named property similar to
    "original-states".  It maps an each state to an unsigned integer
    such that if two classes are in the same class, they are expected
    to recognize the same language.  The "original-states" should be
    prefered property when that integer correspond to some actual
    state.

  - tgba_determinize() learned to fill the "original-classes" property.
    States of the determinized automaton that correspond to the same
    subset of states of the original automaton belong to the same
    class.  Filling this property is only done on demand as it inccurs
    a small overhead.

  - sbacc() learned to take the "original-classes" property into
    account and to preserve it.

  - The HOA parser and printer learned to map the synthesis-outputs
    property of Spot to the controllable-AP header for the Extended
    HOA format used in SyntComp.   https://arxiv.org/abs/1912.05793

  - The automaton parser learned to parse games in the PGSolver format.
    See the bottom of https://spot.lrde.epita.fr/ipynb/games.html for
    an example.

  - "aliases" is a new named property that is filled by the HOA parser
    using the list of aliases declared in the HOA file, and then used
    by the HOA printer on a best-effort basis.  Aliases can be used to
    make HOA files more compact or more readable.  But another
    possible application is to use aliases to name letters of a 2^AP
    alphabet, in applications where using atomic propositions is
    inconvenient.

  - print_dot() learned option "@" to display aliases, as discussed
    above.

  - to_finite() is a new function that help interpreting automata
    build from LTLf formula using the from_ltlf() function.  It replace
    the previously suggested method of removing and atomic proposition
    and simpifying automata, that failed to deal with states without
    successors.  See updated https://spot.lrde.epita.fr/tut12.html

  - the HOA parser learned to not ignore self-loops labeled with [f]
    and to turn any state that have colors but no outgoing transitions
    into a state with a [f] self-loop.  This helps dealing with
    automata containing states without successors, as in the output of
    to_finite().

  - purge_dead_states() will now also remove edges labeled by false
    (except self-loops).

  - When parsing formulas with a huge number of operands for an n-ary
    operator (for instance 'p1 | p2 | ... | p1000') the LTL parser
    would construct that formula two operand at a time, and the
    formula constructor for that operator would be responsible for
    inlining, sorting, deduplicating, ... all operands at each step.
    This resulted in a worst-than-quadratic slowdown.  This is now
    averted in the parser by delaying the construction of such n-ary
    nodes until all children are known.

  - complement() used to always turn tautological acceptance conditions
    into Büchi.  It now only does that if the automaton is modified.

  - The zielonka_tree construction was optimized using the same
    memoization trick that is used in ACD.  Additionally it can now be
    run with additional options to abort when the tree as an unwanted
    shape, or to turn the tree into a DAG.

  - contains() can now take a twa as a second argument, not just a
    twa_graph.  This allows for instance to do contains(ltl, kripke)
    to obtain a simple model checker (that returns true or false,
    without counterexample).

  - degeneralize() and degeneralize_tba() learned to work on
    generalized-co-Büchi as well.

  - product() learned that the product of two co-Büchi automata
    is a co-Büchi automaton.  And product_or() learned that the
    "or"-product of two Büchi automata is a Büchi automaton.

  - spot::postprocessor has a new extra option "merge-states-min" that
    indicates above how many states twa_graph::merge_states() (which
    perform a very cheap pass to fuse states with identicall
    succesors) should be called before running simulation-based
    reductions.

  - A new function delay_branching_here(aut) can be used to simplify
    some non-deterministic branching.  If two transitions (q₁,ℓ,M,q₂)
    and (q₁,ℓ,M,q₃) differ only by their destination state, and are
    the only incoming transitions of their destination states, then q₂
    and q₃ can be merged (taking the union of their outgoing
    transitions).  This is cheap function is automatically called by
    spot::translate() after translation of a formula to GBA, before
    further simplification.  This was introduced to help with automata
    produced from formulas output by "genltl --eil-gsi" (see above).

  - spot::postprocessor has new configuration variable branch-post
    that can be used to control the use of branching-postponement
    (disabled by default) or delayed-branching (see above, enabled by
    default).  See the spot-x(7) man page for details.

  - spot::postprocessor is now using acd_transform() by default when
    building parity automata.  Setting option "acd=0" will revert
    to using "to_parity()" instead.

  - to_parity() has been almost entirely rewritten and is a bit
    faster.

  - When asked to build parity automata, spot::translator is now more
    aggressively using LTL decomposition, as done in the Generic
    acceptance case before paritizing the result.  This results in
    much smaller automata in many cases.

  - spot::parallel_policy is an object that can be passed to some
    algorithm to specify how many threads can be used if Spot has been
    compiled with --enable-pthread.  Currently, only
    twa_graph::merge_states() supports it.

 Python bindings:

  - The to_str() method of automata can now export a parity game into
    the PG-Solver format by passing option 'pg'.  See
    https://spot.lrde.epita.fr/ipynb/games.html for an example.

 Deprectation notice:

  - spot::pg_print() has been deprecated in favor of spot::print_pg()
    for consistency with the rest of the API.

  Bugs fixed:

  - calling twa_graph::new_univ_edge(src, begin, end, cond, acc) could
    produce unexpected result if begin and end where already pointing
    into the universal edge vector, since the later can be
    reallocated during that process.

  - Printing an alternating automaton with print_dot() using 'u' to
    hide true state could produce some incorrect GraphViz output if
    the automaton as a true state as part of a universal group.

  - Due to an optimization introduces in 2.10 to parse HOA label more
    efficiently, the automaton parser could crash when parsing random
    input (not HOA) containing '[' (issue #509).

New in spot 2.10.6  (2022-05-18)

  Bugs fixed:

  - Fix compilation error on MacOS X.

  - Using -Ffile/N to read column N of a CSV file would not reset the
    /N specification for the next file.

  - make_twa_graph() will now preserve state number when copying a
    kripke_graph object.  As a consequence, print_dot() and
    print_hoa() will now use state numbers matching those of the
    kripke_graph (issue #505).

  - Fix several compilation warning introduced by newer versions
    of GCC and Clang.

New in spot 2.10.5  (2022-05-03)

  Bugs fixed:

  - reduce_parity() produced incorrect results when applied to
    automata with deleted edges.

  - An optimization of Zielonka could result in incorrect results
    in some cases.

  - ltlsynt --print-pg incorrectly solved the game in addition to
    printing it.

  - ltlsynt would fail if only one of --ins or --outs was set, and
    if it was set empty.

  - Work around a portability issue in Flex 2.6.4 preventing
    compilation on OpenBSD.

  - Do not use the seq command in test cases, it is not available
    everywhere.

  - Do not erase the previous contents of the PYTHONPATH environment
    variable when running tests, prepend to it instead.

  - Simplify Debian instructions for LTO build to work around newer
    libtool version.

  - Fix invalid read in digraph::sort_edges_of_(), currently unused in
    Spot.

New in spot 2.10.4  (2022-02-01)

  Bug fixed:

  - Fix memory leaks in Python bindings for several iteration objects.
    This occured while itering on twa_graph.out(), twa_graph.edges(),
    twa_graph.univ_dests(), kripke_graph.out(), kripke_graph.edges(),
    mark_t.sets(), scc_info.edges_of(), scc_info.inner_edges_of(), and
    on an scc_info instance.

New in spot 2.10.3  (2022-01-15)

  Bugs fixed:

  - On automata where the absence of color is not rejecting
    (e.g. co-Büchi) and where the initial state was in a rejecting
    SCC, sbacc() could output a superfluous state.  (Issue #492)

  - Compared to 2.9.8, complement() (and many functions using it) was
    slower and produced larger outputs on some automata with more than
    32 states due to an optimization of the determinization being
    unintentionally disabled.

  - The split_2step() function used to create a synthesis game could
    complain that it was unable to complement a monitor (Issue #495).

  - Work around GraphViz bug #2179 by avoiding unnecessary empty
    lines in the acceptance conditions shown in dot.

  - Fix a case where generic_accepting_run() incorrectly returns a
    cycle around a rejecting self-loop.

  - Mealy machines resulting from SAT-based minimization could differ
    on architectures with different hash tables implementations.

  - org-mode moved to GNU ELPA (Issue #496).

New in spot 2.10.2  (2021-12-03)

  Bugs fixed:

  - twa_graph::purge_dead_states() now also removes edges labeled by
    bddfalse.

  - only use sched_getcpu() and pthread_setaffinity_np() when they are
    available.

  - Using ##300 in a SERE will report that the repeatition exceeds the
    allowed value using a parse error, not as an exception.

  - spot::formula::is_literal() should be const.

New in spot 2.10.1  (2021-11-19)

  Build:

  - Python 3.5 or later is now required (unless Python bindings are
    disabled).  Python 3.5 reached end-of-life one year ago, so we
    believe support for older versions of Python will not be missed.

  Library:

  - The sbacc() function now defines the "original-states" property,
    allowing to map an output state back to its input.

  Bugs fixed:

  - Ranged repetition operators from LTL/PSL, like [*a..b],
    [->a..b] or even F[a..b] used to store and handle a and b
    as 8-bit values without overflow checks.  So for instance
       {a[*260];b} was silently interpreted as {a[*4];b}
    This version still restricts those bounds to 8 bits, but now
    diagnose overflow checks while parsing and constructing formulas.
    Also the so called "trivial simplification rules" will not be
    applied in case they would lead to an overflow.  For instance
       {a;[*150];[*50];b} is rewritten to {a;[*200];b}
    but
       {a;[*150];[*150];b} is untouched.

  - Fix a compilation error on system with glibc older than 2.18,
    where <inttypes.h> does not define some C99 macro by default when
    compiled in C++ mode.  (Such old libraries are used when compiling
    packages for conda-forge, which use CentOS 6 as building environment.)

  - Fix a link error of tests/ltsmin/modelcheck not finding BDD functions
    on some systems (observed on CentOS 6 again).

  - Fix spurious test-suite failures under newer Jupyter
    installations.  IPykernel 6.0 now captures stdout/stderr from
    subprocesses and displays them in the notebook.  The spot.ltsmin
    code that loads modules compiled with SpinS had to be rewritten to
    capture and display the output of SpinS, so that our test-cases
    can be reproduced regardless of the IPykernel version.  This fix
    was easier to do in Python 3.5.

  - Fix sevaral minor documentation errors.

New in spot 2.10  (2021-11-13)

  Build:

  - Spot is now built in C++17 mode, so you need at least GCC 7 or
    clang 5 to compile it.  The current versions of all major linux
    distributions ship with at least GCC 7.4, so this should not be a
    problem.  There is also an --enable-c++20 option to configure in
    case you want to force a build of Spot in C++20 mode.

  Command-line tools:

  - ltlsynt has been seriously overhauled, while trying to
    preserve backward compatibility.  Below is just a summary
    of the main user-visible changes.  Most of the new options
    are discussed on https://spot-dev.lrde.epita.fr/ltlsynt.html

    + ltlsynt now accepts only one of the --ins or --outs options to
      be given and will deduce the value of the other one.

    + ltlsynt learned --print-game-hoa to output its internal parity
      game in the HOA format (with an extension described below).

    + ltlsynt --aiger option now takes an optional argument indicating
      how the bdd and states are to be encoded in the aiger output.
      The option has to be given in the form
      ite|isop|both[+ud][+dc][+sub0|sub1|sub2] where the first only
      obligatory argument decides whether "if-then-else" ("ite") or
      irreducible-sum-of-products ("isop") is to be used.  "both"
      executes both strategies and retains the smaller circuits.  The
      additional options are for fine-tuning. "ud" also encodes the
      dual of the conditions and retains the smaller circuits.  "dc"
      computes if for some inputs we do not care whether the output is
      high or low and try to use this information to compute a smaller
      circuit. "subX" indicates different strategies to find common
      subexpressions, with "sub0" indicating no extra computations.

    + ltlsynt learned --verify to check the computed strategy or aiger
      circuit against the specification.

    + ltlsynt now has a --decompose=yes|no option to specify if
      the specification should be decomposed into independent
      sub-specifications, the controllers of which can then be glued
      together to satisfy the input specification.
      (cf. [finkbeiner.21.nfm] in doc/spot.bib)

    + ltlsynt learned --simplify=no|bisim|bwoa|sat|bisim-sat|bwoa-sat
      to specifies how to simplify the synthesized controler.

  - ltl2tgba, autfilt, dstar2tgba, and randaut learned a --buchi
    option, or -b for short.  This outputs Büchi automata without
    forcing state-based acceptance.

    The following aliases have also been added for consistency:
    --gba  is an alias for --tgba, because the "t" of --tgba is
           never forced.
    --sba  is an alias for -B/--ba, because -B really
           implies -S and that is not clear from --ba.
    As a consequence of the introduction of --sba, the option
    --sbacc (an alias for --state-based-acceptance) cannot be
    abbreviated as --sba anymore.

  - autfilt learned a --kill-states=NUM[,NUM...] option.

  - ltlcross, autcross, ltldo have learned shorthands for
    the commands of Owl 21.0.  For instance running
       ltlcross 'owl-21 ltl2nba' ...
    is equivalent to running
       ltlcross '{owl-21 ltl2nba}owl-21 ltl2nba -f %f>%O' ...

  Library:

  - Spot now provides convenient functions to create and solve games.
    (twaalgos/game.hh) and some additional functions for reactive
    synthesis (twaalgos/synthesis.hh, twaalgos/mealy_machine.hh).

  - A new class called aig represent an and-inverter-graphs, and is
    currently used during synthesis. (twaalgos/aiger.hh)

  - Some new *experimental* classes, called "twacube" and "kripkecube"
    implement a way to store automata or Kripke structures without
    using BDDs as labels.  Instead they use a structure called cube,
    that can only encode a conjunction of litterals.  Avoiding BDDs
    (or rather, avoiding the BuDDy library) in the API for automata
    makes it possible to build multithreaded algorithms.

    /!\ Currently these classes should be considered as experimental,
    and their API is very likely to change in a future version.

    Functions like twacube_to_twa(), twa_to_twacube(), can be used
    to bridge between automata with BDD labels, and automata with
    cube labels.

    In addition to the previous conversion, we implemented various
    state-of-the-art multi-threaded emptiness-check algorithms:
    [bloemen.16.ppopp], [bloemen.16.hvc], [evangeslita.12.atva],
    [renault.13.lpar], [holzmann.11.ieee] (see doc/spot.bib).

    The mc_instanciator class provides an abstraction for launching,
    pinning, and stopping threads inside of parallel model-checking
    algorithms.

    The class ltsmin_model class, providing support for loading
    various models that follow ltsmin's PINS interinterface, has
    been adjusted and can now produce either a kripke instance, or
    a kripkecube instance.

  - The postprocessor::set_type() method can now accept
    options postprocessor::Buchi and postprocessor::GeneralizedBuchi.

    These syntaxes is more homogeneous with the rest of the supported
    types.  Note that postprocessor::BA and postprocessor::TGBA, while
    still supported and not yet marked as deprecated, are best avoided
    in new code.

         postprocessor::set_type(postprocessor::TGBA)

    can be replaced by

         postprocessor::set_type(postprocessor::GeneralizedBuchi)

    and

         postprocessor::set_type(postprocessor::BA)

    by

         postprocessor::set_type(postprocessor::Buchi)
         postprocessor::set_pref(postprocessor::Small
                                 | postprocessor::SBAcc)

    Note that the old postprocessor::BA option implied state-based
    acceptance (and was unique in that way), but the new
    postprocessor::Buchi specifies Büchi acceptance without requiring
    state-based acceptance (something that postprocessor did not
    permit before).

  - Translations for formulas such as FGp1 & FGp2 &...& FGp32 which
    used to take ages are now instantaneous.  (Issue #412.)

  - remove_fin() learned to remove more unnecessary edges by using
    propagate_marks_vector(), both in the general case and in the
    Rabin-like case.

  - simplify_acceptance() learned to simplify acceptence formulas of
    the form Inf(x)&Inf(y) or Fin(x)|Fin(y) when the presence of x
    always implies that of y.  (Issue #406.)

  - simplifiy_acceptance_here() and propagate_marks_here() learned to
    use some patterns of the acceptance condition to remove or add
    (depending on the function) colors on edges.  As an example, with
    a condition such as Inf(0) | Fin(1), an edge labeled by {0,1} can
    be simplied to {0}, but the oppose rewriting can be useful as
    well.  (Issue #418.)

  - The parity_game class has been removed, now any twa_graph_ptr that
    has a "state-player" property is considered to be a game.

  - the "state-player" property is output by the HOA printer, and read
    back by the automaton parser, which means that games can be saved
    and loaded.

  - print_dot() will display states from player 1 using a diamond
    shape.

  - print_dot() learned to display automata that correspond to Mealy
    machines specifically.

  - print_dot() has a new option "i" to define id=... attributes for
    states (S followed by the state number), edges (E followed by the
    edge number), and SCCs (SCC followed by SCC number).  The option
    "i(graphid)" will also set the id of the graph.  These id
    attributes are useful if an automaton is converted into SVG and
    one needs to refer to some particular state/edge/SCC with CSS or
    javascript.  See https://spot.lrde.epita.fr/oaut.html#SVG-and-CSS

  - spot::postproc has new configuration variables that define
    thresholds to disable some costly operation in order to speedup
    the processing of large automata.  (Those variables are typically
    modified using the -x option of command-line tools, and are all
    documented in the spot-x(7) man page.)

      wdba-det-max   Maximum number of additional states allowed in
                     intermediate steps of WDBA-minimization.  If the
                     number of additional states reached in the
                     powerset construction or in the followup products
                     exceeds this value, WDBA-minimization is aborted.
                     Defaults to 4096. Set to 0 to disable.  This limit
                     is ignored when -D used or when det-max-states is
                     set.

      simul-max      Number of states above which simulation-based
                     reductions are skipped.  Defaults to 4096.  Set
                     to 0 to disable.  This also applies to the
                     simulation-based optimization of the
                     determinization algorithm.

      simul-trans-pruning
                     For simulation-based reduction, this is the
                     number of equivalence classes above which
                     transition-pruning for non-deterministic automata
                     is disabled.  Defaults to 512.  Set to 0 to
                     disable.  This applies to all simulation-based
                     reduction, as well as to the simulation-based
                     optimization of the determinization algorithm.
                     Simulation-based reduction perform a number of
                     BDD implication checks that is quadratic in the
                     number of classes to implement transition pruning.

      dpa-simul      Set to 0/1 to disable/enable simulation-based
                     reduction performed after running a Safra-like
                     determinization to optain a DPA.  By default, it is
                     only disabled with --low or if simul=0.

      dba-simul      Set to 0/1 to disable/enable simulation-based
                     reduction performed after running the
                     powerset-like construction (enabled by option
                     tba-det) to obtain a DBA.  By default, it is
                     only disabled with --low or if simul=0.

    spot::translator additionally honor the following new variables:

      tls-max-states  Maximum number of states of automata involved in
                      automata-based implication checks for formula
                      simplifications.  Defaults to 64.

      exprop          When set, this causes the core LTL translation to
                      explicitly iterate over all possible valuations of
                      atomic propositions when considering the successors
                      of a BDD-encoded state, instead of discovering
                      possible successors by rewriting the BDD as a sum of
                      product.  This is enabled by default for --high,
                      and disabled by default otherwise.  When unambiguous
                      automata are required, this option forced and
                      cannot be disabled.  (This feature is not new, but
                      was not tunable before.)

  - In addition the to above new variables, the default value for
    ba-simul (controling how degeneralized automata are reduced) and
    for det-simul (simulation-based optimization of the
    determinization) is now equal to the default value of simul.  This
    changes allows "-x simul=0" to disable all of "ba-simul",
    "dba-simul", "dpa-simul", and "det-simul" at once.

  - tgba_powerset() now takes an extra optional argument to specify a
    list of accepting sinks states if some are known.  Doing so can
    cut the size of the powerset automaton by 2^|sinks| in favorable
    cases.

  - twa_graph::merge_edges() had its two passes swapped.  Doing so
    improves the determinism of some automata.

  - twa_graph::kill_state(num) is a new method that removes the
    outgoing edge of some state.  It can be used together with
    purge_dead_states() to remove selected states.

  - The new functions reduce_direct_sim(), reduce_direct_cosim() and
    reduce_iterated() (and their state based acceptance version,
    reduce_direct_sim_sba(), reduce_direct_cosim_sba() and
    reduce_iterated_sba()), are matrix-based implementation of
    simulation-based reductions.   This new version is faster
    than the signature-based BDD implementation in most cases,
    however there are some cases it does not capture yet.. By
    default, the old one is used. This behavior can be changed by
    setting SPOT_SIMULATION_REDUCTION environment variable or using
    the "-x simul-method=..." option (see spot-x(7)).

  - The automaton parser will now recognize lines of the form
      #line 10 "filename"
    that may occur *between* the automata of an input stream to specify
    that error messages should be emitted as if the next automaton
    was read from "filename" starting on line 10.  (Issue #232)

  - The automaton parser for HOA is now faster at parsing files where
    a label is used multiple times (i.e., most automata): it uses a
    hash table to avoid reconstructing BDDs corresponding to label
    that have already been parsed.  This trick was already used in the
    never claim parser.

  - spot::twa_graph::merge_states() will now sort the vector of edges
    before attempting to detect mergeable states.  When merging states
    with identical outgoing transitions, it will now also consider
    self-looping transitions has having identical destinations.
    Additionally, this function now returns the number of states that
    have been merged (and therefore removed from the automaton).

  - spot::zielonka_tree and spot::acd are new classes that implement the
    Zielonka Tree and Alternatic Cycle Decomposition, based on a paper
    by Casares et al. (ICALP'21).  Those structures can be used to
    paritize any automaton, and more.  The graphical rendering of ACD
    in Jupyter notebooks is Spot's first interactive output.  See
    https://spot.lrde.epita.fr/ipynb/zlktree.html for more.

  Python:

  - Bindings for functions related to games.
    See https://spot.lrde.epita.fr/ipynb/games.html and
    https://spot.lrde.epita.fr/tut40.html for examples.

  - Bindings for functions related to synthesis and aig-circuits.
    See https://spot.lrde.epita.fr/ipynb/synthesis.html

  - spot::twa::prop_set was previously absent from the Python
    bindings, making it impossible to call make_twa_graph() for copying
    a twa.  It is now available as spot.twa_prop_set (issue #453).
    Also for convenience, the twa_graph.__copy__() method, called by
    copy.copy(), will duplicate a twa_graph (issue #470).

  Bugs fixed:

  - tgba_determinize() could create parity automata using more colors
    than necessary.  (Related to issue #298)

  - There were two corners cases under which sat_minimize() would
    return the original transition-based automaton when asked to
    produce state-based output without changing the acceptance
    condition.

  Deprecation notices:

  - twa_graph::defrag_states() and digraph::defrag_states() now take
    the renaming vector by reference instead of rvalue reference.  The
    old prototype is still supported but will emit a deprecation
    warning.

New in spot 2.9.8 (2021-08-10)

  Documentation:

  - Mention the conda-forge package.

  Bugs fixed:

  - left->intersacting_run(right) could return a run with incorrect
    colors (likely not corresponding to any existing transition of
    left) if left was a weak automaton.  (Issue #471)

  - Fix bitset::operator-() causing issues when Spot was configured
    with more than 32 colors for instance with --enable-max-sets=64.
    (Issue #469)

  - Work around some warnings from newer compilers.


New in spot 2.9.7 (2021-05-12)

  Bugs fixed:

  - Some formulas using ->, <->, or xor were not properly detected as
    purely universal or pure eventualities.

  - autfilt --keep-states=... could incorrectly mention --mask-acc
    when diagnosing errors.

  - Work around GraphViz issue 1931, causing spurious failures in the
    test suite.

  - Miscelaneous documentation fixes.

New in spot 2.9.6 (2021-01-18)

  Build:

  - Build scripts are now compatible with Autoconf 2.70 (Issue #447)
    and require at least Autoconf 2.69 (released in 2012).

  - Debian builds use updated standards.

  Bugs fixed:

  - twa_graph::merge_edges() could fail to merge two transitions if the
    destination transition was the first of the automaton.  (Issue #441)

  - On non-deterministic automata, iterated_simulations() was
    performing an (unnecessary) second iteration even when the first
    one failed to reduce the automaton.  (Issue #442)

  - When passed an incomplete automaton as input, tgba_determinize()
    would sometimes produce a complete automaton but incorrectly mark
    it as incomplete.  (Related to issue #298)

  - gf_guarantee_to_ba(), used to translate some specific classes of
    subformulas, could segfault in some condition (Issue #449).

  - Work around some spurious test suite failures.

New in spot 2.9.5 (2020-11-19)

  Bugs fixed:

  - Fix multiple spurious test suite failures on Darwin
    (Issues #426, #427, #428, #429) or when the
    Pandas package was not installed.

  - The filename python/spot/aux.py caused a problem on Windows and
    has been renamed.  Existing "import spot.aux" statements should
    *not* be updated.  (Issue #437)

  - Fix memory leak in minimize_wdba() when determinization is aborted
    because an output_aborter is passed.

  - Distribution used to contain unnecessary large SVG files (Issue #422)

New in spot 2.9.4 (2020-09-07)

  Bugs fixed:

  - Handle xor and <-> in a more natural way when translating
    LTL formulas to generic acceptance conditions.

  - Multiple typos in documentation, --help texts, or error messages.

New in spot 2.9.3 (2020-07-22)

  Bug fixed:

  - Completely fix the ltlcross issue mentioned in previous release.

New in spot 2.9.2 (2020-07-21)

  Bugs fixed:

  - ltlcross --csv=... --product=N with N>0 could output spurious
    diagnostics claiming that words were rejected when evaluating
    the automaton on state-space.

  - spot::scc_info::determine_unknown_acceptance() now also update the
    result of spot::scc_info::one_accepting_scc().

New in spot 2.9.1 (2020-07-15)

  Command-line tools:

  - 'ltl2tgba -G -D' is now better at handling formulas that use
    '<->' and 'xor' operators at the top level.   For instance
         ltl2tgba -D -G '(Fa & Fb & Fc & Fd) <-> GFe'
    now produces a 16-state automaton (instead of 31 in Spot 2.9).

  - Option '-x wdba-minimize=N' used to accept N=0 (off), or N=1 (on).
    It can now take three values:
      0: never attempt this optimization,
      1: always try to determinize and minimize automata as WDBA,
         and check (if needed) that the result is correct,
      2: determinize and minimize automata as WDBA only if it is known
         beforehand (by cheap syntactic means) that the result will be
         correct (e.g., the input formula is a syntactic obligation).
    The default is 1 in "--high" mode, 2 in "--medium" mode or in
    "--deterministic --low" mode, and 0 in other "--low" modes.

  - ltlsynt learned --algo=ps to use the default conversion to
    deterministic parity automata (the same as ltl2tgba -DP'max odd').

  - ltlsynt now has a -x option to fine-tune the translation.  See the
    spot-x(7) man page for detail.  Unlike ltl2tgba, ltlsynt defaults
    to simul=0,ba-simul=0,det-simul=0,tls-impl=1,wdba-minimize=2.

  Library:

  - product_xor() and product_xnor() are two new versions of the
    synchronized product.  They only work with operands that are
    deterministic automata, and build automata whose language are
    respectively the symmetric difference of the operands, and the
    complement of that.

  - tl_simplifier_options::keep_top_xor is a new option to keep Xor
    and Equiv operator that appear at the top of LTL formulas (maybe
    below other Boolean or X operators).  This is used by the
    spot::translator class when creating deterministic automata with
    generic acceptance.

  - remove_fin() learned a trick to remove some superfluous
    transitions.

  Bugs fixed:

  - Work around undefined behavior reported by clang 10.0.0's UBsan.

  - spot::twa_sub_statistics was very slow at computing the number of
    transitons, and could overflow.  It is now avoiding some costly
    loop of BDD operations, and storing the result using at least 64
    bits.  This affects the output of "ltlcross --csv" and
    "--stats=%t" for many tools.

  - simplify_acceptance() missed some patterns it should have been
    able to simplify.  For instance Fin(0)&(Fin(1)|Fin(2)|Fin(4))
    should simplify to Fin(1)|Fin(2)|Fin(4) after adding {1,2,4} to
    every place where 0 occur, and then the acceptance would be
    renumbered to Fin(0)|Fin(1)|Fin(2).  This is now fixed.

  - Improve ltldo diagnostics to fix spurious test-suite failure on
    systems with antique GNU libc.

  - twa_run::reduce() could reduce accepting runs incorrectly on
    automata using Fin in their acceptance condition.  This caused
    issues in intersecting_run(), exclusive_run(),
    intersecting_word(), exclusive_word(), which all call reduce().

  - ltlcross used to crash with "Too many acceptance sets used.  The
    limit is 32." when complementing an automaton would require more
    acceptance sets than supported by Spot.  This is now diagnosed
    with --verbose, but does not prevent ltlcross from continuing.

  - scc_info::edges_of() and scc_info::inner_edges_of() now correctly
    honor any filter passed to scc_info.

New in spot 2.9 (2020-04-30)

  Command-line tools:

  - When the --check=stutter-sensitive-example option is passed to
    tools like ltl2tgba, autfilt, genaut, or ltldo, the produced
    automata are checked for stutter-invariance (as in the
    --check=stutter-invariant case), additionally a proof of
    stutter-sensitiveness is provided as two stutter-equivalent words:
    one accepted, and one rejected.  These sample words are printed in
    the HOA output.

    % ltl2tgba --check=stutter-sensitive-example Xa | grep word:
    spot-accepted-word: "!a; cycle{a}"
    spot-rejected-word: "!a; !a; cycle{a}"

  - autfilt learned the --partial-degeneralize option, to remove
    conjunctions of Inf, or disjunctions of Fin that appears in
    arbitrary conditions.

  - ltlfilt and autfilt learned a --nth=RANGE (a.k.a. -N) option to
    select a range of their input formulas or automata (assuming a
    1-based numbering).

  - When running translators, ltlcross will now display {names} when
    supplied.

  - ltlcross is now using the generic emptiness check procedure
    introduced in Spot 2.7, as opposed to removing Fin acceptance
    before using a classical emptiness check.

  - ltlcross learned a --save-inclusion-products=FILENAME option.  Its
    use cases are probably quite limited.  We use that to generate
    benchmarks for our generic emptiness check procedure.

  - ltlsynt --algo=lar uses the new version of to_parity() mentionned
    below.  The old version is available via --algo=lar.old

  - ltlsynt learned --csv=FILENAME, to record some statistics about
    the duration of its different phases.

  - The dot printer is now automatically using rectangles with rounded
    corners for automata states if one state label have five or more
    characters.  This saves space with very long labels.  Use --dot=c,
    --dot=e, or --dot=E to force the use of Circles, Ellipses, or
    rEctangles.

  Library:

  - Historically, Spot only supports LTL with infinite semantics
    so it had automatic simplifications reducing X(1) and X(0) to
    1 and 0 whenever such formulas are constructed.  This
    caused issues for users of LTLf formulas, where it is important
    to distinguish a "weak next" (for which X(1)=1 but X(0)!=0) from
    a "strong next" (for which X(0)=0 but X(1)!=1).

    To accommodate this, this version introduces a new operator
    op::strong_X in addition to the existing op::X (whose
    interpretation is now weak in LTLf).  The syntax for the strong
    next is X[!] as a reference to the PSL syntax (where the strong
    next is written X!).

    Trivial simplification rules for X are changed to just
       X(1) = 1       (and not X(0)=0 anymore)
    while we have
       X[!]0 = 0

    The X(0)=0 and X[!]1=1 reductions are now preformed during LTL
    simplification, not automatically.  Aside from the from_ltlf()
    function, the other functions of the library handle X and X[!] in
    the same way, since there is no difference between X and X[!] over
    infinite words.

    Operators F[n:m!] and G[n:m!] are also supported as strong
    variants of F[n:m] and G[n:m], but those four are only implemented
    as syntactic sugar.

  - acc_cond::acc_code::parity(bool, bool, int) was not very readable,
    as it was unclear to the reader what the boolean argument meant.
    The following sister functions can now improve readability:
       parity_max(is_odd, n) = parity(true, is_odd, n)
       parity_max_odd(n) = parity_max(true, n)
       parity_max_even(n) = parity_max(false, n)
       parity_min(is_odd, n) = parity(false, is_odd, n)
       parity_min_odd(n) = parity_min(true, n)
       parity_min_even(n) = parity_min(false, n)

  - partial_degeneralize() is a new function performing partial
    degeneralization to get rid of conjunctions of Inf terms, or
    disjunctions of Fin terms in acceptance conditions.

  - simplify_acceptance_here() and simplify_acceptance() learned to
    simplify subformulas like Fin(i)&Fin(j) or Inf(i)|Inf(j), or some
    more complex variants of those.  If i is uniquely used in the
    acceptance condition, these become respectively Fin(i) or Inf(i),
    and the automaton is adjusted so that i also appears where j
    appeared.

  - acc_code::unit_propagation() is a new method for performing unit
    propagation in acceptance condition.  E.g.  Fin(0) | (Inf(0) &
    Inf(1)) becomes Fin(0) | Inf(1).  This is now called by
    simplify_acceptance_here().

  - propagate_marks_vector() and propagate_marks_here() are helper
    functions for propagating marks on the automaton: ignoring
    self-loops and out-of-SCC transitions, marks common to all the
    input transitions of a state can be pushed to all its outgoing
    transitions, and vice-versa.  This is repeated until a fix point
    is reached.  propagate_marks_vector() does not modify the
    automaton and returns a vector of the acc_cond::mark_t that should
    be on each transition; propagate_marks_here() actually modifies
    the automaton.

  - rabin_to_buchi_if_realizable() is a new variant of
    rabin_to_buchi_maybe() that converts a Rabin-like automaton into a
    Büchi automaton only if the resulting Büchi automaton can keep the
    same transition structure (where the ..._maybe() variant would
    modify the Rabin automaton if needed).

  - to_parity() has been rewritten. It now combines several strategies
    for paritizing automata with any acceptance condition.

  - relabel_bse(), used by ltlfilt --relabel-bool, is now better at
    dealing with n-ary operators and isolating subsets of operands
    that can be relabeled as a single term.

  - print_dot()'s default was changed to use circles for automata with
    fewer than 10 unamed states, ellipses for automata with up to 1000
    unamed states (or named states with up to 4 characters), and
    rounded rectangles otherwise.  Rectangles are also used for
    automata with acceptance bullets on states.  The new "E" option
    can be used to force rectangles in all situations.

  - The generic emptiness check has been slightly improved (doing
    fewer recursive calls in the worst case).

  Backward-incompatible changes:

  - iar() and iar_maybe() have been moved from
    spot/twaalgos/rabin2parity.hh spot/twaalgos/toparity.hh and marked
    as deprecated, they should be replaced by to_parity().  In case
    the input is Rabin-like or Streett-like, to_parity() should be at
    least as good as iar().

  - The twa_graph::is_alternating() and digraph::is_alternating() methods,
    deprecated in Spot 2.3.1 (2017-02-20), have been removed.

  Bugs fixed:

  - Relabeling automata could introduce false edges.  Those are now
    removed.

  - Emptiness checks, and scc_info should now ignore edges labeled
    with false.

  - relabel_bse() could incorrectly relabel Boolean subformulas that
    had some atomic propositions in common.

New in spot 2.8.7 (2020-03-13)

  Bugs fixed:

  - Building a product between two complete automata where one operand
    had false acceptance could create an incomplete automaton
    incorrectly tagged as complete, causing the print_hoa() function
    to raise an exception.

  - autfilt --uniq was not considering differences in acceptance
    conditions or number of states when discarding automata.

  - In an automaton with acceptance condition containing some Fin, and
    whose accepting cycle could be reduced to a self-loop in an
    otherwise larger SCC, the generation of an accepting run could be
    wrong.  This could in turn cause segfaults or infinite loops while
    running autcross or autfilt --stats=%w.

  - The generic emptiness check used a suboptimal selection of "Fin"
    to remove, not matching the correct line in our ATVA'19 paper.
    This could cause superfluous recursive calls, however benchmarks
    have shown the difference to be insignificant in practice.

  - The sl(), and sl2() functions for computing the "self-loopization"
    of an automaton, and used for instance in algorithms for computing
    proof of stutter-sensitiveness (e.g., in our web application),
    were incorrect when applied on automata with "t" acceptance (or
    more generaly, any automaton where a cycle without mark is
    accepting).

  - ltlcross was not diagnosing write errors associated to
    options --grind=FILENAME and --save-bogus=FILENAME.

New in spot 2.8.6 (2020-02-19)

  Bugs fixed:

  - calling spot.translate() with two conflicting preferences like
    spot.translate(..., 'det', 'any') triggered a syntax error in the
    Python code to handle this error.

  - degeneralize_tba() was incorrectly not honnoring the "skip_level"
    optimization after creating an accepting transition; as a
    consequence, some correct output could be larger than necessary
    (but still correct).

  - Some formulas with Boolean sub-formulas equivalent to true or
    false could be translated into automata with false labels.

  - The HOA printer would incorrectly output the condition such
    'Fin(1) & Inf(2) & Fin(0)' as 'Fin(0) | Fin(1) & Inf(2)'
    due to a bug in the is_generalized_rabin() matching function.

New in spot 2.8.5 (2020-01-04)

  Bugs fixed:

  - ltl2tgba -B could return automata with "t" acceptance, instead
    of the expected Inf(0).

  - twa_graph::merge_edges() (a.k.a. autfilt --merge-transitions) was
    unaware that merging edges can sometimes transform a
    non-deterministic automaton into a deterministic one, causing the
    following unexpected diagnostic:
    "print_hoa(): automaton is universal despite prop_universal()==false"
    The kind of non-deterministic automata where this occurs is not
    naturally produced by any Spot algorithm, but can be found for
    instance in automata produced by Goal 2015-10-18.

New in spot 2.8.4 (2019-12-08)

  Bugs fixed:

  - The Rabin-to-Büchi conversion could misbehave when applied to
    Rabin-like acceptance with where some pairs have missing Inf(.)
    (e.g. Fin(0)|(Inf(1)&Fin(2))) and when some of the SCCs do not
    visit the remaining Inf(.).  This indirectly caused the
    complementation algorithm to produce incorrect results on such
    inputs, causing false positives in ltlcross and autcross.

  - Work around a small difference between Python 3.7 and 3.8, causing
    spurious failures of the test suite.

New in spot 2.8.3 (2019-11-06)

  Build:

  - Minor portabilities improvements with C++17 deprecations.

  Python:

  - Doing "import spot.foo" will now load any spot-extra/foo.py on
    Python's search path.  This can be used to install third-party
    packages that want to behave as plugins for Spot.

  - Work around GraphViz bug #1605 in Jupyter notebooks.
    https://gitlab.com/graphviz/graphviz/issues/1605

New in spot 2.8.2 (2019-09-27)

  Command-line tools:

  - ltl2tgba and ltldo learned a --negate option.

  Bugs fixed:

  - Calling "autfilt --dualize" on an alternating automaton with
    transition-based acceptance and universal initial states would
    fail with "set_init_state() called with nonexisting state".

  - The numbering of nodes in the AIGER output of ltlsynt was
    architecture dependent.

  - Various compilation issues.  In particular, this release is the
    first one that can be compiled (and pass tests) on a Raspberry PI.

New in spot 2.8.1 (2019-07-18)

  Command-line tools:

  - genltl learned --pps-arbiter-standard and --pps-arbiter-strict.

  - ltlcross and autcross have learned a new option -q (--quiet) to
    remain quiet until an error is found.  This helps for instance in
    scenarios where multiple instances of ltlcross/autcross are run in
    parallel (using "xargs -P" or "GNU Parallel", for instance).  See
    https://spot.lrde.epita.fr/ltlcross.html#parallel for examples.

  Bugs fixed:

  - When complement() was called with an output_aborter, it could
    return an alternating automaton on large automata.  This in turn
    caused ltlcross to emit errors like "remove_alternation() only
    works with weak alternating automata" or "product() does not
    support alternating automata".

  - Work around Emacs bug #34341 when rebuilding documentation on a
    system with Emacs <26.3, GNU TLS >= 3.6, and without ESS or with
    an obsolete org-mode installed.

New in spot 2.8 (2019-07-10)

  Command-line tools:

  - autfilt learned --highlight-accepting-run=NUM to highlight some
    accepting run with color NUM.

  - ltldo, ltlcross, and autcross are now preferring posix_spawn()
    over fork()+exec() when available.

  - ltlcross has new options --determinize-max-states=N and
    --determinize-max-edges=M to restrict the use of
    determinization-based complementation to cases where it produces
    automata with at most N states and M edges.  By default
    determinization is now attempted up to 500 states and 5000
    edges.  This is an improvement over the previous default where
    determinization-based complementation was not performed at all,
    unless -D was specified.

  - ltlcross will now skip unnecessary cross-checks and
    consistency-checks (they are unnecessary when all automata
    could be complemented and statistics were not required).

  - genaut learned --m-nba=N to generate Max Michel's NBA family.
    (NBAs with N+1 states whose determinized have at least N! states.)

  - Due to some new simplification of parity acceptance, the output of
    "ltl2tgba -P -D" is now using a minimal number of colors.  This
    means that recurrence properties have an acceptance condition
    among "Inf(0)", "t", or "f", and persistence properties have an
    acceptance condition among "Fin(0)", "t", or "f".

  - ltldo and ltlcross learned shorthands to call the tools ltl2na,
    ltl2nba, and ltl2ngba from Owl 19.06.  Similarly, autcross learned
    a shorthand for Owl's dra2dpa.

  Documentation:

  - https://spot.lrde.epita.fr/tut90.html is a new file that explains
    the purpose of the spot::bdd_dict object.

  Library:

  - Add generic_accepting_run() as a variant of generic_emptiness_check() that
    returns an accepting run in an automaton with any acceptance condition.

  - twa::accepting_run() and twa::intersecting_run() now work on
    automata using Fin in their acceptance condition.

  - simulation-based reductions have learned a trick that sometimes
    improve transition-based output when the input is state-based.
    (The automaton output by 'ltl2tgba -B GFa | autfilt --small' now
    has 1 state instead of 2 in previous versions.  Similarly,
    'ltldo ltl2dstar -f 'GFa -> GFb' | autfilt --small' produces 1
    state instead of 4.)

  - simulation-based reductions has learned another trick to better
    merge states from transient SCCs.

  - acc_cond::top_disjuncts() and acc_cond::top_conjuncts() can be
    used to split an acceptance condition on the top-level & or |.
    These methods also exist in acc_cond::acc_code.

  - minimize_obligation() learned to work on very weak automata even
    if the formula or negated automaton are not supplied.  (This
    allows "autfilt [-D] --small" to minimize very-weak automata
    whenever they are found to represent obligation properties.)

  - There is a new spot::scc_and_mark_filter object that simplifies the
    creation of filters to restrict spot::scc_info to some particular
    SCC while cutting new SCCs on given acceptance sets.  This is used
    by spot::generic_emptiness_check() when processing SCCs
    recursively, and makes it easier to write similar code in Python.

  - print_dot has a new option 'g', to hide edge labels.  This is
    helpful to display automata as "graphs", e.g., when illustrating
    algorithms that do not care about labels.

  - A new complement() function returns complemented automata with
    unspecified acceptance condition, allowing different algorithms to
    be used.  The output can be alternating only if the input was
    alternating.

  - There is a new class output_aborter that is used to specify
    upper bounds on the size of automata produced by some algorithms.
    Several functions have been changed to accept an output_aborter.
    This includes:
      * tgba_determinize()
      * tgba_powerset()
      * minimize_obligation()
      * minimize_wdba()
      * remove_alternation()
      * product()
      * the new complement()
      * the postprocessor class, via the "det-max-state" and
        "det-max-edges" options.

  - SVA's first_match operator can now be used in SERE formulas and
    that is supported by the ltl_to_tgba_fm() translation.  See
    doc/tl/tl.pdf for the semantics.  *WARNING* Because this adds a
    new operator, any code that switches over the spot::op type may
    need a new case for op::first_match.  Furthermore, the output of
    "randltl --psl" will be different from previous releases.

  - The parser for SERE learned to recognize the ##n and ##[i:j]
    operators from SVA.  So {##2 a ##0 b[+] ##1 c ##2 e} is another
    way to write {[*2];a:b[+];c;1;e}.  The syntax {a ##[i:j] b} is
    replaced in different ways depending on the values of i, a, and b.
    The formula::sugar_delay() function implements this SVA operator in
    terms of the existing PSL operators.  ##[+] and ##[*] are sugar
    for ##[1:$] and ##[0:$].

  - The F[n:m] and G[n:m] operators introduced in Spot 2.7 now
    support the case where m=$.

  - spot::relabel_apply() makes it easier to reverse the effect
    of spot::relabel() or spot::relabel_bse() on formula.

  - The LTL simplifier learned the following rules:
    F(G(a | Fb)) = FGa | GFb  (if option "favor_event_univ")
    G(F(a | Gb)) = GFa | FGb  (if option "favor_event_univ")
    F(G(a & Fb) = FGa & GFb   (unless option "reduce_size_strictly")
    G(F(a & Gb)) = GFa & FGb  (unless option "reduce_size_strictly")
    GF(f) = GF(dnf(f))        (unless option "reduce_size_strictly")
    FG(f) = FG(cnf(f))        (unless option "reduce_size_strictly")
    (f & g) R h = f R h       if h implies g
    (f & g) M h = f M h       if h implies g
    (f | g) W h = f W h       if g implies h
    (f | g) U h = f U h       if g implies h
    Gf | F(g & eventual) = f W (g & eventual)    if !f implies g
    Ff & G(g | universal) = f M (g | universal)  if f implies !g
    f U (g & eventual) = F(g & eventual)         if !f implies g
    f R (g | universal) = G(g | universal)       if f implies !g

  - cleanup_parity() and colorize_parity() were cleaned up a bit,
    resulting in fewer colors used in some cases.  In particular,
    colorize_parity() learned that coloring transient edges does not
    require the introduction of a new color.

  - A new reduce_parity() function implements and generalizes the
    algorithm for minimizing parity acceptance by Carton and Maceiras
    (Computing the Rabin index of a parity automaton, 1999).  This is
    a better replacement for cleanup_parity() and colorize_parity().
    See https://spot.lrde.epita.fr/ipynb/parity.html for examples.

  - The postprocessor and translator classes are now using
    reduce_parity() for further simplifications.

  - The code for checking recurrence/persistence properties can also
    use the fact the reduce_parity() with return "Inf(0)" (or "t" or
    "f") for deterministic automata corresponding to recurrence
    properties, and "Fin(0)" (or "t" or "f") for persistence
    properties.  This can be altered with the SPOT_PR_CHECK
    environment variable.

  Deprecation notices:

  - The virtual function twa::intersecting_run() no longer takes a
    second "from_other" Boolean argument.  This is a backward
    incompatibility only for code that overrides this function in a
    subclass.  For backward compatibility with programs that simply
    call this function with two argument, a non-virtual version of the
    function has been introduced and marked as deprecated.

  - The spot::acc_cond::format() methods have been deprecated.  These
    were used to display acceptance marks, but acceptance marks are
    unrelated to acceptance conditions, so it's better to simply print
    marks with operator<< without this extra step.

  Bugs fixed:

  - The gf_guarantee_to_ba() is relying on an in-place algorithm that
    could produce a different number of edges for the same input in
    two different transition order.

  - A symmetry-based optimization of the LAR algorithm performed in
    spot::to_parity() turned out to be incorrect.  The optimization
    has been removed.  "ltlsynt --algo=lar" is the only code using
    this function currently.

New in spot 2.7.5 (2019-06-05)

  Build:

  - Although the Python bindings in this release are still done with
    Swig3.0, the code has been updated to be compatible with Swig4.0.

  Library:

  - print_dot will replace labels that have more 2048 characters by a
    "(label too long)" string.  This works around a limitation of
    GraphViz that aborts when some label exceeds 16k characters, and
    also helps making large automata more readable.

  Bugs fixed:

  - spot::translator was not applying Boolean sub-formula rewriting
    by default unless a spot::option_map was passed.  This caused some
    C++ code for translating certain formulas to be noticeably slower
    than the equivalent call to the ltl2tgba binary.

  - The remove_ap algorithm was preserving the "terminal property" of
    automata, but it is possible that a non-terminal input produces a
    terminal output after some propositions are removed.

New in spot 2.7.4 (2019-04-27)

  Bugs fixed:

  - separate_sets_here() (and therefore autfilt --separate-sets) could
    loop infinitely on some inputs.

  - In some situation, ltl2tgba -G could abort with
    "direct_simulation() requires separate Inf and Fin sets".  This
    was fixed by teaching simulation-based reductions how to deal
    with such cases.

  - The code for detecting syntactically stutter-invariant PSL
    formulas was incorrectly handling the ";" operator, causing some
    stutter-sensitive formulas to be flagged a stutter-invariant.

New in spot 2.7.3 (2019-04-19)

  Bugs fixed:

  - When processing CSV files with MSDOS-style \r\n line endings,
    --stats would output the \r as part of the %> sequence instead
    of ignoring it.

  - Fix serious typo in remove_alternation() causing incorrect
    output for some VWAA.  Bug introduced in Spot 2.6.

  Documentation:

  - Multiple typos and minor updates.

New in spot 2.7.2 (2019-03-17)

  Python:

  - Improved support for explicit Kripke structures.  It is now
    possible to iterate over a kripke_graph object in a way similar to
    twa_graph.

  Documentation:

  - A new page shows how to create explicit Kripke structures in C++
    and Python.  See https://spot.lrde.epita.fr/tut52.html
  - Another new page shows how to deal with LTLf formulas (i.e., LTL
    with finite semantics) and how to translate those.
    See https://spot.lrde.epita.fr/tut12.html

  Build:

  - Work around a spurious null dereference warning when compiling
    with --coverage and g++ 8.3.0-3 from Debian unstable.

New in spot 2.7.1 (2019-02-14)

  Build:

  - Work around GCC bug #89303 that causes memory leaks and std::weak_bad_ptr
    exceptions when Spot is compiled with the version of g++ 8.2 currently
    distributed by Debian unstable (starting with g++ 8.2.0-15).

  Python:

  - The following methods of spot::bdd_dict are now usable in Python when
    fine control over the lifetime of associations between BDD variables
    and atomic propositions is needed.
    - register_proposition(formula, for_me)
    - register_anonymous_variables(count, for_me)
    - register_all_propositions_of(other, for_me)
    - unregister_all_my_variables(for_me)
    - unregister_variable(var, for_me)

  - Better support for explicit Kripke structures:
    - the kripke_graph type now has Python bindings
    - spot.automaton() and spot.automata() now support a want_kripke=True
      to return a kripke_graph
      See the bottom of https://spot.lrde.epita.fr/ipynb/ltsmin-dve.html
      for some examples.

  Library:

  - Printing Kripke structures via print_hoa() will save state names.

  - kripke_graph_ptr objects now honor any "state-names" property
    when formatting states.

  Bugs fixed:

  - The print_dot_psl() function would incorrectly number all but the
    first children of commutative n-ary operators: in this case no
    numbering was expected.

  - std::out_of_range C++ exceptions raised from Python code are now
    converted into IndexError Python exceptions (instead of aborting
    the program).

  - The LTL parser would choke on carriage returns when command-line
    tools such as ltlfilt, ltlcross, or ltl2tgba were run on files of
    formulas with MS-DOS line endings.

  - The core translation for unambiguous automata was incorrectly
    tagging some non-weak automata as weak.

  - The product_susp() function used to multiply an automaton with a
    suspendable automaton could incorrectly build transition-based
    automata when multiplying two state-based automata.  This caused
    ltl2tgba to emit error messages such as: "automaton has
    transition-based acceptance despite prop_state_acc()==true".

New in spot 2.7 (2018-12-11)

  Command-line tools:

  - ltlsynt now has three algorithms for synthesis:
    --algo=sd     is the historical one. The automaton of the formula
                  is split to separate inputs and outputs, then
                  determinized (with Safra construction).
    --algo=ds     the automaton of the formula is determinized (Safra),
                  then split to separate inputs and outputs.
    --algo=lar    translate the formula to a deterministic automaton
                  with an arbitrary acceptance condition, then turn it
                  into a parity automaton using LAR, and split it.
    In all three cases, the obtained parity game is solved using
    Zielonka algorithm. Calude's quasi-polynomial time algorithm has
    been dropped as it was not used.

  - ltlfilt learned --liveness to match formulas representing liveness
    properties.

  - the --stats= option of tools producing automata learned how to
    tell if an automaton uses universal branching (%u), or more
    precisely how many states (%[s]u) or edges (%[e]u) use universal
    branching.

  Python:

  - spot.translate() and spot.postprocess() now take an xargs=
    argument similar to the -x option of ltl2tgba and autfilt, making
    it easier to fine tune these operations.  For instance
        ltl2tgba 'GF(a <-> XXa)' --det -x gf-guarantee=0
    would be written in Python as
        spot.translate('GF(a <-> XXa)', 'det', xargs='gf-guarantee=0')
    (Note: those extra options are documented in the spot-x(7) man page.)

  - spot.is_generalized_rabin() and spot.is_generalized_streett() now return
    a tuple (b, v) where b is a Boolean, and v is the vector of the sizes
    of each generalized pair.  This is a backward incompatible change.

  Library:

  - The LTL parser learned syntactic sugar for nested ranges of X
    using the X[n], F[n:m], and G[n:m] syntax of TSLF.  (These
    correspond to the next!, next_e!, and next_a! operators of PSL,
    but we do not support those under these names currently.)

      X[6]a = XXXXXXa
      F[2:4]a = XX(a | X(a | Xa))
      G[2:4]a = XX(a & X(a & Xa))

    The corresponding constructors (for C++ and Python) are
      formula::X(unsigned, formula)
      formula::F(unsigned, unsigned, formula)
      formula::G(unsigned, unsigned, formula)

  - spot::unabbreviate(), used to rewrite away operators such as M or
    W, learned to use some shorter rewritings when an argument (e) is
    a pure eventuality or (u) is purely universal:

      Fe = e
      Gu = u
      f R u = u
      f M e = F(f & e)
      f W u = G(f | u)

  - The twa_graph class has a new dump_storage_as_dot() method
    to show its data structure.  This is more conveniently used
    as aut.show_storage() in a Jupyter notebook.  See
    https://spot.lrde.epita.fr/ipynb/twagraph-internals.html

  - spot::generic_emptiness_check() is a new function that performs
    emptiness checks of twa_graph_ptr (i.e., automata not built
    on-the-fly) with an *arbitrary* acceptance condition.  Its sister
    spot::generic_emptiness_check_scc() can be used to decide the
    emptiness of an SCC.  This is now used by
    twa_graph_ptr::is_empty(), twa_graph_ptr::intersects(), and
    scc_info::determine_unknown_acceptance().

  - The new function spot::to_parity() translates an automaton with
    arbitrary acceptance condition into a parity automaton, based on a
    last-appearance record (LAR) construction.  (It is used by ltlsynt
    but not yet by autfilt or ltl2tgba.)

  - The new function is_liveness() and is_liveness_automaton() can be
    used to check whether a formula or an automaton represents a
    liveness property.

  - Two new functions count_univbranch_states() and
    count_univbranch_edges() can help measuring the amount of
    universal branching in alternating automata.

  Bugs fixed:

  - translate() would incorrectly mark as stutter-invariant
    some automata produced from formulas of the form X(f...)
    where f... is syntactically stutter-invariant.

  - acc_cond::is_generalized_rabin() and
    acc_cond::is_generalized_streett() did not recognize the cases
    where a single generalized pair is used.

  - The pair of acc_cond::mark_t returned by
    acc_code::used_inf_fin_sets(), and the pair (bool,
    vector_rs_pairs) by acc_cond::is_rabin_like() and
    acc_cond::is_streett_like() were not usable in Python.

  - Many object types had __repr__() methods that would return the
    same string as __str__(), contrary to Python usage where repr(x)
    should try to show how to rebuild x.  The following types have
    been changed to follow this convention:
       spot.acc_code
       spot.acc_cond
       spot.atomic_prop_set
       spot.formula
       spot.mark_t
       spot.twa_run (__repr__ shows type and address)
       spot.twa_word (likewise, but _repr_latex_ used in notebooks)

    Note that this you were relying on the fact that Jupyter calls
    repr() to display returned values, you may want to call print()
    explicitly if you prefer the old representation.

  - Fix compilation under Cygwin and Alpine Linux, both choking
    on undefined secure_getenv().

New in spot 2.6.3 (2018-10-17)

  Bugs fixed:

  - Running "ltl2tgba -B" on formulas of the type FG(safety) would
    unexpectedly use a co-Büchi automaton as an intermediate step.
    This in turn caused "ltl2tgba -U -B" to not produce unambiguous
    automata.

  - ltl2tgba --low now disables the "gf-guarantee" feature, as
    documented.

  - ltlfilt's --accept-word and --reject-word options were ignored
    unless used together.

New in spot 2.6.2 (2018-09-28)

  Build:

  - We no longer distribute the Python-based CGI script + javascript
    code for the online translator.  Its replacement has its own
    repository: https://gitlab.lre.epita.fr/spot/spot-web-app/

  Library:

  - When states are highlighted with more than 8 colors, print_dot()
    will add some extra markers to help distinguishing the colors.
    This helps with the fact that colors 8-15 are lighter versions of
    colors 0-7, and that higher color numbers cycle into this 16-color
    palette.

  Bugs fixed:

  - exclusive_ap::constrain() (called by autfilt --exclusive-ap=...)
    would incorrectly copy the "universal" property of the input
    automaton, causing print_hoa() to fail.

  - configure --disable-doxygen would actually enable it.

  - Fix several warnings emitted by the development version of GCC.

New in spot 2.6.1  (2018-08-04)

  Command-line tools:

  - "ltlfilt --suspendable" is now a synonym for
    "ltlfilt --universal --eventual".

  Bugs fixed:

  - scc_info::split_on_sets() did not correctly register the
    atomic propositions of the returned automata.

  - The spot::tl_simplifier class could raise an exception while
    attempting to reduce formulas containing unsimplified <->, -> or
    xor, if options nenoform_stop_on_boolean and synt_impl are both
    set.  (This combination of options is not available from
    command-line tools.)

  - The spot::contains(a, b) function introduced in 2.6 was testing
    a⊆b instead of a⊇b as one would expect.  Unfortunately the
    documentation was also matching the code, so this is a backward
    incompatible change, but a short-lived one.

  - The Python binding of the getter of spot::parsed_formula::f was
    returning a reference instead of a copy, causing issues if the
    reference outlasted the parsed_formula struct.


New in spot 2.6  (2018-07-04)

  Command-line tools:

  - autfilt learned --is-colored to filter automata that use
    exactly one acceptance set per mark or transition.

  - autfilt learned --has-univ-branching and --has-exist-branching
    to keep automata that have universal branching, or that make
    non-deterministic choices.

  - autcross' tool specifications now have %M replaced by the name of
    the input automaton.

  - autcross now aborts if there is any parser diagnostic for the
    input automata (previous versions would use the input automaton
    whenever the parser would manage to read something).

  - ltlcross, ltldo, and autcross learned shorthands to call
    delag, ltl2dra, ltl2dgra, and nba2dpa.

  - When autfilt is asked to build a Büchi automaton from
    of a chain of many products, as in
    "autfilt -B --product 1.hoa ... --product n.hoa in.hoa"
    then it will automatically degeneralize the intermediate
    products to avoid exceeding the number of supported
    acceptance sets.

  - genltl learned to generate six new families of LTL formulas:
      --gf-equiv-xn=RANGE    GF(a <-> X^n(a))
      --gf-implies-xn=RANGE  GF(a -> X^n(a))
      --sejk-f=RANGE[,RANGE] f(0,j)=(GFa0 U X^j(b)), f(i,j)=(GFai U
                             G(f(i-1,j)))
      --sejk-j=RANGE         (GFa1&...&GFan) -> (GFb1&...&GFbn)
      --sejk-k=RANGE         (GFa1|FGb1)&...&(GFan|FGbn)
      --sejk-patterns[=RANGE]   φ₁,φ₂,φ₃ from Sikert et al's [CAV'16]
                             paper (range should be included in 1..3)

  - genltl --ms-example can now take a two-range argument,
    (as --sejk-f above).

  Build:

  - ./configure --enable-max-accsets=N let you specify a maximum
    number of acceptance sets that Spot should support.  The default
    is still 32, but this limit is no longer hard-coded.  Larger values
    will cost additional memory and time.

  - We know distribute RPM packages for Fedora 28.  See
    file:///home/adl/git/spot/doc/userdoc/install.html#Fedora

  Library:

  - The PSL/LTL simplification routine learned the following:

      q R Xf = X(q R f)  if q is suspendable
      q U Xf = X(q U f)  if q is suspendable
      {SERE;1} = {1}     if {SERE} accepts [*0]
      {SERE;1} = {SERE}  if {SERE} does not accept [*0]

  - gf_guarantee_to_ba() is a specialized construction for translating
    formulas of the form GF(guarantee) to BA or DBA, and
    fg_safety_to_dca() is a specialized construction for translating
    formulas of the form FG(safety) to DCA.  These are generalizations
    of some constructions proposed by J. Esparza, J. Křentínský, and
    S. Sickert (LICS'18).

    These are now used by the main translation routine, and can be
    disabled by passing -x '!gf-guarantee' to ltl2tgba.  For example,
    here are the size of deterministic transition-based generalized
    Büchi automata constructed from four GF(guarantee) formulas with
    two versions of Spot, and converted to other types of
    deterministic automata by other tools distributed with Owl 18.06.
    "x(y)" means x states and y acceptance sets.

                               ltl2tgba -D  delag ltl2dra
                                2.5   2.6   18.06  18.06
    -------------------------  -----------  -------------
    GF(a <-> XXa)               9(1)  4(1)   4(2)   9(4)
    GF(a <-> XXXa)             27(1)  8(1)   8(2)  25(4)
    GF(((a & Xb) | XXc) & Xd)   6(1)  4(1)  16(1)   5(2)
    GF((b | Fa) & (b R Xb))     6(2)  2(1)   3(4)   3(4)

    Note that in the above the automata produced by 'ltl2tgba -D' in
    version 2.5 were not deterministic (because -D is only a
    preference).  They are deterministic in 2.6 and other tools.

  - spot::product() and spot::product_or() learned to produce an
    automaton with a simpler acceptance condition if one of the
    argument is a weak automaton.  In this case the resulting
    acceptance condition is (usually) that of the other argument.

  - spot::product_susp() and spot::product_or_susp() are new
    functions for building products between an automaton A and
    a "suspendable" automaton B.  They are also optimized for
    the case that A is weak.

  - When 'generic' acceptance is enabled, the translation routine will
    split the input formula on Boolean operators into components that
    are syntactically 'obligation', 'suspendable', or 'something
    else'.  Those will be translated separately and combined with
    product()/product_susp().  This is inspired by the ways things are
    done in ltl2dstar or delag, and can be disabled by passing option
    -xltl-split=0, as in ltl2tgba -G -D -xltl-split=0.  Here are the
    sizes of deterministic automata (except for ltl3tela which produces
    non-deterministic automata) produced with generic acceptance
    using two versions of ltl2tgba and other tools for reference.

                                             ltl2tgba -DG  delag  ltl3tela
                                              2.5    2.6   18.06    1.1.2
                                             ------------ ------- --------
    FGa0&GFb0                                 2(2)   1(2)   1(2)    1(2)
    (FGa1&GFb1)|FGa0|GFb0                    16(6)   1(4)   1(4)    1(9)
    (FGa2&GFb2)|((FGa1|GFb1)&FGa0&GFb0)      29(8)   1(6)   1(6)    3(11)
    FGa0|GFb0                                 5(4)   1(2)   1(2)    1(5)
    (FGa1|GFb1)&FGa0&GFb0                     8(4)   1(4)   1(4)    1(7)
    (FGa2|GFb2)&((FGa1&GFb1)|FGa0|GFb0)     497(14)  1(6)   1(6)    1(14)
    GFa1 <-> GFz                              4(6)   1(3)   1(4)    1(7)
    (GFa1 & GFa2) <-> GFz                     8(4)   1(3)   1(6)    3(10)
    (GFa1 & GFa2 & GFa3) <-> GFz             21(4)   1(4)   1(8)    3(13)
    GFa1 -> GFz                               5(4)   1(2)   1(2)    1(5)
    (GFa1 & GFa2) -> GFz                     12(4)   1(3)   1(3)    1(6)
    (GFa1 & GFa2 & GFa3) -> GFz              41(4)   1(4)   1(4)    1(7)
    FG(a|b)|FG(!a|Xb)                         4(2)   2(2)   2(2)    2(3)
    FG(a|b)|FG(!a|Xb)|FG(a|XXb)              21(2)   4(3)   4(3)    4(4)
    FG(a|b)|FG(!a|Xb)|FG(a|XXb)|FG(!a|XXXb) 170(2)   8(4)   8(4)    8(5)

  - For 'parity' output, the 'ltl-split' optimization just separates
    obligation sub-formulas from the rest, where a determinization is
    still performed.
                                 ltl2tgba -DP  ltl3dra  ltl2dpa
                                   2.5   2.6    0.2.3    18.06
                                -------------- -------  -------
    FGp0 & (Gp1 | XFp2)           6(2)  4(1)     4(1)     4(2)
    G!p0 | F(p0 & (!p1 W p2))     5(2)  4(2)     n/a      5(2)
    (p0 W XXGp0) & GFp1 & FGp2    6(2)  5(2)     n/a      6(3)

    (The above just show a few cases that were improved.  There are
    many cases where ltl2dpa still produces smaller automata.)

  - The automaton postprocessor will now simplify acceptance
    conditions more aggressively, calling spot::simplify_acceptance()
    or spot::cleanup_acceptance() depending on the optimization level.

  - print_dot(), used to print automata in GraphViz's format,
    underwent several changes:

    * option "a", for printing the acceptance condition, is now
      enabled by default.  Option "A", introduced in Spot 2.4, can be
      used to hide the acceptance condition in case you do not want
      it.  This change of course affects the --dot option of all the
      command-line tools, as well as the various way to display
      automata using the Python bindings.

    * when option "1" is used to hide state names and force the
      display of state numbers, the actual state names is now moved to
      the "tooltip" field of the state.  The SVG files produced by
      "dot -Tsvg" will show those as popups.  This is also done for
      state labels of Kripke structures.

    * the output digraph is now named using the name of the automaton
      if available, or the empty string otherwise.  (Previous versions
      used to call all digraphs "G".)  This name appears as a tooltip
      in SVG figures when the mouse is over the acceptance condition.

    * a new option "u" hides "true states" behind "exiting
      transitions".  This can be used to display alternating automata
      in a way many people expect.

    * a new option "K" cancels the effect of "k" (which uses state
      labels whenever possible).  This is most useful when one want to
      force transition-labeling of a Kripke structure, where "k" is
      usually the default.

  - spot::twa_graph::merge_states() is a new method that merges states
    with the exact same outgoing edges. As it performs no reordering
    of the edges, it is better to call it when you know that the edges
    in the twa_graph are sorted (e.g. after a call to merge_edges()).

  - spot::twa_graph::purge_unreachable_states() now takes a function
    which is called with the new numbering of states. This is useful
    to update an external structure that references states of the twa
    that we want to purge.

  - spot::scc_filter() now automatically turns automata marked as
    inherently-weak into weak automata with state-based acceptance.
    The acceptance condition is set to Büchi unless the input had
    co-Büchi or t acceptance.  spot::scc_filter_states() will pass
    inherently-weak automata to spot::scc_filter().

  - spot::cleanup_parity() and spot::cleanup_parity_here() are smarter
    and now remove from the acceptance condition the parity colors
    that are not used in the automaton.

  - spot::contains() and spot::are_equivalent() can be used to
    check language containment between two automata or formulas.
    They are most welcome in Python, since we used to redefine
    them every now and them.  Some examples are shown in
    https://spot.lrde.epita.fr/ipynb/contains.html

  - aut1->exclusive_word(aut2) is a new method that returns a word
    accepted by aut1 or aut2 but not both.  The exclusive_run()
    variant will return.  This is useful when comparing automata and
    looking for differences.  See also
    https://spot.lrde.epita.fr/ipynb/contains.html

  - spot::complement_semidet(aut) is a new function that returns the
    complement of aut, where aut is a semi-deterministic automaton. The
    function uses the NCSB complementation algorithm proposed by
    F. Blahoudek, M. Heizmann, S. Schewe, J. Strejček, and MH. Tsai
    (TACAS'16).

  - spot::remove_alternation() was slightly improved on very-weak
    alternating automata: the labeling of the outgoing transitions in
    the resulting TGBA makes it more likely that simulation-based
    reductions will reduce it.

  - When applied to automata that are not WDBA-realizable,
    spot::minimize_wdba() was changed to produce an automaton
    recognizing a language that includes the original one.  As a
    consequence spot::minimize_obligation() and
    spot::is_wdba_realizable() now only need one containment check
    instead of two.

  - Slightly improved log output for the SAT-based minimization
    functions.  The CSV log files now include an additional column
    with the size of the reference automaton, and they now have a
    header line.  These log files give more details and are more
    accurate in the case of incremental SAT-solving.

  Python:

  - New spot.jupyter package.  This currently contains a function for
    displaying several arguments side by side in a Jupyter notebook.
    See https://spot.lrde.epita.fr/ipynb/alternation.html for some
    examples.

  - Strings are now implicitly converted into formulas when passed
    as arguments to functions that expect formulas.  Previously this
    was done only for a few functions.

  - The Python bindings for sat_minimize() now have display_log and
    return_log options; these are demonstrated on the new
    https://spot.lrde.epita.fr/ipynb/satmin.html page.

  Bugs fixed:

  - Python *.py and *.so files are now always installed into the same
    directory.  This was an issue on systems like Fedora that separate
    platform-specific packages from non-platform-specific ones.

  - print_dot() will correctly escape strings containing \n in HTML
    mode.

  - The HOA parser will now accept Alias: declarations that occur
    before AP:.

  - The option --allow-dups of randltl now works properly.

  - Converting generalized-co-Büchi to Streett using dnf_to_nca()
    could produce bogus automata if the input had rejecting SCCs.

  Deprecation notices:

  - The type spot::acc_cond::mark_t has been overhauled and uses
    a custom bit-vector to represent acceptance sets instead of
    storing everything in a "unsigned int".  This change is
    to accommodate configure's --enable-max-accsets=N option and
    has several effect:

    * The following shortcuts are deprecated:
        acc_cond::mark_t m1 = 0U;
        acc_cond::mark_t m2 = -1U;
      instead, use:
        acc_cond::mark_t m1 = {};
        acc_cond::mark_t m2 = acc_cond::mark_t::all();

    * acc_cond::mark_t::value_t is deprecated.  It is now only
      defined when --enable-max-accsets=32 (the default) and
      equal to "unsigned" for backward compatibility reasons.

  Backward incompatibilities:

  - Functions spot::parity_product() and spot::parity_product_or()
    were removed.  The code was unused, hard to maintain, and bogus.

  - The output of print_dot() now include the acceptance condition.
    Add option "A" (supported since version 2.4) to cancel that.

  - Because genltl now supports LTL pattern with two arguments, using
    --format=%L may output two comma-separated integer.  This is an
    issue if you used to produce CSV files using for instance:
      genltl --format='%F,%L,%f' ...
    Make sure to quote %L to protect the potential commas:
      genltl --format='%F,"%L",%f' ...

  - In Spot 2.5 and prior running "ltl2tgba --generic --det" on some
    formula would attempt to translate it as deterministic TGBA or
    determinize it into an automaton with parity acceptance.  Version
    2.5 introduced --parity to force parity acceptance on the output.
    This version finally gives --generic some more natural semantics:
    any acceptance condition can be used.


New in spot 2.5.3 (2018-04-20)

  Bugs fixed:

  - "autfilt --cobuchi --small/--det" would turn a transition-based
    co-Büchi automaton into a state-based co-Büchi.

  - Fix cryptic error message from Python's spot.translate() and
    spot.postprocess() when supplying conflicting arguments.

  - "autfilt -B --sat-minimize" was incorrectly producing
    transition-based automata.

  - Using spot.automata("cmd...|") to read just a few automata out of
    an infinite stream would not properly terminate the command.

  - The is_unambiguous() check (rewritten in Spot 2.2) could mark some
    unambiguous automata as ambiguous.

New in spot 2.5.2 (2018-03-25)

  Bugs fixed:

  - acc_cond::is_generalized_rabin() and
    acc_cond::is_generalized_streett() would segfault on weird
    acceptance conditions such as "3 t" or "3 f".

  - remove_fin() and streett_to_generalized_buchi() should never
    return automata with "f" acceptance.

  - "autfilt --acceptance-is=Fin-less" no longer accept automata
    with "f" acceptance.

  - twa_run methods will now diagnose cases where the cycle is
    unexpectedly empty instead of segfaulting.

  - spot::closure(), used by default for testing stutter-invariance,
    was using an optimization incorrect if the acceptance condition
    had some Fin(x).  Consequently stutter-invariance tests for
    automata, for instance with "autfilt --is-stutter-invariant",
    could be to be wrong (even if the input does not use Fin(x), its
    complement, used in the stutter-invariance test likely will).
    Stutter-invariance checks of LTL formulas are not affected.

New in spot 2.5.1 (2018-02-20)

  Library:

  - iar() and iar_maybe() now also handle Streett-like conditions.

  Bugs fixed:

  - iar() and iar_maybe() properly handle Rabin-like conditions.

  - streett_to_generalized_buchi() could produce incorrect result on
    Streett-like input with acceptance like (Inf(0)|Fin(1))&Fin(1)
    where some Fin(x) is used both with and without a paired Fin(y).

  - is_generalized_rabin() had a typo that caused some non-simplified
    acceptance conditions like Fin(0)|(Fin(0)&Inf(1)) to be
    incorrectly detected as generalized-Rabin 2 0 1 and then output
    as Fin(0)|(Fin(1)&Inf(2)) instead.  Likewise for
    is_generalized_streett

  - the conversion from Rabin to Büchi was broken on Rabin-like
    acceptance condition where one pair used Fin(x) and another pair
    used Inf(x) with the same x.  Unfortunately, this situation could
    also occur as a side effect of simplifying the acceptance
    condition (by merging identical set) of some automaton prior to
    converting it to Büchi.

  - dnf_to_dca() was mishandling some Rabin-like input.

New in spot 2.5 (2018-01-20)

  Build:

  - We no longer distribute the doxygen-generated documentation in
    the tarball of Spot to save space.  This documentation is still
    available online at https://spot.lrde.epita.fr/doxygen/.  If you
    want to build a local copy you can configure Spot with
    --enable-doxygen, or simply run "cd doc && make doc".

  - All the images that illustrate the documentation have been
    converted to SVG, to save space and improve quality.

  Tools:

  - ltlsynt is a new tool for synthesizing a controller from LTL/PSL
    specifications.

  - ltlcross learned --reference=COMMANDFMT to specify a translator
    that should be trusted.  Doing so makes it possible to reduce the
    number of tests to be performed, as all other translators will be
    compared to the reference's output when available.  Multiple
    reference can be given; in that case other tools are compared
    against the smallest reference automaton.

  - autcross, ltlcross, and ltldo learned --fail-on-timeout.

  - ltl2tgba, autfilt, and dstar2tgba have some new '--parity' and
    '--colored-parity' options to force parity acceptance on the
    output.  Different styles can be requested using for instance
    --parity='min odd' or --parity='max even'.

  - ltl2tgba, autfilt, and dstar2tgba have some new '--cobuchi' option
    to force co-Büchi acceptance on the output.  Beware: if the input
    language is not co-Büchi realizable the output automaton will
    recognize a superset of the input.  Currently, the output is
    always state-based.

  - genltl learned to generate six new families of formulas, taken from
    the SYNTCOMP competition on reactive synthesis, and from from
    Müller & Sickert's GandALF'17 paper:
    --gf-equiv=RANGE       (GFa1 & GFa2 & ... & GFan) <-> GFz
    --gf-implies=RANGE     (GFa1 & GFa2 & ... & GFan) -> GFz
    --ms-example=RANGE     GF(a1&X(a2&X(a3&...)))&F(b1&F(b2&F(b3&...)))
    --ms-phi-h=RANGE       FG(a|b)|FG(!a|Xb)|FG(a|XXb)|FG(!a|XXXb)|...
    --ms-phi-r=RANGE       (FGa{n}&GFb{n})|((FGa{n-1}|GFb{n-1})&(...))
    --ms-phi-s=RANGE       (FGa{n}|GFb{n})&((FGa{n-1}&GFb{n-1})|(...))

  - autfilt learned --streett-like to convert automata with DNF
    acceptance into automata with Streett-like acceptance.

  - autfilt learned --acceptance-is=ACC to filter automata by
    acceptance condition.  ACC can be the name of some acceptance
    class (e.g. Büchi, Fin-less, Streett-like) or a precise acceptance
    formula in the HOA syntax.

  - ltldo learned to limit the number of automata it outputs using -n.

  - ltlfilt learned to measure wall-time using --format=%r, and
    cpu-time with --format=%R.

  - The --format=%g option of tools that output automata used to
    print the acceptance condition as a *formula* in the HOA format.
    This %g may now take optional arguments to print the acceptance
    *name* in different formats.  For instance

       ... | autfilt -o '%[s]g.hoa'

    will separate a stream of automata into different files named
    by their acceptance name (Buchi, co-Buchi, Streett, etc.) or
    "other" if no name is known for the acceptance condition.

  - Tools that produce formulas now support --format=%[OP]n to
    display the nesting depth of operator OP.

  - The new -x tls-impl=N option allows to fine-tune the
    implication-based simplification rules of ltl2tgba.  See the
    spot-x(7) man page for details.

  - All tools learned to check the SPOT_OOM_ABORT environment
    variable.  This is only useful for debuging out-of-memory
    conditions. See the spot-x(7) man page for details.

  New functions in the library:

  - spot::iar() and spot::iar_maybe() use index appearance records (IAR)
    to translate Rabin-like automata into equivalent parity automata.
    This translation preserves determinism and is especially useful when
    the input automaton is deterministic.

  - spot::print_aiger() encodes an automaton as an AIGER circuit, as
    required by the SYNTCOMP competition.  It relies on a new named
    property "synthesis outputs" that describes which atomic
    propositions are to be encoded as outputs of the circuits.

  - spot::dnf_to_streett() converts any automaton with a DNF
    acceptance condition into a Streett-like automaton.

  - spot::nsa_to_nca(), spot::dfn_to_nca(), spot::dfn_to_dca(), and
    spot::nsa_to_dca(), convert automata with DNF or Streett-like
    acceptance into deterministic or non-deterministic co-Büchi
    automata.  spot::to_dca() and spot::to_nca() dispatch between
    these four functions.  The language of produced automaton includes
    the original language, but may be larger if the original automaton
    is not co-Büchi realizable.  Based on Boker & Kupferman FOSSACS'11
    paper.  Currently only supports state-based output.

  - spot::scc_info::states_on_acc_cycle_of() returns all states
    visited by any accepting cycle of the specified SCC.  It only
    works on automata with Streett-like acceptance.

  - spot::is_recurrence(), spot::is_persistence(), and
    spot::is_obligation() test if a formula is a recurrence,
    persistance or obligation.  The SPOT_PR_CHECK and SPOT_O_CHECK
    environment variables can be used to select between multiple
    implementations of these functions (see the spot-x(7) man page).

  - spot::is_colored() checks if an automaton is colored
    (i.e., every transition belongs to exactly one acceptance set).

  - spot::change_parity() converts between different parity acceptance
    conditions.

  - spot::colorize_parity() transforms an automaton with parity
    acceptance into a colored automaton (which is often what people
    working with parity automata expect).

  - spot::cleanup_parity_acceptance() simplifies a parity acceptance
    condition.

  - spot::parity_product() and spot::parity_product_or() are
    specialized (but expensive) products that preserve parity
    acceptance.

  - spot::remove_univ_otf() removes universal transitions on the fly
    from an alternating Büchi automaton using Miyano and Hayashi's
    breakpoint algorithm.

  - spot::stutter_invariant_states(),
    spot::stutter_invariant_letters(),
    spot::highlight_stutter_invariant_states(), ...  These functions
    help study a stutter-sensitive automaton and detect the subset of
    states that are stutter-invariant.  See
    https://spot.lrde.epita.fr/ipynb/stutter-inv.html for examples.

  - spot::acc_cond::name(fmt) is a new method that names well-known
    acceptance conditions.  The fmt parameter specifies the format to
    use for that name (e.g. to the style used in HOA, or that used by
    print_dot()).

  - spot::formula::is_leaf() method can be used to detect formulas
    without children (atomic propositions, or constants).

  - spot::nesting_depth() computes the nesting depth of any LTL
    operator.

  - spot::check_determinism() sets both prop_semi_deterministic()
    and prop_universal() appropriately.

  Improvements to existing functions in the library:

  - spot::tgba_determinize() has been heavily rewritten and
    optimized. The algorithm has (almost) not changed, but it is
    much faster now. It also features an optimization for
    stutter-invariant automata that may produce slightly smaller
    automata.

  - spot::scc_info now takes an optional argument to disable some
    features that are expensive and not always necessary.  By
    default scc_info tracks the list of all states that belong to an
    SCC (you may now ask it not to), tracks the successor SCCs of
    each SCC (that can but turned off), and explores all SCCs of the
    automaton (you may request to stop on the first SCC that is
    found accepting).

  - In some cases, spot::degeneralize() would output Büchi automata
    with more SCCs than its input.  This was hard to notice, because
    very often simulation-based simplifications remove those extra
    SCCs.  This situation is now detected by spot::degeneralized()
    and fixed before returning the automaton.  A new optional
    argument can be passed to disable this behavior (or use -x
    degen-remscc=0 from the command-line).

  - The functions for detecting stutter-invariant formulas or
    automata have been overhauled.  Their interface changed
    slightly.  They are now fully documented.

  - spot::postprocessor::set_type() can now request different forms
    of parity acceptance as output.  However currently the
    conversions are not very smart: if the input does not already
    have parity acceptance, it will simply be degeneralized or
    determinized.

  - spot::postprocessor::set_type() can now request co-Büchi
    acceptance as output.  This calls the aforementioned to_nca() or
    to_dca() functions.

  - spot::remove_fin() will now call simplify_acceptance(),
    introduced in 2.4, before attempting its different Fin-removal
    strategies.

  - spot::simplify_acceptance() was already merging identical
    acceptance sets, and detecting complementary sets i and j to
    perform the following simplifications
      Fin(i) & Inf(j) = Fin(i)
      Fin(i) | Inf(j) = Inf(i)
    It now additionally applies the following rules (again assuming i
    and j are complementary):
      Fin(i) & Fin(j) = f          Inf(i) | Inf(j) = t
      Fin(i) & Inf(i) = f          Fin(i) | Inf(i) = t

  - spot::formula::map(fun) and spot::formula::traverse(fun) will
    accept additionnal arguments and pass them to fun().  See
    https://spot.lrde.epita.fr/tut03.html for some examples.

  Python-specific changes:

  - The "product-states" property of automata is now accessible via
    spot.twa.get_product_states() and spot.twa.set_product_states().

  - twa_word instances can be displayed as SVG pictures, with one
    signal per atomic proposition.  For some examples, see the use of
    the show() method in https://spot.lrde.epita.fr/ipynb/word.html

  - The test suite is now using v4 of the Jupyter Notebook format.

  - The function rabin_is_buchi_realizable() as its name suggests checks
    if a rabin aut. is Büchi realizable. It is heavily based on
    tra_to_tba() algorithm.

  Deprecation notices:

  (These functions still work but compilers emit warnings.)

  - spot::scc_info::used_acc(), spot::scc_info::used_acc_of() and
    spot::scc_info::acc() are deprecated.  They have been renamed
    spot::scc_info::marks(), spot::scc_info::marks_of() and
    spot::scc_info::acc_sets_of() respectively for clarity.

  Backward incompatible changes:

  - The spot::closure(), spot::sl2(), spot::is_stutter_invariant()
    functions no longer take && arguments.  The former two have
    spot::closure_inplace() and spot::sl2_inplace() variants.  These
    functions also do not take a list of atomic propositions as an
    argument anymore.

  - The spot::bmrand() and spot::prand() functions have been removed.
    They were not used at all in Spot, and it is not Spot's objective
    to provide such random functions.

  - The spot::bdd_dict::register_all_propositions_of() function has
    been removed.  This low-level function was not used anywhere in
    Spot anymore, since it's better to use spot::twa::copy_ap_of().

  Bugs fixed:

  - spot::simplify_acceptance() could produce incorrect output if the
    first edge of the automaton was the only one with no acceptance
    set.  In spot 2.4.x this function was only used by autfilt
    --simplify-acceptance; in 2.5 it is now used in remove_fin().

  - spot::streett_to_generalized_buchi() could generate incorrect
    automata with empty language if some Fin set did not intersect all
    accepting SCCs (in other words, the fix from 2.4.1 was incorrect).

  - ltlcross could crash when calling remove_fin() on an automaton
    with 32 acceptance sets that would need an additional set.

  - the down_cast() helper function used in severaly headers of Spot
    was not in the spot namespace, and caused issues with some
    configurations of GCC.

New in spot 2.4.4 (2017-12-25)

  Bugs fixed:

  - The generic to_generalized_buchi() function would fail if the
    Fin-less & CNF version of the acceptance condition had several
    unit clauses.

  - If the automaton passed to sbacc() was incomplete or
    non-deterministic because of some unreachable states, then it was
    possible that the output would marked similarly while it was in
    fact complete or deterministic.

New in spot 2.4.3 (2017-12-19)

  Bugs fixed:

  - couvreur99_new() leaked memory when processing TωA that allocate
    states.

  - Some mismatched placement-new/delete reported by ASAN were fixed.

  - Static compilation was broken on MinGW.

  - Execution of Jupyter notebooks from the testsuite failed with
    recent versions of Python.

  - Fix some warnings triggered by the development version of g++.

New in spot 2.4.2 (2017-11-07)

  Tools:

  - ltlcross and ltldo support ltl3tela, the new name of ltl3hoa.

  Bugs fixed:

  - Automata produced by "genaut --ks-nca=N" were incorrectly marked
    as not complete.

  - Fix some cases for which the highest setting of formulas
    simplification would produce worse results than lower settings.

New in spot 2.4.1 (2017-10-05)

  Bugs fixed:

  - The formula class failed to build {a->c[*]} although it is
    allowed by our grammar.

  - spot::scc_info::determine_unknown_acceptance() incorrectly
    considered some rejecting SCC as accepting.

  - spot::streett_to_generalized_buchi() could generate automata with
    empty language if some Fin set did not intersect all accepting
    SCCs.  As a consequence, some Streett-like automata were
    considered empty even though they were not.  Also, the same
    function could crash on input that had a Streett-like acceptance
    not using all declared sets.

  - The twa_graph::merge_edges() function relied on BDD IDs to sort
    edges.  This in turn caused some algorithms (like the
    degeneralization) to produce slighltly different (but still correct)
    outputs depending on the BDD operations performed before.

  - spot::simulation() could incorrectly flag an automaton as
    non-deterministic.

New in spot 2.4 (2017-09-06)

  Build:

  - Spot is now built in C++14 mode, so you need at least GCC 5 or
    clang 3.4.  The current version of all major linux distributions
    ship with at least GCC 6, which defaults to C++14, so this should
    not be a problem.  In *this* release of Spot, most of the header
    files are still C++11 compatible, so you should be able to include
    Spot in a C++11 project in case you do not yet want to upgrade.
    There is also an --enable-c++17 option to configure in case you
    want to force a build of Spot in C++17 mode.

  Tools:

  - genaut is a new binary that produces families of automata defined
    in the literature (in the same way as we have genltl for LTL
    formulas).

  - autcross is a new binary that compares the output of tools
    transforming automata (in the same way as we have ltlcross for
    LTL translators).

  - genltl learned two generate two new families of formulas:
      --fxg-or=RANGE         F(p0 | XG(p1 | XG(p2 | ... XG(pn))))
      --gxf-and=RANGE        G(p0 & XF(p1 & XF(p2 & ... XF(pn))))
    The later is a generalization of --eh-pattern=9, for which a
    single state TGBA always exists (but previous version of Spot
    would build larger automata).

  - autfilt learned to build the union (--sum) or the intersection
    (--sum-and) of two languages by putting two automata side-by-side
    and fiddling with the initial states.  This complements the already
    implemented intersection (--product) and union (--product-or),
    both based on a product.

  - autfilt learned to complement any alternating automaton with
    option --dualize.   (See spot::dualize() below.)

  - autfilt learned --split-edges to convert labels that are Boolean
    formulas into labels that are min-terms.  (See spot::split_edges()
    below.)

  - autfilt learned --simplify-acceptance to simplify some acceptance
    conditions.  (See spot::simplify_acceptance() below.)

  - autfilt --decompote-strength has been renamed to --decompose-scc
    because it can now extract the subautomaton leading to an SCC
    specified by number.  (The old name is still kept as an alias.)

  - The --stats=%c option of tools producing automata can now be
    restricted to count complete SCCs, using %[c]c.

  - Tools producing automata have a --stats=... option, and tools
    producing formulas have a --format=... option.  These two options
    work similarly, the use of different names is just historical.
    Starting with this release, all tools that recognize one of these
    two options also accept the other one as an alias.

  Library:

  - A new library, libspotgen, gathers all functions used to generate
    families of automata or LTL formulas, used by genltl and genaut.

  - spot::sum() and spot::sum_and() implements the union and the
    intersection of two automata by putting them side-by-side and
    using non-deterministim or universal branching on the initial
    state.

  - twa objects have a new property: prop_complete().  This obviously
    acts as a cache for the is_complete() function.

  - spot::dualize() complements any alternating automaton.  Since
    the dual of a deterministic automaton is still deterministic, the
    function spot::dtwa_complement() has been deprecated and simply
    calls spot::dualize().

  - spot::decompose_strength() was extended and renamed to
    spot::decompose_scc() as it can now also extract a subautomaton
    leading to a particular SCC.  A demonstration of this feature via
    the Python bindings can be found at
    https://spot.lrde.epita.fr/ipynb/decompose.html

  - The print_dot() function will now display names for well known
    acceptance conditions under the formula when option 'a' is used.
    We plan to enable 'a' by default in a future release, so a new
    option 'A' has been added to hide the acceptance condition.

  - The print_dot() function has a new experimental option 'x' to
    output labels are LaTeX formulas.  This is meant to be used in
    conjunction with the dot2tex tool.  See
      https://spot.lrde.epita.fr/oaut.html#dot2tex

  - A new named property for automata called "original-states" can be
    used to record the origin of a state before transformation.  It is
    currently defined by the degeneralization algorithms, and by
    transform_accessible() and algorithms based on it (like
    remove_ap::strip(), decompose_scc()).  This is realy meant as an
    aid for writing algorithms that need this mapping, but it can also
    be used to debug these algorithms: the "original-states"
    information is displayed by the dot printer when the 'd' option is
    passed.  For instance in

      % ltl2tgba 'GF(a <-> Fb)' --dot=s
      % ltl2tgba 'GF(a <-> Fb)' | autfilt -B --dot=ds

    the second command outputs an automaton with states that show
    references to the first one.

  - A new named property for automata called "degen-levels" keeps track
    of the level of a state in a degeneralization. This information
    complements the one carried in "original-states".

  - A new named property for automata called "simulated-states" can be
    used to record the origin of a state through simulation. The
    behavior is similar to "original-states" above. Determinization
    takes advantage of this in its pretty print.

  - The new function spot::acc_cond::is_streett_like() checks whether
    an acceptance condition is conjunction of disjunctive clauses
    containing at most one Inf and at most one Fin.  It builds a
    vector of pairs to use if we want to assume the automaton has
    Streett acceptance.  The dual function is
    spot::acc_cond::is_rabin_like() works similarly.

  - The degeneralize() function has learned to consider acceptance
    marks common to all edges comming to a state to select its initial
    level.  A similar trick was already used in sbacc(), and saves a
    few states in some cases.

  - There is a new spot::split_edges() function that transforms edges
    (labeled by Boolean formulas over atomic propositions) into
    transitions (labeled by conjunctions where each atomic proposition
    appear either positive or negative).  This can be used to
    preprocess automata before feeding them to algorithms or tools
    that expect transitions labeled by letters.

  - spot::scc_info has two new methods to easily iterate over the
    edges of an SCC: edges_of() and inner_edges_of().

  - spot::scc_info can now be passed a filter function to ignore
    or cut some edges.

  - spot::scc_info now keeps track of acceptance sets that are common
    to all edges in an SCC.  These can be retrieved using
    scc_info::common_sets_of(scc), and they are used by scc_info to
    classify some SCCs as rejecting more easily.

  - The new function acc_code::remove() removes all the given
    acceptance sets from the acceptance condition.

  - It is now possible to change an automaton acceptance condition
    directly by calling twa::set_acceptance().

  - spot::cleanup_acceptance_here now takes an additional boolean
    parameter specifying whether to strip useless marks from the
    automaton.  This parameter is defaulted to true, in order to
    keep this modification backward-compatible.

  - The new function spot::simplify_acceptance() is able to perform
    some simplifications on an acceptance condition, and might lead
    to the removal of some acceptance sets.

  - The function spot::streett_to_generalized_buchi() is now able to
    work on automata with Streett-like acceptance.

  - The function for converting deterministic Rabin automata to
    deterministic Büchi automata (when possible), internal to the
    remove_fin() procedure, has been updated to work with
    transition-based acceptance and with Rabin-like acceptance.

  - spot::relabel_here() was used on automata to rename atomic
    propositions, it can now replace atomic propositions by Boolean
    subformula.  This makes it possible to use relabeling maps
    produced by relabel_bse() on formulas.

  - twa_graph::copy_state_names_from() can be used to copy the state
    names from another automaton, honoring "original-states" if
    present.

  - Building automata for LTL formula with a large number N of atomic
    propositions can be costly, because several loops and
    data-structures are exponential in N.  However a formula like
      ((a & b & c) | (d & e & f)) U ((d & e & f) | (g & h & i))
    can be translated more efficiently by first building an automaton
    for (p0 | p1) U (p1 | p2), and then substituting p0, p1, p2 by the
    appropriate Boolean formula.  Such a trick is now attempted
    for translation of formulas with 4 atomic propositions or
    more (this threshold can be changed, see -x relabel-bool=N in
    the spot-x(7) man page).

  - The LTL simplification routines learned that an LTL formula like
    G(a & XF(b & XFc & Fd) can be simplified to G(a & Fb & Fc & Fd),
    and dually F(a | XG(b | XGc | Gd)) = F(a | Gb | Gc | Gd).

    When working with SERE, the simplification of "expr[*0..1]" was
    improved.  E.g. {{a[*]|b}[*0..1]} becomes {a[*]|b} instead of
    {{a[+]|b}[*0..1]}.

  - The new function spot::to_weak_alternating() is able to take an
    input automaton with generalized Büchi/co-Büchi acceptance and
    convert it to a weak alternating automaton.

  - spot::sbacc() is can now also convert alternating automata
    to state-based acceptance.

  - spot::sbacc() and spot::degeneralize() learned to merge
    accepting sinks.

  - If the SPOT_BDD_TRACE environment variable is set, statistics
    about BDD garbage collection and table resizing are shown.

  - The & and | operators for acceptannce conditions have been changed
    slightly to be more symmetrical.  In older versions, operator &
    would move Fin() terms to the front, but that is not the case
    anymore.  Also operator & was already grouping all Inf() terms
    (for efficiency reasons), in this version operator | is
    symmetrically grouping all Fin() terms.

  - The automaton parser is now reentrant, making it possible to
    process automata from different streams at the same time (i.e.,
    using multiple spot::automaton_stream_parser instances at once).

  - The print_hoa() and parse_automaton() functions have been updated
    to recognize the "exist-branch" property of the non-released HOA
    v1.1, as well as the new meaning of property "deterministic".  (In
    HOA v1 "properties: deterministic" means that the automaton has no
    existential branching; in HOA v1.1 it disallows universal
    branching as well.)  The meaning of "deterministic" in Spot has
    been adjusted to these new semantics, see "Backward-incompatible
    changes" below.

  - The parser for HOA now recognizes and verifies correct use of the
    "univ-branch" property.  This is known to be a problem with option
    -H1 of ltl3ba 1.1.2 and ltl3dra 0.2.4, so the environment variable
    SPOT_HOA_TOLERANT can be set to disable the diagnostic.

  Python:

  - The 'spot.gen' package exports the functions from libspotgen.
    See https://spot.lrde.epita.fr/ipynb/gen.html for examples.

  Bugs fixed:

  - When the remove_fin() function was called on some automaton with
    Inf-less acceptance involving at least one disjunction (e.g.,
    generalized co-Büchi), it would sometimes output an automaton with
    transition-based acceptance but marked as state-based.

  - The complete() function could complete an empty co-Büchi automaton
    into an automaton accepting everything.

  Backward-incompatible changes:

  - spot::acc_cond::mark_t::operator bool() has been marked as
    explicit.  The implicit converion to bool (and, via bool, to int)
    was a source of bugs.

  - spot::twa_graph::set_init_state(const state*) has been removed.
    It was never used.  You always want to use
    spot::twa_graph::set_init_state(unsigned) in practice.

  - The previous implementation of spot::is_deterministic() has been
    renamed to spot::is_universal().  The new version of
    spot::is_deterministic() requires the automaton to be both
    universal and existential.  This should not make any difference in
    existing code unless you work with the recently added support for
    alternating automata.

  - spot::acc_cond::mark_t::sets() now returns an internal iterable
    object instead of an std::vector<unsigned>.

  - The color palette optionally used by print_dot() has been extended
    from 9 to 16 colors.  While the first 8 colors are similar, they
    are a bit more saturated now.

  deprecation notices:

  (these functions still work but compilers emit warnings.)

  - spot::decompose_strength() is deprecated, it has been renamed
    to spot::decompose_scc().

  - spot::dtwa_complement() is deprecated.  prefer the more generic
    spot::dualize() instead.

  - the spot::twa::prop_deterministic() methods have been renamed to
    spot::twa::prop_universal() for consistency with the change to
    is_deterministic() listed above.  we have kept
    spot::twa::prop_deterministic() as a deprecated synonym for
    spot::twa::prop_universal() to help backward compatibility.

  - the spot::copy() function is deprecated.  use
    spot::make_twa_graph() instead.

New in spot 2.3.5 (2017-06-22)

  Bugs fixed:

  - We have fixed new cases where translating multiple formulas in a
    single ltl2tgba run could produce automata different from those
    produced by individual runs.

  - The print_dot() function had a couple of issues when printing
    alternating automata: in particular, when using flag 's' (display
    SCC) or 'y' (split universal destination by colors) universal
    edges could be connected to undefined states.

  - Using --stats=%s or --stats=%s or --stats=%t could take an
    unnecessary long time on automata with many atomic propositions,
    due to a typo.  Furthermore, %s/%e/%t/%E/%T were printing
    a number of reachable states/edges/transitions, but %S was
    incorrectly counting all states even unreachable.

  - Our verson of BuDDy had an incorrect optimization for the
    biimp operator.

New in spot 2.3.4 (2017-05-11)

  Bugs fixed:

  - The transformation to state-based acceptance (spot::sbacc()) was
    incorrect on automata where the empty acceptance mark is accepting.

  - The --help output of randaut and ltl2tgba was showing an
    unsupported %b stat.

  - ltldo and ltlcross could leave temporary files behind when
    aborting on error.

  - The LTL simplifcation rule that turns F(f)|q into F(f|q)
    when q is a subformula that is both eventual and universal
    was documented but not applied in some forgotten cases.

  - Because of some caching inside of ltl2tgba, translating multiple
    formula in single ltl2tgba run could produce automata different
    from those produced by individual runs.

New in spot 2.3.3 (2017-04-11)

  Tools:

  - ltldo and ltlcross learned shorthands to talk to ltl2da, ltl2dpa,
    and ltl2ldba (from Owl) without needing to specify %f>%O.

  - genltl learned --spec-patterns as an alias for --dac-patterns; it
    also learned two new sets of LTL formulas under --hkrss-patterns
    (a.k.a. --liberouter-patterns) and --p-patterns
    (a.k.a. --beem-patterns).

  Bugs fixed:

  - In "lenient" mode the formula parser would fail to recover from a
    missing closing brace.

  - The output of 'genltl --r-left=1 --r-right=1 --format=%F' had
    typos.

  - 'ltl2tgba Fa | autfilt --complement' would incorrectly claim that
    the output is "terminal" because dtwa_complement() failed to reset
    that property.

  - spot::twa_graph::purge_unreachable_states() was misbehaving on
    alternating automata.

  - In bench/stutter/ the .cc files were not compiling due to warnings
    being caught as errors.

  - The code in charge of detecting DBA-type Rabin automata is
    actually written to handle a slightly larger class of acceptance
    conditions (e.g., Fin(0)|(Fin(1)&Inf(2))), however it failed to
    correctly detect DBA-typeness in some of these non-Rabin
    acceptance.

New in spot 2.3.2 (2017-03-15)

  Tools:

  - In tools that output automata, the number of atomic propositions
    can be output using --stats=%x (output automaton) or --stats=%X
    (input automaton).  Additional options can be passed to list
    atomic propositions instead of counting them.  Tools that output
    formulas also support --format=%x for this purpose.

  Python:

  - The bdd_to_formula(), and to_generalized_buchi() functions can now
    be called in Python.

  Documentation:

  - https://spot.lrde.epita.fr/tut11.html is a new page describing
    how to build monitors in Shell, Python, or C++.

  Bugs fixed:

  - The tests using LTSmin's patched version of divine would fail
    if the current (non-patched) version of divine was installed.

  - Because of a typo, the output of --stats='...%P...' was correct
    only if %p was used as well.

  - genltl was never meant to have (randomly attributed) short
    options for --postive and --negative.

  - a typo in the code for transformating transition-based acceptance
    to state-based acceptance could cause a superfluous initial state
    to be output in some cases (the result was still correct).

  - 'ltl2tgba --any -C -M ...' would not complete automata.

  - While not incorrect, the HOA properties output by 'ltl2tgba -M'
    could be 'inherently-weak' or 'terminal', while 'ltl2tgba -M -D'
    would always report 'weak' automata.  Both variants now report the
    most precise between 'weak' or 'terminal'.

  - spot::twa_graph::set_univ_init_state() could not be called with
    an initializer list.

  - The Python wrappers for spot::twa_graph::state_from_number and
    spot::twa_graph::state_acc_sets were broken in 2.3.

  - Instantiating an emptiness check on an automaton with unsupported
    acceptance condition should throw an exception.  This used to be
    just an assertion, disabled in release builds; the difference
    matters for the Python bindings.

  Deprecation notice:

  - Using --format=%a to print the number of atomic propositions in
    ltlfilt, genltl, and randltl still works, but it is not documented
    anymore and should be replaced by the newly-introduced --format=%x
    for consistency with tools producing automata, where %a means
    something else.


New in spot 2.3.1 (2017-02-20)

  Tools:

  - ltldo learnt to act like a portfolio: --smallest and --greatest
    will select the best output automaton for each formula translated.
    See https://spot.lrde.epita.fr/ltldo.html#portfolio for examples.

  - The colors used in the output of ltlcross have been adjusted to
    work better with white backgrounds and black backgrounds.

  - The option (y) has been added to --dot. It splits the universal
    edges with the same targets but different colors.

  - genltl learnt three new families for formulas: --kr-n2=RANGE,
    --kr-nlogn=RANGE, and --kr-n=RANGE.  These formulas, from
    Kupferman & Rosenberg [MoChArt'10] are recognizable by
    deterministic Büchi automata with at least 2^2^n states.

  Library:

  - spot::twa_run::as_twa() has an option to preserve state names.

  - the method spot::twa::is_alternating(), introduced in Spot 2.3 was
    badly named and has been deprecated.  Use the negation of the new
    spot::twa::is_existential() instead.

  Bugs fixed:

  - spot::otf_product() was incorrectly registering atomic
    propositions.

  - spot::ltsmin_model::kripke() forgot to register the "dead"
    proposition.

  - The spot::acc_word type (used to construct acceptance condition)
    was using some non-standard anonymous struct.  It is unlikely that
    this type was actually used outside Spot, but if you do use it,
    spot::acc_word::op and spot::acc_word::type had to be renamed as
    spot::acc_word::sub.op and spot::acc_word::sub.type.

  - alternation_removal() was not always reporting the unsupported
    non-weak automata.

  - a long-standing typo in the configure code checking for Python
    caused any user-defined CPPFLAGS to be ignored while building
    Spot.

  - The display of clusters with universal edges was confused, because the
    intermediate node was not in the cluster even if one of the target was
    in the same one.

New in spot 2.3 (2017-01-19)

  Build:

  * While Spot only relies on C++11 features, the configure script
    learned --enable-c++14 to compile in C++14 mode.  This allows us
    check that nothing breaks when we will switch to C++14.

  * Spot is now distributed with PicoSAT 965, and uses it for
    SAT-based minimization of automata without relying on temporary
    files.  It is still possible to use an external SAT solver by
    setting the SPOT_SATSOLVER environment variable.

  * The development Debian packages for Spot now install static
    libraries as well.

  * We now install configuration files for users of pkg-config.

  Tools:

  * ltlcross supports translators that output alternating automata in
    the HOA format.  Cross-comparison checks will only work with weak
    alternating automata (not necessarily *very* weak), but "ltlcross
    --no-check --product=0 --csv=..." will work with any alternating
    automaton if you just want satistics.

  * autfilt can read alternating automata.  This is still experimental
    (see below). Some of the algorithms proposed by autfilt will
    refuse to work because they have not yet been updated to work with
    alternating automata, but in any case they should display a
    diagnostic: if you see a crash, please report it.

  * autfilt has three new filters: --is-very-weak, --is-alternating,
    and --is-semi-deterministic.

  * the --check option of autfilt/ltl2tgba/ltldo/dstar2tgba can now
    take "semi-determinism" as argument.

  * autfilt --highlight-languages will color states that recognize
    identical languages.  (Only works for deterministic automata.)

  * 'autfilt --sat-minimize' and 'ltl2tgba -x sat-minimize' have
    undergone some backward incompatible changes.  They use binary
    search by default, and support different options than they used
    too.  See spot-x(7) for details.  The defaults are those that were
    best for the benchmark in bench/dtgbasat/.

  * ltlfilt learned --recurrence and --persistence to match formulas
    belonging to these two classes of the temporal hierarchy.  Unlike
    --syntactic-recurrence and --syntactic-persistence, the new checks
    are automata-based and will also match pathological formulas.
    See https://spot.lrde.epita.fr/hierarchy.html

  * The --format option of ltlfilt/genltl/randltl/ltlgrind learned to
    print the class of a formula in the temporal hierarchy of Manna &
    Pnueli using %h.  See https://spot.lrde.epita.fr/hierarchy.html

  * ltldo and ltlcross learned a --relabel option to force the
    relabeling of atomic propositions to p0, p1, etc.  This is more
    useful with ltldo, as it allows calling a tool that restricts the
    atomic propositions it supports, and the output automaton will
    then be fixed to use the original atomic propositions.

  * ltldo and ltlcross have learned how to call ltl3hoa, so
    'ltl3hoa -f %f>%O' can be abbreviated to just 'ltl3hoa'.

  Library:

  * A twa is required to have at least one state, the initial state.
    An automaton can only be empty while it is being constructed,
    but should not be passed to other algorithms.

  * Couvreur's emptiness check has been rewritten to use the explicit
    interface when possible, to avoid overkill memory allocations.
    The new version has further optimizations for weak and terminal
    automata.  Overall, this new version is roughly 4x faster on
    explicit automata than the former one.  The old version has been
    kept for backward compatibility, but will be removed eventually.

  * The new version of the Couvreur emptiness check is now the default
    one, used by twa::is_empty() and twa::accepting_run().  Always
    prefer these functions over an explicit call to Couvreur.

  * experimental support for alternating automata:

    - twa_graph objects can now represent alternating automata.  Use
      twa_graph::new_univ_edge() and twa_graph::set_univ_init_state()
      to create universal edges an initial states; and use
      twa_graph::univ_dests() to iterate over the universal
      destinations of an edge.

    - the automaton parser will now read alternating automata in the
      HOA format.  The HOA and dot printers can output them.

    - the file twaalgos/alternation.hh contains a few algorithms
      specific to alternating automata:
      + remove_alternation() will transform *weak* alternating automata
        into TGBA.
      + the class outedge_combiner can be used to perform "and" and "or"
        on the outgoing edges of some alternating automaton.

    - scc_info has been adjusted to handle universal edges as if they
      were existential edges.  As a consequence, acceptance
      information is not accurate.

    - postprocessor will call remove_alternation() right away, so
      it can be used as a way to transform alternating automata
      into different sub-types of (generalized) Büchi automata.
      Note that although prostprocessor optimize the resulting
      automata, it still has no simplification algorithms that work
      at the alternating automaton level.

    - See https://spot.lrde.epita.fr/tut23.html
      https://spot.lrde.epita.fr/tut24.html and
      https://spot.lrde.epita.fr/tut31.html for some code examples.

  * twa objects have two new properties, very-weak and
    semi-deterministic, that can be set or retrieved via
    twa::prop_very_weak()/twa::prop_semi_deterministic(), and that can
    be tested by is_very_weak_automaton()/is_semi_deterministic().

  * twa::prop_set has a new attribute used in twa::prop_copy() and
    twa::prop_keep() to indicate that determinism may be improved by
    an algorithm.  In other words properties like
    deterministic/semi-deterministic/unambiguous should be preserved
    only if they are positive.

  * language_map() and highlight_languages() are new functions that
    implement autfilt's --highlight-languages option mentionned above.

  * dtgba_sat_minimize_dichotomy() and dwba_sat_minimize_dichotomy()
    use language_map() to estimate a lower bound for binary search.

  * The encoding part of SAT-based minimization consumes less memory.

  * SAT-based minimization of automata can now be done using two
    incremental techniques that take a solved minimization and attempt
    to forbid the use of some states.  This is done either by adding
    clauses, or by using assumptions.

  * If the environment variable "SPOT_XCNF" is set during incremental
    SAT-based minimization, XCNF files suitable for the incremental SAT
    competition will be generated.  This requires the use of an exteral
    SAT solver, setup with "SPOT_SATSOLVER".  See spot-x(7).

  * The new function mp_class(f) returns the class of the formula
    f in the temporal hierarchy of Manna & Pnueli.

Python:

  * spot.show_mp_hierarchy() can be used to display the membership of
    a formula to the Manna & Pnueli hierarchy, in notebooks.  An
    example is in https://spot.lrde.epita.fr/ipynb/formulas.html

  * The on-line translator will now display the temporal hierarchy
    in the "Formula > property information" output.

Bugs fixed:

  * The minimize_wdba() function was not correctly minimizing automata
    with useless SCCs.  This was not an issue for the LTL translation
    (where useless SCCs are always removed first), but it was an issue
    when deciding if a formula was safety or guarantee.  As a
    consequence, some tricky safety or guarantee properties were only
    recognized as obligations.

  * When ltlcross was running a translator taking the Spin syntax as
    input (%s) it would not automatically relabel any unsupported
    atomic propositions as ltldo already do.

  * When running "autfilt --sat-minimize" on a automaton representing
    an obligation property, the result would always be a complete
    automaton even without the -C option.

  * ltlcross --products=0 --csv should not output any product-related
    column in the CSV output since it has nothing to display there.

New in spot 2.2.2 (2016-12-16)

  Build:

  * If the system has an installed libltdl library, use it instead of
    the one we distribute.

  Bug fixes:

  * scc_filter() had a left-over print statement that would print
    "names" when copying the name of the states.

  * is_terminal() should reject automata that have accepting
    transitions going into rejecting SCCs.  The whole point of
    being a terminal automaton is that reaching an accepting
    transition guarantees that any suffix will be accepted.

  * The HOA parser incorrectly read "Acceptance: 1 Bar(0)" as a valid
    way to specify "Acceptance: 1 Fin(0)" because it assumed that
    everything that was not Inf was Fin.  These errors are now
    diagnosed.

  * Some of the installed headers (spot/misc/fixpool.hh,
    spot/misc/mspool.hh, spot/twaalgos/emptiness_stats.hh) were not
    self-contained.

  * ltlfilt --from-ltlf should ensure that "alive" holds initially in
    order to reject empty traces.

  * the on-line translator had a bug where a long ltl3ba process would
    continue running even after the script had timeout'ed.

New in spot 2.2.1 (2016-11-21)

  Bug fix:

  * The bdd_noderesize() function, as modified in 2.2, would always
    crash.

New in spot 2.2 (2016-11-14)

  Command-line tools:

  * ltlfilt has a new option --from-ltlf to help reducing LTLf (i.e.,
    LTL over finite words) model checking to LTL model checking.  This
    is based on a transformation by De Giacomo & Vardi (IJCAI'13).

  * "ltldo --stats=%R", which used to display the serial number of the
    formula processed, was renamed to "ltldo --stats=%#" to free %R
    for the following feature.

  * autfilt, dstar2tgba, ltl2tgba, ltlcross, ltldo learned to measure
    cpu-time (as opposed to wall-time) using --stats=%R.  User or
    system time, for children or parent, can be measured separately by
    adding additional %[LETTER]R options.  The difference between %r
    (wall-clock time) and %R (CPU time) can also be used to detect
    unreliable measurements.  See
      https://spot.lrde.epita.fr/oaut.html#timing

  Library:

  * from_ltlf() is a new function implementing the --from-ltlf
    transformation described above.

  * is_unambiguous() was rewritten in a more efficient way.

  * scc_info learned to determine the acceptance of simple SCCs made
    of a single self-loop without resorting to remove_fin() for complex
    acceptance conditions.

  * remove_fin() has been improved to better deal with automata with
    "unclean" acceptance, i.e., acceptance sets that are declared but
    not used.  In particular, this helps scc_info to be more efficient
    at deciding the acceptance of SCCs in presence of Fin acceptance.

  * Simulation-based reductions now implement just bisimulation-based
    reductions on deterministic automata, to save time.  As an example,
    this halves the run time of
       genltl --rv-counter=10 | ltl2tgba

  * scc_filter() learned to preserve state names and highlighted states.

  * The BuDDy library has been slightly optimized: the initial setup
    of the unicity table can now be vectorized by GCC, and all calls
    to setjmp are now avoided when variable reordering is disabled
    (the default).

  * The twa class has three new methods:
      aut->intersects(other)
      aut->intersecting_run(other)
      aut->intersecting_word(other)
    currently these are just convenient wrappers around
      !otf_product(aut, other)->is_empty()
      otf_product(aut, other)->accepting_run()->project(aut)
      otf_product(aut, other)->accepting_word()
    with any Fin-acceptance removal performed before the product.
    However the plan is to implement these more efficiently in the
    future.

  Bug fixes:

  * ltl2tgba was always using the highest settings for the LTL
    simplifier, ignoring the --low and --medium options.  Now
      genltl --go-theta=12 | ltl2tgba --low --any
    is instantaneous as it should be.

  * The int-vector compression code could encode 6 and 22 in the
    same way, causing inevitable issues in our LTSmin interface.
    (This affects tests/ltsmin/modelcheck with option -z, not -Z.)

  * str_sere() and str_utf8_sere() were not returning the same
    string that print_sere() and print_utf8_sere() would print.

  * Running the LTL parser in debug mode would crash.

  * tgba_determinize() could produce incorrect deterministic automata
    when run with use_simulation=true (the default) on non-simplified
    automata.

  * When the automaton_stream_parser reads an automaton from an
    already opened file descriptor, it will not close the file
    anymore.  Before that the spot.automata() python function used to
    close a file descriptor twice when reading from a pipe, and this
    led to crashes of the 0MQ library used by Jupyter, killing the
    Python kernel.

  * remove_fin() could produce incorrect result on incomplete
    automata tagged as weak and deterministic.

  * calling set_acceptance() several times on an automaton could
    result in unexpected behaviors, because set_acceptance(0,...)
    used to set the state-based acceptance flag automatically.

  * Some buffering issue caused syntax errors to be displayed out
    of place by the on-line translator.

New in spot 2.1.2 (2016-10-14)

  Command-line tools:

  * genltl learned 5 new families of formulas
    (--tv-f1, --tv-f2, --tv-g1, --tv-g2, --tv-uu)
    defined in Tabakov & Vardi's RV'10 paper.

  * ltlcross's --csv and --json output was changed to not include
    information about the ambiguity or strength of the automata by
    default.   Computing those can be costly and not needed by
    every user, so it should now be requested explicitely using
    options --strength and --ambiguous.

  Library:

  * New LTL simplification rule:

    - GF(f & q) = G(F(f) & q) if q is
      purely universal and a pure eventuality.  In particular
      GF(f & GF(g)) now ultimately simplifies to G(F(f) & F(g)).

  Bug fixes:

  * Fix spurious uninitialized read reported by valgrind when
    is_Kleene_star() is compiled by clang++.

  * Using "ltlfilt some-large-file --some-costly-filter" could take
    to a lot of time before displaying the first results, because the
    output of ltlfilt is buffered: the buffer had to fill up before
    being flushed.  The issue did not manifest when the input is
    standard input, because of the C++ feature that reading std::cin
    should flush std::cout; however it was well visible when reading
    from files.  Flushing is now done more regularly.

  * Fix compilation warnings when -Wimplicit-fallthrough it enabled.

  * Fix python errors on Darwin when using methods from the spot module
    inside of the spot.ltsmin submodule.

  * Fix ltlcross crash when combining --no-check with --verbose.

  * Adjust paths and options used in bench/dtgbasat/ to match the
    changes introduced in Spot 2.0.

New in spot 2.1.1 (2016-09-20)

  Command-line tools:

  * ltlfilt, randltl, genltl, and ltlgrind learned to display the size
    (%s), Boolean size (%b), and number of atomic propositions (%a)
    with the --format and --output options.  A typical use-case is to
    sort formulas by size:
       genltl --dac --format='%s,%f' | sort -n | cut -d, -f2
    or to group formulas by number of atomic propositions:
       genltl --dac --output='ap-%a.ltl'

  * autfilt --stats learned the missing %D, %N, %P, and %W sequences,
    to complete the existing %d, %n, %p, and %w.

  * The --stats %c option of ltl2tgba, autfilt, ltldo, and dstar2tgba
    now accepts options to filter the SCCs to count.  For instance
    --stats='%[awT]c' will count the SCCs that are (a)ccepting and
    (w)eak, but (not t)erminal.  See --help for all supported filters.

  Bugs fixed:

  * Fix several cases where command-line tools would fail to diagnose
    write errors (e.g. when writing to a filesystem that is full).
  * Typos in genltl --help and in the man page spot-x(7).

New in spot 2.1 (2016-08-08)

  Command-line tools:

  * All tools that input formulas or automata (i.e., autfilt,
    dstar2tgba, ltl2tgba, ltl2tgta, ltlcross, ltldo, ltlfilt,
    ltlgrind) now have a more homogeneous handling of the default
    input.

    - If no formula/automaton have been specified, and the standard
    input is a not a tty, then the default is to read that. This is a
    change for ltl2tgba and ltl2tgta.  In particular, it simplifies

        genltl --dac | ltl2tgba -F- | autfilt ...

    into

        genltl --dac | ltl2tgba | autfilt ...

    - If standard input is a tty and no other input has been
    specified, then an error message is printed.  This is a change for
    autfilt, dstar2tgba, ltlcross, ltldo, ltlfilt, ltlgrind, that used
    to expect the user to type formula or automata at the terminal,
    confusing people.

    - All tools now accept - as a shorthand for -F-, to force reading
    input from the standard input (regardless of whether it is a tty
    or not).  This is a change for ltl2tgba, ltl2tgta, ltlcross, and
    ltldo.

  * ltldo has a new option --errors=... to specify how to deal
    with errors from executed tools.

  * ltlcross and ltldo learned to bypass the shell when executing
    simple commands (with support for single or double-quoted
    arguments, and redirection of stdin and stdout, but nothing more).

  * ltlcross and ltldo learned a new syntax to specify that an input
    formula should be written in some given syntax after rewriting
    some operators away.  For instance the defaults arguments passed
    to ltl2dstar have been changed from
         --output-format=hoa %L %O
    into
         --output-format=hoa %[WM]L %O
    where [WM] specifies that operators W and M should be rewritten
    away.  As a consequence running
         ltldo ltl2dstar -f 'a M b'
    will now work and call ltl2dstar on the equivalent formula
    'b U (a & b)' instead.  The operators that can be listed between
    brackets are the same as those of ltlfilt --unabbreviate option.

  * ltlcross learned to show some counterexamples when diagnosing
    failures of cross-comparison checks against random state spaces.

  * autfilt learned to filter automata by count of SCCs (--sccs=RANGE)
    or by type of SCCs (--accepting-sccs=RANGE,
    --rejecting-sccs=RANGE, trivial-sccs=RANGE, --terminal-sccs=RANGE,
    --weak-sccs=RANGE, --inherently-weak-sccs=RANGE).

  * autfilt learned --remove-unused-ap to remove atomic propositions
    that are declared in the input automaton, but not actually used.
    This of course makes sense only for input/output formats that
    declare atomic propositions (HOA & DSTAR).

  * autfilt learned two options to filter automata by count of used or
    unused atomic propositions: --used-ap=RANGE --unused-ap=RANGE.
    These differ from --ap=RANGE that only consider *declared* atomic
    propositions, regardless of whether they are actually used.

  * autfilt learned to filter automata by count of nondeterministsic
    states with --nondet-states=RANGE.

  * autfilt learned to filter automata representing stutter-invariant
    properties with --is-stutter-invariant.

  * autfilt learned two new options to highlight non-determinism:
    --highlight-nondet-states=NUM and --highlight-nondet-states=NUM
    where NUM is a color number.  Additionally --highlight-nondet=NUM is
    a shorthand for using the two.

  * autfilt learned to highlight a run matching a given word using the
    --highlight-word=[NUM,]WORD option.  However currently this only
    work on automata with Fin-less acceptance.

  * autfilt learned two options --generalized-rabin and
    --generalized-streett to convert the acceptance conditions.

  * genltl learned three new families: --dac-patterns=1..45,
    --eh-patterns=1..12, and --sb-patterns=1..27.  Unlike other options
    these do not output scalable patterns, but simply a list of formulas
    appearing in these three papers: Dwyer et al (FMSP'98), Etessami &
    Holzmann (Concur'00), Somenzi & Bloem (CAV'00).

  * genltl learned two options, --positive and --negative, to control
    wether formulas should be output after negation or not (or both).

  * The formater used by --format (for ltlfilt, ltlgrind, genltl,
    randltl) or --stats (for autfilt, dstar2tgba, ltl2tgba, ltldo,
    randaut) learned to recognize double-quoted fields and double the
    double-quotes output inbetween as expected from RFC4180-compliant
    CSV files.  For instance
      ltl2tgba -f 'a U "b+c"' --stats='"%f",%s'
    will output
      "a U ""b+c""",2

  * The --csv-escape option of genltl, ltlfilt, ltlgrind, and randltl
    is now deprecated.  The option is still here, but hidden and
    undocumented.

  * The --stats option of autfilt, dstar2tgba, ltl2tgba, ltldo,
    randaut learned to display the output (or input if that makes
    sense) automaton as a HOA one-liner using %h (or %H), helping to
    create CSV files contained automata.

  * autfilt and dstar2tgba learned to read automata from columns in
    CSV files, specified using the same filename/COLUMN syntax used by
    tools reading formulas.

  * Arguments passed to -x (in ltl2tgba, ltl2tgta, autfilt, dstar2tgba)
    that are not used are now reported as they might be typos.
    This ocurred a couple of times in our test-suite.  A similar
    check is done for the arguments of autfilt --sat-minimize=...

  Library:

  * The print_hoa() function will now output version 1.1 of the HOA
    format when passed the "1.1" option (i.e., use -H1.1 from any
    command-line tool).  As far as Spot is concerned, this allows
    negated properties to be expressed.  Version 1 of the HOA format
    is still the default, but we plan to default to version 1.1 in the
    future.

  * The "highlight-states" and "highlight-edges" named properties,
    which were introduced in 1.99.8, will now be output using
    "spot.highlight.edges:" and "spot.highlight.states:" headers if
    version 1.1 of the HOA format is selected.  The automaton parser
    was secretly able of reading that since 1.99.8, but that is now
    documented at https://spot.lrde.epita.fr/hoa.html#extensions

  * highlight_nondet_states() and highlight_nondet_edges() are
    new functions that define the above two named properties.

  * is_deterministic(), is_terminal(), is_weak(), and
    is_inherently_weak(), count_nondet_states(),
    highlight_nondet_edges(), highlight_nondet_states() will update
    the corresponding properties of the automaton as a side-effect of
    their check.

  * the sbacc() function, used by "ltl2tgba -S" and "autfilt -S" to
    convert automata to state-based acceptance, learned some tricks
    (using SCCs, pulling accepting marks common to all outgoing edges,
    and pushing acceptance marks common to all incoming edges) to
    reduce the number of additional states needed.

  * to_generalized_rabin() and to_generalized_streett() are two new
    functions that convert the acceptance condition as requested
    without changing the transition structure.

  * language_containment_checker now has default values for all
    parameters of its constructor.

  * spot::twa_run has a new method, project(), that can be used to
    project a run found in a product onto one of the original operand.
    This is for instance used by autfilt --highlight-word.

    The old spot::project_twa_run() function has been removed: it was
    not used anywhere in Spot, and had an obvious bug in its
    implementation, so it cannot be missed by anyone.

  * spot::twa has two new methods that supplement is_empty():
    twa::accepting_run() and twa::accepting_word().  They compute
    what their names suggest.  Note that twa::accepting_run(), unlike
    the two others, is currently restricted to automata with Fin-less
    acceptance.

  * spot::check_stutter_invariance() can now work on non-deterministic
    automata for which no corresponding formula is known.  This
    implies that "autfilt --check=stutter" will now label all
    automata, not just deterministic automata.

  * New LTL and PSL simplification rules:
    - if e is pure eventuality and g => e, then e U g = Fg
    - if u is purely universal and u => g, then u R g = Gg
    - {s[*0..j]}[]->b = {s[*1..j]}[]->b
    - {s[*0..j]}<>->b = {s[*1..j]}<>->b

  * spot::twa::succ_iterable renamed to
    spot::internal::twa_succ_iterable to make it clear this is not for
    public consumption.

  * spot::fair_kripke::state_acceptance_conditions() renamed to
    spot::fair_kripke::state_acceptance_mark() for consistency.  This
    is backward incompatible, but we are not aware of any actual use
    of this method.

  Python:

  * The __format__() method for formula supports the same
    operator-rewritting feature introduced in ltldo and ltlcross.
    So "{:[i]s}".format(f) is the same as
    "{:s}".format(f.unabbreviate("i")).

  * Bindings for language_containment_checker were added.

  * Bindings for randomize() were added.

  * Iterating over edges via "aut.out(s)" or "aut.edges()"
    now allows modifying the edge fields.

  * Under IPython the spot.ltsmin module now offers a
    %%pml magic to define promela models, compile them
    with spins, and dynamically load them.   This is
    akin to the %%dve magic that was already supported.

  * The %%dve and %%pml magics honor the SPOT_TMPDIR and
    TMPDIR environment variables.  This especially helps
    when the current directory is read-only.

  Documentation:

  * A new example page shows how to test the equivalence of
    two LTL/PSL formulas.  https://spot.lrde.epita.fr/tut04.html

  * A new page discusses explicit vs. on-the-fly interfaces for
    exploring automata in C++.  https://spot.lrde.epita.fr/tut50.html

  * Another new page shows how to implement an on-the-fly Kripke
    structure for a custom state space.
    https://spot.lrde.epita.fr/tut51.html

  * The concepts.html page now lists all named properties
    used by automata.

  Bug fixes:

  * When ltlcross found a bug using a product of complemented
    automata, the error message would report "Comp(Ni)*Comp(Pj)" as
    non-empty while the actual culprit was "Comp(Nj)*Comp(Pi)".

  * Fix some non-deterministic execution of minimize_wdba(), causing
    test-suite failures with the future G++ 7, and clang 3.9.

  * print_lbtt() had a memory leak when printing states without
    successors.

New in spot 2.0.3 (2016-07-11)

  Bug fixes:

  * The degen-lcache=1 option of the degeneralization algorithm (which
    is a default option) did not behave exactly as documented: instead
    of reusing the first level ever created for a state where the
    choice of the level is free, it reused the last level ever used.
    This caused some posterior simulation-based reductions to be less
    efficient at reducing automata (on the average).
  * The generalized testing automata displayed by the online
    translator were incorrect (those output by ltl2tgta were OK).
  * ltl2tgta should not offer options --ba, --monitor, --tgba and such.
  * the relabel() function could incorrectly unregister old atomic
    propositions even when they were still used in the output (e.g.,
    if a&p0 is relabeled to p0&p1).  This could cause ltldo and the
    online translator to report errors.

New in spot 2.0.2 (2016-06-17)

  Documentation:

  * We now have a citing page at https://spot.lrde.epita.fr/citing.html
    providing a list of references about Spot.
  * The Python examples have been augmented with the two examples
    from our ATVA'16 tool paper.

  Bug fixes:

  * Fix compilation error observed with Clang++ 3.7.1 and GCC 6.1.1
    headers.
  * Fix an infinite recursion in relabel_bse().
  * Various small typos and cosmetic cleanups.

New in spot 2.0.1 (2016-05-09)

  Library:

  * twa::unregister_ap() and twa_graph::remove_unused_ap() are new
    methods introduced to fix some of the bugs listed below.

  Documentation:

  * Add missing documentation for the option string passed to
    spot::make_emptiness_check_instantiator().

  * There is now a spot(7) man page listing all installed
    command-line tools.

  Python:

  * The tgba_determinize() function is now accessible in Python.

  Bug fixes:

  * The automaton parser would choke on comments like /******/.
  * check_strength() should also set negated properties.
  * Fix autfilt to apply --simplify-exclusive-ap only after
    the simplifications of (--small/--deterministic) have
    been performed.
  * The automaton parser did not fully register atomic propositions
    for automata read from never claim or as LBTT.
  * spot::ltsmin::kripke() had the same issue.
  * The sub_stats_reachable() function used to count the number
    of transitions based on the number of atomic propositions
    actually *used* by the automaton instead of using the number
    of AP declared.
  * print_hoa() will now output all the atomic propositions that have
    been registered, not only those that are used in the automaton.
    (Note that it will also throw an exception if the automaton uses
    an unregistered AP; this is how some of the above bugs were
    found.)
  * For Small or Deterministic preference, the postprocessor
    will now unregister atomic propositions that are no longer
    used in labels.   Simplification of exclusive properties
    and remove_ap::strip() will do similarly.
  * bench/ltl2tgba/ was not working since the source code
    reorganization of 1.99.7.
  * Various typos and minor documentation fixes.


New in spot 2.0 (2016-04-11)

  Command-line tools:

  * ltlfilt now also support the --accept-word=WORD and
    --reject-word=WORD options that were introduced in autfilt in the
    previous version.

  Python:

  * The output of spot.atomic_prop_collect() is printable and
    can now be passed directly to spot.ltsmin.model.kripke().

  Library:

  * digraph::valid_trans() renamed to digraph::is_valid_edge().

  Documentation:

  * The concepts page (https://spot.lrde.epita.fr/concepts.html) now
    includes a highlevel description of the architecture, and some
    notes aboute automata properties.

  * More Doxygen documentation for spot::formula and spot::digraph.

New in spot 1.99.9 (2016-03-14)

  Command-line tools:

  * autfilt has two new options: --accept-word=WORD and
    --reject-word=WORD for filtering automata that accept or reject
    some word.  The option may be used multiple times.

  Library:

  * The parse_word() function can be used to parse a lasso-shaped
    word and build a twa_word.  The twa_word::as_automaton()
    method can be used to create an automaton out of that.
  * twa::ap_var() renamed to twa::ap_vars().
  * emptiness_check_instantiator::min_acceptance_conditions() and
    emptiness_check_instantiator::max_acceptance_conditions() renamed
    to emptiness_check_instantiator::min_sets() and
    emptiness_check_instantiator::max_sets().
  * tgba_reachable_iterator (and subclasses) was renamed to
    twa_reachable_iterator for consistency.

  Documentation:

  * The page https://spot.lrde.epita.fr/upgrade2.html should help
    people migrating old C++ code written for Spot 1.2.x, and update
    it for (the upcoming) Spot 2.0.

  Bug fixes:

  * spot/twaalgos/gtec/gtec.hh was incorrectly installed as
  spot/tgbaalgos/gtec/gtec.hh.
  * The shared libraries should now compile again on Darwin.

New in spot 1.99.8 (2016-02-18)

  Command-line tools:

  * ltl2tgba now also support the --generic option (already supported
    by ltldo and autfilt) to lift any restriction on the acceptance
    condition produced.  This option now has a short version: -G.

  * ltl2tgba and autfilt have learnt how to determinize automata.
    For this to work, --generic acceptance should be enabled (this
    is the default for autfilt, but not for ltl2tgba).

    "ltl2tgba -G -D" will now always outpout a deterministic automaton.
    It can be an automaton with transition-based parity acceptance in
    case Spot could not find a deterministic automaton with (maybe
    generalized) Büchi acceptance.

    "ltl2tgba -D" is unchanged (the --tgba acceptance is the default),
    and will output a deterministic automaton with (generalized) Büchi
    acceptance only if one could be found.  Otherwise a
    non-deterministic automaton is output, but this does NOT mean that
    no deterministic Büchi automaton exist for this formula.  It only
    means Spot could not find it.

    "autfilt -D" will determinize any automaton, because --generic
    acceptance is the default for autfilt.

    "autfilt -D --tgba" will behave like "ltl2tgba -D", i.e., it may
    fail to find a deterministic automaton (even if one exists) and
    return a nondeterministic automaton.

  * "autfilt --complement" now also works for non-deterministic
    automata but will output a deterministic automaton.
    "autfilt --complement --tgba" will likely output a
    nondeterministic TGBA.

  * autfilt has a new option, --included-in, to filter automata whose
    language are included in the language of a given automaton.

  * autfilt has a new option, --equivalent-to, to filter automata
    that are equivalent (language-wise) to a given automaton.

  * ltlcross has a new option --determinize to instruct it to
    complement non-deterministic automata via determinization.  This
    option is not enabled by default as it can potentially be slow and
    generate large automata.  When --determinize is given, option
    --product=0 is implied, since the tests based on products with
    random state-space are pointless for deterministic automata.

  * ltl2tgba and ltldo now support %< and %> in the string passed
    to --stats when reading formulas from a CSV file.

  * ltlfilt's option --size-min=N, --size-max=N, --bsize-min=N, and
    --bsize-max=N have been reimplemented as --size=RANGE and
    --bsize=RANGE.  The old names are still supported for backward
    compatibility, but they are not documented anymore.

  * ltlfilt's option --ap=N can now take a RANGE as parameter.

  * autfilt now has a --ap=RANGE option to filter automata by number
    of atomic propositions.

  Library:

  * Building products with different dictionaries now raise an
    exception instead of using an assertion that could be disabled.

  * The load_ltsmin() function has been split in two.  Now you should
    first call ltsmin_model::load(filename) to create an ltsmin_model,
    and then call the ltsmin_model::kripke(...) method to create an
    automaton that can be iterated on the fly.  The intermediate
    object can be queried about the supported variables and their
    types.

  * print_dot() now accepts several new options:
    - use "<N" to specify a maximum number of states to output.
      Incomplete states are then marked appropriately.  For a quick
      example, compare
	ltl2tgba -D 'Ga | Gb | Gc' -d'<3'
      with
        ltl2tgba -D 'Ga | Gb | Gc' -d
    - use "C(color)" to specify the color to use for filling states.
    - use "#" to display the internal number of each transition
    - use "k" to use state-based labels when possible.  This is
      similar to the "k" option already supported by print_hoa(), and
      is useful when printing Kripke structures.

  * Option "k" is automatically used by print_dot() and print_hoa()
    when printing Kripke structures.

  * print_dot() also honnor two new automaton properties called
    "highlight-edges" and "highlight-states".  These are used to color
    a subset of edges or transitions.

  * There is a new tgba_determinize() function.  Despite its name, it
    in facts works on transition-based Büchi automaton, and will first
    degeneralize any automaton with generalized Büchi acceptance.

  * The twa_safra_complement class has been removed.  Use
    tgba_determinize() and dtwa_complement() instead.

  * The twa::transition_annotation() and
    twa::compute_support_conditions() methods have been removed.

  * The interface for all functions parsing formulas (LTL, PSL, SERE,
    etc.) has been changed to use an interface similar to the one used
    for parsing automata.  These function now return a parsed_formula
    object that includes both a formula and a list of syntax errors.

    Typically a function written as

       spot::formula read(std::string input)
       {
	  spot::parse_error_list pel;
	  spot::formula f = spot::parse_infix_psl(input, pel);
	  if (spot::format_parse_errors(std::cerr, input, pel))
	    exit(2);
	  return f;
       }

    should be updated to

       spot::formula read(std::string input)
       {
	  spot::parsed_formula pf = spot::parse_infix_psl(input);
	  if (pf.format_errors(std::cerr))
	    exit(2);
	  return pf.f;
       }

  Python:

  * The ltsmin interface has been binded in Python.  It also
    comes with a %%dve cell magic to edit DiVinE models in the notebook.
    See https://spot.lrde.epita.fr/ipynb/ltsmin.html for a short example.

  * spot.setup() sets de maximum number of states to display in
    automata to 50 by default, as more states is likely to be
    unreadable (and slow to process by GraphViz).  This can be
    overridden by calling spot.setup(max_states=N).

  * Automata now have methods to color selected states and
    transitions.  See
    https://spot.lrde.epita.fr/ipynb/highlighting.html for an example.

  Documentation:

  * There is a new page giving informal illustrations (and extra
    pointers) for some concepts used in Spot.
    See https://spot.lrde.epita.fr/concepts.html

  Bug fixes:

  * Using ltl2tgba -U would fail to output the unambiguous property
    (regression introduced in 1.99.7)
  * ltlfilt, autfilt, randltl, and randaut could easily crash when
    compiled statically (i.e., with configure --disable-shared).
  * "1 U (a | Fb)" was not always simplified to "F(a | b)".
  * destroying the operands of an otf_product() before the product
    itself could crash.

New in spot 1.99.7 (2016-01-15)

  Command-line tools:

  * BACKWARD INCOMPATIBLE CHANGE: All tools that output automata now
    use the HOA format by default instead of the GraphViz output.
    This makes it easier to pipe several commands together.

    If you have an old script that relies on GraphViz being the default
    output and that you do not want to update it, use
       export SPOT_DEFAULT_FORMAT=dot
    to get the old behavior back.

  * Tools that output automata now accept -d as a shorthand for --dot
    since requesting the GraphViz (a.k.a. dot) output by hand is now
    more frequent.
    randaut's short option for specifying edge-density used to be -d:
    it has been renamed to -e.

  * The SPOT_DEFAULT_FORMAT environment variable can be set to 'dot'
    or 'hoa' to force a default output format for automata.  Additional
    options may also be added, as in SPOT_DEFAULT_FORMAT='hoa=iv'.

  * autfilt has a new option: --is-inherently-weak.

  * The --check=strength option of all tools that produce automata
    will also test if an automaton is inherently weak.

  Library:

  * Installed headers now assume that they will be included as
    #include <spot/subdir/header.hh>
    instead of
    #include <subdir/header.hh>

    This implies that when Spot headers are installed in
    /usr/include/spot/... (the default when using the Debian packages)
    or /usr/local/include/spot/... (the default when compiling from
    source), then it is no longuer necessary to add
    -I/usr/include/spot or -I/usr/local/include/spot when compiling,
    as /usr/include and /usr/local/include are usually searched by
    default.

    Inside the source distribution, the subdirectory src/ has been
    renamed to spot/, so that the root of the source tree can also be
    put on the preprocessor's search path to compile against a
    non-installed version of Spot.  Similarly, iface/ltsmin/ has been
    renamed to spot/ltsmin/, so that installed and non-installed
    directories can be used similarly.

  * twa::~twa() is now calling
      get_dict()->unregister_all_my_variable(this);
    so this does not need to be done in any subclass.

  * is_inherently_weak_automaton() is a new function, and
    check_strength() has been modified to also check inherently weak
    automata.

  * decompose_strength() is now extracting inherently weak SCCs
    instead of just weak SCCs.  This gets rid of some corner cases
    that used to require ad hoc handling.

  * acc_cond::acc_code's methods append_or(), append_and(), and
    shift_left() have been replaced by operators |=, &=, <<=, and for
    completeness the operators |, &, and << have been added.

  * Several methods have been removed from the acc_cond
    class because they were simply redundant with the methods of
    acc_cond::mark_t, and more complex to use.

       acc_cond::marks(...)      -> use acc_cond::mark_t(...)
       acc_cond::sets(m)         -> use m.sets()
       acc_cond::has(m, u)       -> use m.has(u)
       acc_cond::cup(l, r)       -> use l | r
       acc_cond::cap(l, r)       -> use l & r
       acc_cond::set_minus(l, r) -> use l - r

    Additionally, the following methods/functions have been renamed:

       acc_cond::is_tt() -> acc_cond::is_t()
       acc_cond::is_ff() -> acc_cond::is_f()
       parse_acc_code()  -> acc_cond::acc_code(...)

  * Automata property flags (those that tell whether the automaton is
    deterministic, weak, stutter-invariant, etc.) are now stored using
    three-valued logic: in addition to "maybe"/"yes" they can now also
    represent "no".  This is some preparation for the upcomming
    support of the HOA v1.1 format, but it also saves time in some
    algorithms (e.g, is_deterministic() can now return immediately on
    automata marked as not deterministic).

  * The automaton parser now accept negated properties as they will be
    introduced in HOA v1.1, and will check for some inconsistencies.
    These properties are stored and used when appropriate, but they
    are not yet output by the HOA printer.

  Python:

  * Iterating over the transitions leaving a state (the
    twa_graph::out() C++ function) is now possible in Python.  See
    https://spot.lrde.epita.fr/tut21.html for a demonstration.

  * Membership to acceptance sets can be specified using Python list
    when calling for instance the Python version of twa_graph::new_edge().
    See https://spot.lrde.epita.fr/tut22.html for a demonstration.

  * Automaton states can be named via the set_state_names() method.
    See https://spot.lrde.epita.fr/ipynb/product.html for an example.

  Documentation:

  * There is a new page explaining how to compile example programs and
    and link them with Spot.  https://spot.lrde.epita.fr/compile.html

  * Python bindings for manipulating acceptance conditions are
    demonstrated by https://spot.lrde.epita.fr/ipynb/acc_cond.html,
    and a Python implementation of the product of two automata is
    illustrated by https://spot.lrde.epita.fr/ipynb/product.html

  Source code reorganisation:

  * A lot of directories have been shuffled around in the
    distribution:
      src/                   ->  spot/    (see rational above)
      iface/ltsmin/  (code)  ->  spot/ltsmin/
      wrap/python/           ->  python/
      src/tests/             ->  tests/core/
      src/sanity/            ->  tests/sanity/
      iface/ltsmin/  (tests) ->  tests/ltsmin/
      wrap/python/tests      ->  tests/python/

  Bug fixes:

  * twa_graph would incorrectly replace named-states during
    purge_dead_states and purge_unreachable_states.
  * twa::ap() would contain duplicates when an atomic proposition
    was registered several times.
  * product() would incorrectly mark the product of two
    sttuter-sensitive automata as stutter-sensitive as well.
  * complete() could incorrectly reuse an existing (but accepting!)
    state as a sink.

New in spot 1.99.6 (2015-12-04)

  Command-line tools:

  * autfilt has two new filters: --is-weak and --is-terminal.

  * autfilt has a new transformation: --decompose-strength,
    implementing the decomposition of our TACAS'13 paper.
    A demonstration of this feature via the Python bindings
    can be found at https://spot.lrde.epita.fr/ipynb/decompose.html

  * All tools that output HOA files accept a --check=strength option
    to request automata to be marked as "weak" or "terminal" as
    appropriate.  Without this option, these properties may only be
    set as a side-effect of running transformations that use this
    information.

  Library:

  * Properties of automata (like the "properties:" line of the HOA
    format) are stored as bits whose interpretation is True=yes,
    False=maybe.  Having getters like "aut->is_deterministic()" or
    "aut->is_unambiguous()" was confusing, because there are separate
    functions "is_deterministic(aut)" and "is_unambiguous(aut)" that
    do actually check the automaton.  The getters have been renamed to
    avoid confusion, and get names more in line with the HOA format.

    - twa::has_state_based_acc() -> twa::prop_state_acc()
    - twa::prop_state_based_acc(bool) -> twa::prop_state_acc(bool)
    - twa::is_inherently_weak() -> twa::prop_inherently_weak()
    - twa::is_deterministic() -> twa::prop_deterministic()
    - twa::is_unambiguous() -> twa::prop_unambiguous()
    - twa::is_stutter_invariant() -> twa::prop_stutter_invariant()
    - twa::is_stutter_sensitive() -> twa::prop_stutter_sensitive()

    The setters have the same names as the getters, except they take a
    Boolean argument.  This argument used to be optionnal (defaulting
    to True), but it no longer is.

  * Automata now support the "weak" and "terminal" properties in
    addition to the previously supported "inherently-weak".

  * By default the HOA printer tries not to bloat the output with
    properties that are redundant and probably useless.  The HOA
    printer now has a new option "v" (use "-Hv" from the command line)
    to output more verbose "properties:".  This currently includes
    outputing "no-univ-branch", outputting "unambiguous" even for
    automata already tagged as "deterministic", and "inherently-weak"
    or "weak" even for automata tagged "weak" or "terminal".

  * The HOA printer has a new option "k" (use "-Hk" from the command
    line) to output automata using state labels whenever possible.
    This is useful to print Kripke structures.

  * The dot output will display pair of states when displaying an
    automaton built as an explicit product.  This works in IPython
    with spot.product() or spot.product_or() and in the shell with
    autfilt's --product or --product-or options.

  * The print_dot() function supports a new option, +N, where N is a
    positive integer that will be added to all set numbers in the
    output.  This is convenient when displaying two automata before
    building their product: use +N to shift the displayed sets of the
    second automaton by the number of acceptance sets N of the first
    one.

  * The sat minimization for DTωA now does a better job at selecting
    reference automata when the output acceptance is the the same as
    the input acceptance.  This can provide nice speedups when tring
    to syntethise larged automata with different acceptance
    conditions.

  * Explicit Kripke structures (i.e., stored as explicit graphs) have
    been rewritten above the graph class, using an interface similar
    to the twa class.  The new class is called kripke_graph.  The ad
    hoc Kripke parser and printer have been removed, because we can
    now use print_hoa() with the "k" option to print Kripke structure
    in the HOA format, and furthermore the parse_aut() function now
    has an option to load such an HOA file as a kripke_graph.

  * The HOA parser now accepts identifier, aliases, and headernames
    containing dots, as this will be allowed in the next version of
    the HOA format.

  * Renamings:
    is_guarantee_automaton() -> is_terminal_automaton()
    tgba_run -> twa_run
    twa_word::print -> operator<<
    dtgba_sat_synthetize() -> dtwa_sat_synthetize()
    dtgba_sat_synthetize_dichotomy() -> dtwa_sat_synthetize_dichotomy()

  Python:

  * Add bindings for is_unambiguous().
  * Better interface for sat_minimize().

  Bug fixes:

  * the HOA parser was ignoring the "unambiguous" property.
  * --dot=Bb should work like --dot=b, allowing us to disable
    a B option set via an environment variable.

New in spot 1.99.5 (2015-11-03)

  Command-line tools:

  * autfilt has gained a --complement option.
    It currently works only for deterministic automata.

  * By default, autfilt does not simplify automata (this has not
    changed), as if the --low --any options were used.  But now, if
    one of --small, --deterministic, or --any is given, the
    optimization level automatically defaults to --high (unless
    specified otherwise).  For symmetry, if one of --low, --medium, or
    --high is given, then the translation intent defaults to --small
    (unless specified otherwise).

  * autfilt, dstar2tgba, ltlcross, and ltldo now trust the (supported)
    automaton properties declared in any HOA file they read.  This can
    be disabled with option --trust-hoa=no.

  * ltlgrind FILENAME[/COL] is now the same as
    ltlgrind -F FILENAME[/COL] for consistency with ltlfilt.

  Library:

  * dtgba_complement() was renamed to dtwa_complement(), moved to
    complement.hh, and its purpose was restricted to just completing
    the automaton and complementing its acceptance condition.  Any
    further acceptance condition transformation can be done with
    to_generalized_buchi() or remove_fin().

  * The remove_fin() has learnt how to better deal with automata that
    are declared as weak.  This code was previously in
    dtgba_complement().

  * scc_filter_states() has learnt to remove useless acceptance marks
    that are on transitions between SCCs, while preserving state-based
    acceptance.  The most visible effect is in the output of "ltl2tgba
    -s XXXa": it used to have 5 accepting states, it now has only one.
    (Changing removing acceptance of those 4 states has no effect on
    the language, but it speeds up some algorithms like NDFS-based
    emptiness checks, as discussed in our Spin'15 paper.)

  * The HOA parser will diagnose any version that is not v1, unless it
    looks like a subversion of v1 and no parse error was detected.

  * The way to pass option to the automaton parser has been changed to
    make it easier to introduce new options.  One such new option is
    "trust_hoa": when true (the default) supported properties declared
    in HOA files are trusted even if they cannot be easily be verified.
    Another option "raise_errors" now replaces the method
      automaton_stream_parser::parse_strict().

  * The output of the automaton parser now include the list of parse
    errors (that does not have to be passed as a parameters) and the
    input filename (making the output of error messages easier).

  * The following renamings make the code more consistent:
    ltl_simplifier -> tl_simplifier
    tgba_statistics::transitions -> twa_statistics::edges
    tgba_sub_statistics::sub_transitions -> twa_sub_statistics::transitions
    tgba_run -> twa_run
    reduce_run -> twa_run::reduce
    replay_tgba_run -> twa_run::replay
    print_tgba_run -> operator<<
    tgba_run_to_tgba -> twa_run::as_twa
    format_parse_aut_errors -> parsed_aut::format_errors
    twa_succ_iterator::current_state -> twa_succ_iterator::dst
    twa_succ_iterator::current_condition -> twa_succ_iterator::cond
    twa_succ_iterator::current_acceptance_conditions -> twa_succ_iterator::acc
    ta_succ_iterator::current_state -> ta_succ_iterator::dst
    ta_succ_iterator::current_condition -> ta_succ_iterator::cond
    ta_succ_iterator::current_acceptance_conditions -> ta_succ_iterator::acc

  Python:

  * The minimum supported Python version is now 3.3.
  * Add bindings for complete() and dtwa_complement()
  * Formulas now have a custom __format__ function.  See
    https://spot.lrde.epita.fr/tut01.html for examples.
  * The Debian package is now compiled for all Python3 versions
    supported by Debian, not just the default one.
  * Automata now have get_name()/set_name() methods.
  * spot.postprocess(aut, *options), or aut.postprocess(*options)
    simplify the use of the spot.postprocessor object.  (Just like we
    have spot.translate() on top of spot.translator().)
  * spot.automata() and spot.automaton() now have additional
    optional arguments:
    - timeout: to restrict the runtime of commands that
      produce automata
    - trust_hoa: can be set to False to ignore HOA properties
      that cannot be easily verified
    - ignore_abort: can be set to False if you do not want to
      skip automata ended with --ABORT--.

  Documentation:

  * There is a new page showing how to use spot::postprocessor
    to convert any type of automaton to Büchi.
    https://spot.lrde.epita.fr/tut30.html

  Bugs fixed:

  * Work around some weird exception raised when using the
    randltlgenerator under Python 3.5.
  * Recognize "nullptr" formulas as None in Python.
  * Fix compilation of bench/stutter/
  * Handle saturation of formula reference counts.
  * Fix typo in the Python code for the CGI server.
  * "randaut -Q0 1" used to segfault.
  * "ltlgrind -F FILENAME/COL" did not preserve other CSV columns.
  * "ltlgrind --help" did not document FORMAT.
  * unabbreviate could easily use forbidden operators.
  * "autfilt --is-unambiguous" could fail to detect the nonambiguity
    of some automata with empty languages.
  * When parsing long tokens (e.g, state labels representing
    very large strings) the automaton parser could die with
    "input buffer overflow, can't enlarge buffer because scanner uses REJECT"

New in spot 1.99.4 (2015-10-01)

  New features:

  * autfilt's --sat-minimize now takes a "colored" option to constrain
    all transitions (or states) in the output automaton to belong to
    exactly one acceptance sets.  This is useful when targeting parity
    acceptance.

  * autfilt has a new --product-or option.  This builds a synchronized
    product of two (completed of needed) automata in order to
    recognize the *sum* of their languages.  This works by just using
    the disjunction of their acceptance conditions (with appropriate
    renumbering of the acceptance sets).

    For consistency, the --product option (that builds a synchronized
    product that recognizes the *intersection* of the languages) now
    also has a --product-and alias.

  * the parser for ltl2dstar's format has been merged with the parser
    for the other automata formats.  This implies two things:
    - autfilt and dstar2tgba (despite its name) can now both read
      automata written in any of the four supported syntaxes
      (ltl2dstar's, lbtt's, HOA, never claim).
    - "dstar2tgba some files..." now behaves exactly like
      "autfilt --tgba --high --small some files...".
      (But dstar2tgba does not offer all the filtering and
      transformations options of autfilt.)

  Major code changes and reorganization:

  * The class hierarchy for temporal formulas has been entirely
    rewritten.  This change is actually quite massive (~13200 lines
    removed, ~8200 lines added), and brings some nice benefits:
    - LTL/PSL formulas are now represented by lightweight formula
      objects (instead of pointers to children of an abstract
      formula class) that perform reference counting automatically.
    - There is no hierachy anymore: all operators are represented by
      a single type of node in the syntax tree, and an enumerator is
      used to distinguish between operators.
    - Visitors have been replaced by member functions such as map()
      or traverse(), that take a function (usually written as a
      lambda function) and apply it to the nodes of the tree.
    - As a consequence, writing algorithms that manipulate formula
      is more friendly, and several algorithms that spanned a few
      pages have been reduced to a few lines.
    The page https://spot.lrde.epita.fr/tut03.html illustrates the
    new interface, in both C++ and Python.

  * Directories ltlast/, ltlenv/, and ltlvisit/, have been merged into
    a single tl/ directory (for temporal logic).  This is motivated by
    the fact that these formulas are not restricted to LTL, and by the
    fact that we no longuer use the "visitor" pattern.

  * The LTL/PSL parser is now declared in tl/parse.hh (instead of
    ltlparse/public.hh).

  * The spot::ltl namespace has been merged with the spot namespace.

  * The dupexp_dfs() function has been renamed to copy(), and has
    learned to preserve named states if required.

  * Atomic propositions can be declared without going through an
    environment using the spot::formula::ap() static function.  They
    can be registered for an automaton directly using the
    spot::twa::register_ap() method.  The vector of atomic
    propositions used by an automaton can now be retrieved using the
    spot::twa::ap() method.

New in spot 1.99.3 (2015-08-26)

  * The CGI script for LTL translation offers a HOA download link
    for each generated automaton.

  * The html documentation now includes a HTML copies of the man
    pages, and HTML copies of the Python notebooks.

  * scc_filter(aut, true) does not remove Fin marks from rejecting
    SCCs, but it now does remove Fin marks from transitions between
    SCCs.

  * All the unabbreviation functions (unabbreviate_ltl(),
    unabbreviate_logic(), unabbreviate_wm()) have been merged into a
    single unabbreviate() function that takes a string representing
    the list of operators to remove among "eFGiMRW^" where 'e', 'i',
    and '^' stand respectively for <->, ->, and xor.
    This feature is also available via ltlfilt --unabbreviate.

  * In LTL formulas, atomic propositions specified using double-quotes
    can now include \" and \\.  (This is more consistent with the HOA
    format, which already allows that.)

  * All the conversion routines that were written specifically for
    ltl2dstar's output format (DRA->BA & DRA->TGBA) have been ported
    to the new TωA structure supporting the HOA format.  The DRA->TGBA
    conversion was reimplemented in the previous release, and the
    DRA->BA conversion has been reimplemented in this release (but it
    is still restricted to state-based acceptance).  All these
    conversions are called automatically by to_generalized_buchi()
    or remove_fin() so there should be no need to call them directly.

    As a consequence:
    - "autfilt --remove-fin" or "autfilt -B" is better at converting
      state-based Rabin automata: it will produce a DBA if the input is
      deterministic and DBA-realizable, but will preserve as much
      determinism as possible otherwise.
    - a lot of obsolete code that was here only to support the old
      conversion routines has been removed.  (The number of lines
      removed by this release is twice the number of lines added.)
    - ltlcross now uses automata in ltl2dstar's format directly,
      without converting them to Büchi (this makes the statistics
      reported in CSV files more relevant).
    - ltlcross no longer outputs additional columns about the size
      of the input automaton in the case ltl2dstar's format is used.
    - ltldo uses results in ltl2dstar's format directly, without
      converting them to Büchi.
    - dstar2tgba has been greatly simplified and now uses the
      same output routines as all the other tools that output
      automata.  This implies a few minor semantic changes, for
      instance --stats=%A used to output the number of acceptance
      *pairs* in the input automaton, while it now outputs the
      number of acceptance sets like in all the other tools.

  * Bugs fixed
    - Some acceptance conditions like Fin(0)|Fin(1)|Fin(2)&Inf(3)
      were not detected as generalized-Rabin.
    - Unknown arguments for print_hoa() (i.e., option -H in command-line
      tools) are now diagnosed.
    - The CGI script for LTL translation now forces transition-based
      acceptance on WDBA-minimized automata when TGBA is requested.
    - ltlgrind --help output had some options documented twice, or
      in the wrong place.
    - The man page for ltlcross had obsolete examples.
    - When outputting atomic propositions in double quotes, the
      escaping routine used by the two styles of LaTeX output was
      slightly wrong.  For instance ^ was incorrectly escaped, and
      the double quotes where not always properly rendered.
    - A spurious assertion was triggered by streett_to_generalized_buchi(),
      but only when compiled in DEBUG mode.
    - LTL formula rewritten in Spin's syntax no longer have their ->
      and <-> rewritten away.
    - Fix some warnings reported by the development version of GCC 6.
    - The spot.translate() function of the Python binding had a typo
      preventing the use of 'low'/'medium'/'high' as argument.
    - Fix spurious failure of uniq.test under different locales.
    - ltlcross now recovers from out-of-memory errors during
      state-space products.
    - bitvect.test was failing on 32bit architectures with
      assertions enabled because of a bug in the test case.

New in spot 1.99.2 (2015-07-18)

  * The scc_info object, used to build a map of SCCs while gathering
    additional information, has been simplified and speed up.  One
    test case where ltlcross would take more than 13min (to check the
    translation of one PSL formula) now takes only 75s.

  * streett_to_generalized_buchi() is a new function that implements
    what its name suggests, with some SCC-based optimizations over the
    naive definition.  It is used by the to_generalized_buchi() and
    remove_fin() functions when the input automaton is a Streett
    automaton with more than 3 pairs (this threeshold can be changed
    via the SPOT_STREETT_CONV_MIN environment variable -- see the
    spot-x(7) man page for details).

    This is mainly useful to ltlcross, which has to get rid of "Fin"
    acceptance to perform its checks.  As an example, the Streett
    automaton generatated by ltl2dstar (configured with ltl2tgba) for
    the formula
      !((GFa -> GFb) & (GFc -> GFd))
    has 4307 states and 14 acceptance sets.  The new algorithm can
    translate it into a TGBA with 9754 states and 7 acceptance sets,
    while the default approch used for converting any acceptance
    to TGBA would produce 250967 states and 7 acceptance sets.

  * Bugs fixed:
    - p[+][:*2] was not detected as belonging to siPSL.
    - scc_filter() would incorrectly remove Fin marks from
      rejecting SCCs.
    - the libspotltsmin library is installed.
    - ltlcross and ltldo did not properly quote atomic propositions
      and temporary file names containing a single-quote.
    - a missing Python.h is now diagnosed at ./configure time, with
      the suggestion to either install python3-devel, or run
      ./configure --disable-python.
    - Debian packages for libraries have been split from
      the main Spot package, as per Debian guidelines.

New in spot 1.99.1 (2015-06-23)

  * Major changes motivating the jump in version number

    - Spot now works with automata that can represent more than
      generalized Büchi acceptance.  Older versions were built around
      the concept of TGBA (Transition-based Generalized Büchi
      Automata) while this version now deals with what we call TωA
      (Transition-based ω-Automata).  TωA support arbitrary acceptance
      conditions specified as a Boolean formula of transition sets
      that must be visited infinitely often or finitely often.  This
      genericity allows for instance to represent Rabin or Streett
      automata, as well as some generalized variants of those.

    - Spot has near complete support for the Hanoi Omega Automata
      format.  http://adl.github.io/hoaf/  This formats supports
      automata with the generic acceptance condition described above,
      and has been implemented in a number of third-party tools (see
      http://adl.github.io/hoaf/support.html) to ease their
      interactions.  The only part of the format not yet implemented in
      Spot is the support for alternating automata.

    - Spot is now compiling in C++11 mode.  The set of C++11 features
      we use requires GCC >= 4.8 or Clang >= 3.5.  Although GCC 4.8 is
      more than 2-year old, people with older installations won't be
      able to install this version of Spot.

    - As a consequence of the switches to C++11 and to TωA, a lot of
      the existing C++ interfaces have been renamed, and sometime
      reworked.  This makes this version of Spot not backward
      compatible with Spot 1.2.x.  See below for the most important
      API changes.  Furtheremore, the reason this release is not
      called Spot 2.0 is that we have more of those changes planned.

    - Support for Python 2 was dropped.  We now support only Python
      3.2 or later.  The Python bindings have been improved a lot, and
      include some conveniance functions for better integration with
      IPython's rich display system.  Users familiar with IPython's
      notebooks should have a look at the notebook files in
      wrap/python/tests/*.ipynb

  * Major news for the command-line tools

    - The set of tools installed by spot now consists in the following
      11 commands.  Those marked with a '+' are new in this release.

      - randltl    Generate random LTL/PSL formulas.
      - ltlfilt    Filter and convert LTL/PSL formulas.
      - genltl     Generate LTL formulas from scalable patterns.
      - ltl2tgba   Translate LTL/PSL formulas into Büchi automata.
      - ltl2tgta   Translate LTL/PSL formulas into Testing automata.
      - ltlcross   Cross-compare LTL/PSL-to-Büchi translators.
      + ltlgrind   Mutate LTL/PSL formula.
      - dstar2tgba Convert deterministic Rabin or Streett automata into Büchi.
      + randaut    Generate random automata.
      + autfilt    Filter and convert automata.
      + ltldo      Run LTL/PSL formulas through other tools.

      randaut does not need any presentation: it does what you expect.

      ltlgrind is a new tool that mutates LTL or PSL formulas.  If you
      have a tool that is bogus on some formula that is too large to
      debug, you can use ltlgrind to generate smaller derived formulas
      and see if you can reproduce the bug on those.

      autfilt is a new tool that processes a stream of automata.  It
      allows format conversion, filtering automata based on some
      properties, and general transformations (e.g., change of
      acceptance conditions, removal of useless states, product
      between automata, etc.).

      ltldo is a new tool that runs LTL/PSL formulas through other
      tools, but uses Spot's command-line interfaces for specifying
      input and output.  This makes it easier to use third-party tool
      in a pipe, and it also takes care of some necessary format
      conversion.

    - ltl2tgba has a new option, -U, to produce unambiguous automata.
      In unambiguous automata any word is recognized by at most one
      accepting run, but there might be several ways to reject a word.
      This works for LTL and PSL formulas.

    - ltl2tgba has a new option, -S, to produce generalized-Büchi
      automata with state-based acceptance.  Those are obtained by
      converting some transition-based GBA into a state-based GBA, so
      they are usually not as small as one would wish.  The same
      option -S is also supported by autfilt.

    - ltlcross will work with translator producing automata with any
      acceptance condition, provided the output is in the HOA format.
      So it can effectively be used to validate tools producing Rabin
      or Streett automata.

    - ltlcross has several new options:

        --grind attempts to reduce the size of any bogus formula it
        discovers, while still exhibiting the bug.

        --ignore-execution-failures ignores cases where a translator
        exits with a non-zero status.

	--automata save the produced automata into the CSV or JSON
        file.  Those automata are saved using the HOA format.

      ltlcross will also output two extra columns in its CSV/JSON
      output: "ambiguous_aut" and "complete_aut" are Boolean
      that respectively tells whether the automaton is
      ambiguous and complete.

    - "ltlfilt --stutter-invariant" will now work with PSL formulas.
      The implementation is actually much more efficient
      than our previous implementation that was only for LTL.

    - ltlfilt's old -q/--quiet option has been renamed to
      --ignore-errors.  The new -q/--quiet semantic is the
      same as in grep (and also autfilt): disable all normal
      input, for situtations where only the exit status matters.

    - ltlfilt's old -n/--negate option can only be used as --negate
      now.  The short '-n NUM' option is now the same as the new
      --max-count=N option, for consistency with other tools.

    - ltlfilt has a new --count option to count the number of matching
      automata.

    - ltlfilt has a new --exclusive-ap option to constrain formulas
      based on a list of mutually exclusive atomic propositions.

    - ltlfilt has a new option --define to be used in conjunction with
      --relabel or --relabel-bool to print the mapping between old and
      new labels.

    - all tools that produce formulas or automata now have an --output
      (a.k.a. -o) option to redirect that output to a file instead of
      standard output.  The name of this file can be constructed using
      the same %-escape sequences that are available for --stats or
      --format.

    - all tools that output formulas have a -0 option to separate
      formulas with \0.  This helps in conjunction with xargs -0.

    - all tools that output automata have a --check option that
      request extra checks to be performed on the output to fill
      in properties values for the HOA format.  This options
      implies -H for HOA output.  For instance
      	ltl2tgba -H 'formula'
      will declare the output automaton as 'stutter-invariant'
      only if the formula is syntactically stutter-invariant
      (e.g., in LTL\X).  With
        ltl2tgba --check 'formula'
      additional checks will be performed, and the automaton
      will be accurately marked as either 'stutter-invariant'
      or 'stutter-sensitive'.  Another check performed by
      --check is testing whether the automaton is unambiguous.

    - ltlcross (and ltldo) have a list of hard-coded shorthands
      for some existing tools.  So for instance running
      'ltlcross spin ...' is the same as running
      'ltlcross "spin -f %s>%N" ...'.   This feature is much
      more useful for ltldo.

    - For options that take an output filename (i.e., ltlcross's
      --save-bogus, --grind, --csv, --json) you can force the file
      to be opened in append mode (instead of being truncated) by
      by prefixing the filename with ">>".  For instance
      	  --save-bogus=">>bugs.ltl"
      will append to the end of the file.

  * Other noteworthy news

    - The web site moved to http://spot.lrde.epita.fr/.

    - We now have Debian packages.
      See http://spot.lrde.epita.fr/install.html

    - The documentation now includes some simple code examples
      for both Python and C++.  (This is still a work in progress.)

    - The curstomized version of BuDDy (libbdd) used by Spot has be
      renamed as (libbddx) to avoid issues with copies of BuDDy
      already installed on the system.

    - There is a parser for the HOA format
      (http://adl.github.io/hoaf/) available as a
      spot::automaton_stream_parser object or spot::parse_aut()
      function.  The former version is able to parse a stream of
      automata in order to do batch processing.  This format can be
      output by all tools (since Spot 1.2.5) using the --hoa option,
      and it can be read by autfilt (by default) and ltlcross (using
      the %H specifier).  The current implementation does not support
      alternation.  Multiple initial states are converted into an
      extra initial state; complemented acceptance sets Inf(!x) are
      converted to Inf(x); explicit or implicit labels can be used;
      aliases are supported; "--ABORT--" can be used in a stream.

    - The above HOA parser can also parse never claims, and LBTT
      automata, so the never claim parser and the LBTT parser have
      been removed.  This implies that autfilt can input a mix of HOA,
      never claims, and LBTT automata.  ltlcross also use the same
      parser for all these output, and the old %T and %N specifiers
      have been deprecated and replaced by %O (for output).

    - While not all algorithms in the library are able to work with
      any acceptance condition supported by the HOA format, the
      following two new functions mitigate that:

      - remove_fin() takes a TωA whose accepting condition uses Fin(x)
	primitive, and produces an equivalent TωA without Fin(x):
	i.e., the output acceptance is a disjunction of generalized
	Büchi acceptance.  This type of acceptance is supported by
	SCC-based emptiness-check, for instance.

      - similarly, to_tgba() converts any TωA into an automaton with
	generalized-Büchi acceptance.

    - randomize() is a new algorithm that randomly reorders the states
      and transitions of an automaton.  It can be used from the
      command-line using "autfilt --randomize".

    - the interface in iface/dve2 has been renamed to iface/ltsmin
      because it can now interface the dynamic libraries created
      either by Divine (as patched by the LTSmin group) or by
      Spins (the LTSmin compiler for Promela).

    - LTL/PSL formulas can include /* comments */.

    - PSL SEREs support a new operator [:*i..j], the iterated fusion.
      [:*i..j] is to the fusion operator ':' what [*i..j] is to the
      concatenation operator ';'.  For instance (a*;b)[:*3] is the
      same as (a*;b):(a*;b):(a*;b).  The operator [:+], is syntactic
      sugar for [:*1..], and corresponds to the operator ⊕ introduced
      by Dax et al. (ATVA'09).

    - GraphViz output now uses an horizontal layout by default, and
      also use circular states (unless the automaton has more than 100
      states, or uses named-states).  The --dot option of the various
      command-line tools takes an optional parameter to fine-tune the
      GraphViz output (including vertical layout, forced circular or
      elliptic states, named automata, SCC information, ordered
      transitions, and different ways to colorize the acceptance
      sets).  The environment variables SPOT_DOTDEFAULT and
      SPOT_DOTEXTRA can also be used to respectively provide a default
      argument to --dot, and add extra attributes to the output graph.

    - Never claims can now be output in the style used by Spin since
      version 6.2.4 (i.e., using do..od instead of if..fi, and with
      atomic statements for terminal acceptance).  The default output
      is still the old one for compatibility with existing tools.  The
      new style can be requested from command-line tools using option
      --spin=6 (or -s6 for short).

    - Support for building unambiguous automata.  ltl_to_tgba() has a
      new options to produce unambiguous TGBA (used by ltl2tgba -U as
      discussed above).   The function is_unambiguous() will check
      whether an automaton is unambigous, and this is used by
      autfilt --is-unmabiguous.

    - The SAT-based minimization algorithm for deterministic automata
      has been updated to work with ω-Automaton with any acceptance.
      The input and the output acceptance can be different, so for
      instance it is possible to create a minimal deterministic
      Streett automaton starting from a deterministic Rabin automaton.
      This functionnality is available via autfilt's --sat-minimize
      option.  See doc/userdoc/satmin.html for details.

    - The on-line interface at http://spot.lrde.epita.fr/trans.html
      can be used to check stutter-invariance of any LTL/PSL formula.

    - The on-line interface will work around atomic propositions not
      supported by ltl3ba.  (E.g. you can now translate F(A) or
      G("foo < bar").)

  * Noteworthy code changes

    - Boost is not used anymore.

    - Automata are now manipulated exclusively via shared pointers.

    - Most of what was called tgba_something is now called
      twa_something, unless it is really meant to work only for TGBA.
      This includes functions, classes, file, and directory names.
      For instance the class tgba originally defined in tgba/tgba.hh,
      has been replaced by the class twa defined in twa/twa.hh.

    - the tgba_explicit class has been completely replaced by a more
      efficient twa_graph class.  Many of the algorithms that were
      written against the abstract tgba (now twa) interface have been
      rewritten using twa_graph instances as input and output, making
      the code a lot simpler.

    - The tgba_succ_iterator (now twa_succ_iterator) interface has
      changed.  Methods next(), and first() should now return a bool
      indicating whether the current iteration is valid.

    - The twa base class has a new method, release_iter(), that can
      be called to give a used iterator back to its automaton.  This
      iterator is then stored in a protected member, iter_cache_, and
      all implementations of succ_iter() can be updated to recycle
      iter_cache_ (if available) instead of allocating a new iterator.

    - The tgba (now called twa) base class has a new method, succ(),
      to support C++11' range-based for loop, and hide all the above
      change.

      Instead of the following syntax:

	tgba_succ_iterator* i = aut->succ_iter(s);
	for (i->first(); !i->done(); i->next())
	  {
	     // use i->current_state()
	     //     i->current_condition()
	     //     i->current_acceptance_conditions()
	  }
	delete i;

      We now prefer:

	for (auto i: aut->succ(s))
	  {
	    // use i->current_state()
	    //     i->current_condition()
	    //     i->current_acceptance_conditions()
	  }

      And the above syntax is really just syntactic suggar for

	twa_succ_iterator* i = aut->succ_iter(s);
	if (i->first())
	  do
	    {
	      // use i->current_state()
	      //     i->current_condition()
	      //     i->current_acceptance_conditions()
	    }
	  while (i->next());
	aut->release_iter(i); // allow the automaton to recycle the iterator

      Where the virtual calls to done() and delete have been avoided.

    - twa::succ_iter() now takes only one argument.  The optional
      global_state and global_automaton arguments have been removed.

    - The following methods have been removed from the TGBA interface and
      all their subclasses:
        - tgba::support_variables()
	- tgba::compute_support_variables()
	- tgba::all_acceptance_conditions()       // use acc().accepting(...)
	- tgba::neg_acceptance_conditions()
	- tgba::number_of_acceptance_conditions() // use acc().num_sets()

    - Membership to acceptance sets are now stored using bit sets,
      currently limited to 32 bits.  Each TωA has a acc() method that
      returns a reference to an acceptance object (of type
      spot::acc_cond), able to operate on acceptance marks
      (spot::acc_cond::mark_t).

      Instead of writing code like
        i->current_acceptance_conditions() == aut->all_acceptance_conditions()
      we now write
        aut->acc().accepting(i->current_acceptance_conditions())
      (Note that for accepting(x) to return something meaningful, x
      should be a set of acceptance sets visitied infinitely often.  So let's
      imagine that in the above example i is looking at a self-loop.)

      Similarly,
        aut->number_of_acceptance_conditions()
      is now
        aut->acc().num_sets()

    - All functions used for printing LTL/PSL formulas or automata
      have been renamed to print_something().  Likewise the various
      parsers should be called parse_something() (they haven't yet
      all been renamed).

    - All test suites under src/ have been merged into a single one in
      src/tests/.  The testing tool that was called
      src/tgbatest/ltl2tgba has been renamed as src/tests/ikwiad
      (short for "I Know What I Am Doing") so that users should be
      less tempted to use it instead of src/bin/ltl2tgba.

  * Removed features

    - The long unused interface to GreatSPN (or rather, interface to
      a non-public, customized version of GreatSPN) has been removed.
      As a consequence, we could get rid of many cruft in the
      implementation of Couvreur's FM'99 emptiness check.

    - Support for symbolic, BDD-encoded TGBAs has been removed.  This
      includes the tgba_bdd_concrete class and associated supporting
      classes, as well as the ltl_to_tgba_lacim() LTL translation
      algorithm.  Historically, this TGBA implementation and LTL
      translation were the first to be implemented in Spot (by
      mistake!) and this resulted in many bad design decisions.  In
      practice they were of no use as we only work with explicit
      automata (i.e. not symbolic) in Spot, and those produced by
      these techniques are simply too big.

    - All support for ELTL, i.e., LTL logic extended with operators
      represented by automata has been removed.  It was never used in
      practice because it had no practical user interface, and the
      translation was a purely-based BDD encoding producing huge
      automata (when viewed explictely), using the above and non
      longuer supported tgba_bdd_concrete class.

    - Our implementation of the Kupferman-Vardi complementation has
      been removed: it was unused in practice beause of the size of
      the automata built, and it did not survive the conversion of
      acceptance sets from BDDs to bitsets.

    - The unused implementation of state-based alternating Büchi
      automata has been removed.

    - Input and output in the old, Spot-specific, text format for
      TGBA, has been fully removed.  We now use HOA everywhere.  (In
      case you have a file in this format, install Spot 1.2.6 and use
      "src/tgbatest/ltl2tgba -H -X file" to convert the file to HOA.)

New in spot 1.2.6 (2014-12-06)

  * New features:

    - ltlcross --verbose is a new option to see what is being done

  * Bug fixes:

    - Remove one incorrect simplification rule for PSL discovered
      via checks on random formulaes.  (The bug was very unlikely
      to trigger on non-random formulas, because it requires a SERE
      with an entire subexpression that is unsatisfiable.)
    - When the automaton resulting from the translation of a positive
      formula is deterministic, ltlcross will compute its complement
      to perform additional checks against other translations of the
      positive formula.  The same procedure should be performed with
      automata obtained from negated formulas, but because of a typo
      this was not the case.
    - the neverclaim parser will now diagnose redefinitions of
      state labels.
    - the acceptance specification in the HOA format output have been
      adjusted to match recent changes in the format specifications.
    - atomic propositions are correctly escaped in the HOA output.
    - the build rules for documentation have been made compatible with
      version 8.0 of Org-mode.  (This was only a problem if you build
      from the git repository, or if you want to edit the
      documentation.)
    - recent to changes to libstd++ (as shipped by g++ 4.9.2) have
      demonstrated that the order of transitions output by the
      LTL->TGBA translation used to be dependent on the implementation
      of the STL.  This is now fixed.
    - some developpement version of libstd++ had a bug (PR 63698) in
      the assignment of std::set, and that was triggered in two places
      in Spot.  The workaround (not assigning sets) is actually more
      efficient, so we can consider it as a bug fix, even though
      libstd++ has also been fixed.
    - all parsers would report wrong line numbers while processing
      files with DOS style newlines.
    - add support for SWIG 3.0.

New in spot 1.2.5 (2014-08-21)

  * New features:

    - The online ltl2tgba translator will automatically attempt to
      parse a formula using LBT's syntax if it cannot parse it using
      the normal infix syntax.  It also has an option to display
      formulas using LBT's syntax.

    - ltl2tgba and dstar2tgba have a new experimental option --hoaf to
      output automata in the Hanoï Omega Automaton Format whose
      current draft is at http://adl.github.io/hoaf/
      The corresponding C++ function is spot::hoaf_reachable() in
      tgbaalgos/hoaf.hh.

    - 'randltl 4' is now a shorthand for 'randltl p0 p1 p2 p3'.

    - ltlcross has a new option --save-bogus=FILENAME to save any
      formula for which a problem (other than timeout) was detected
      during translation or using the resulting automatas.

  * Documentation:

    - The man page for ltl2tgba has some new notes and references
      about TGBA and about monitors.

  * Bug fixes:

    - Fix incorrect simplification of promises in the translation
      of the M operator (you may suffer from the bug even if you do
      not use this operator as some LTL patterns are automatically
      reduced to it).
    - Fix simplification of bounded repetition in SERE formulas.
    - Fix incorrect translation of PSL formulas of the form !{f} where
      f is unsatisifable.  A similar bug was fixed for {f} in Spot
      1.1.4, but for some reason it was not fixed for !{f}.
    - Fix parsing of neverclaims produced by Modella.
    - Fix a memory leak in the little-used conversion from
      transition-based alternating automata to tgba.
    - Fix a harmless uninitialized read in BuDDy.
    - When writing to the terminal, ltlcross used to display each
      formula in bright white, to make them stand out.  It turns out
      this was actually hiding the formulas for people using a
      terminal with white background... This version displays formula
      in bright blue instead.
    - 'randltl -n -1 --seed 0' and 'randltl -n -1 --seed 1' used to
      generate nearly the same list of formulas, shifted by one,
      because the PRNG write reset with an incremented seed between
      each output formula.  The PRNG is now reset only once.

New in spot 1.2.4 (2014-05-15)

  * New features:

    - "-B -x degen-lskip" can be used to disable level-skipping in the
      degeralization procedure called by ltl2tgba and dstar2tgba.
      This is mostly meant for running experiments.
    - "-B -x degen-lcache=N" can be used to experiment with different
      type of level caching during degeneralization.

  * Bug fixes:

    - Change the Python bindings to make them compatible with Swig 3.0.
    - "ltl2tgta --ta" could crash in certain conditions due to the
      introduction of a simulation-based reduction after
      degeneralization.
    - Fix four incorrect formula-simplification rules, three were
      related to the factorization of Boolean subformulas in
      operands of the non-length-matching "&" SERE operator, and
      a fourth one could only be enabled by explicitely passing the
      favor_event_univ option to the simplifier (not the default).
    - Fix incorrect translation of the fusion operator (":") in SERE
      such as {xx;1}:yy[*] where the left operand has 1 as tail.

New in spot 1.2.3 (2014-02-11)

  * New features:

    - The SPOT_SATLOG environment variable can be set to a filename to
      obtain statistics about the different iterations of the
      SAT-based minimization.  For an example, see
      http://spot.lrde.epita.fr/satmin.html
    - The bench/dtgbasat/ benchmark has been updated to use SPOT_SATLOG
      and record more statistics.
    - The default value for the SPOT_SATSOLVER environment
      variable has been changed to "glucose -verb=0 -model %I >%O".
      This assumes that glucose 3.0 is installed.   For older
      versions of glucose, remove the "-model" option.

  * Bug fixes:

    - More fixes for Python 3 compatibility.
    - Fix calculation of length_boolone(), were 'Xa|b|c' was
      considered as length 6 instead of 4 (because it is 'Xa|(b|a)'
      were (b|a) is Boolean).
    - Fix Clang-3.5 warnings.
    - randltl -S did not honor --boolean-priorities.
    - randltl had trouble generating formulas when all unary, or
      all binary/n-ary operators were disabled.
    - Fix spurious testsuite failure when using Pandas 0.13.
    - Add the time spent in child processes when measuring time
      with the timer class.
    - Fix determinism of the SAT-based minimization encoding.
      (It would sometimes produce different equivalent automata,
      because of a different encoding order.)
    - If the SAT-based minimization is asked for a 10-state automaton
      and returns a 6-state automaton, do not ask for a 9-state
      automaton in the next iteration...
    - Fix some compilation issue with the version of Apple's Clang
      that is installed with MacOS X 10.9.
    - Fix VPATH builds when building from the git repository.
    - Fix UP links in the html documentation for command-line tools.

New in spot 1.2.2 (2014-01-24)

  * Bug fixes:

    - Fix compilation *and* behavior of bitvectors on 32-bit
      architectures.
    - Fix some compilation errors observed using the antique G++ 4.0.1.
    - Fix compatibility with Python 3 in the test suite.
    - Fix a couple of new clang warnings (like "unused private member").
    - Add some missing #includes that are not included indirectly
      when the C++ compiler is in C++11 mode.
    - Fix detection of numbers that are too large in the ELTL parser.
    - Fix a memory leak in the ELTL parser, and avoid some unnecessary
      calls to strlen() at the same time.

New in spot 1.2.1 (2013-12-11)

  * New features:

    - commands for translators specified to ltlcross can now
      be given "short names" to be used in the CSV or JSON output.
      For instance
         ltlcross '{small} ltl2tgba -s --small %f >%N' ...
      will run the command "ltl2tgba -s --small %f >%N", but only
      print "small" in output files.

    - ltlcross' CSV and JSON output now contains two additional
      columns: exit_status and exit_code, used to report failures of
      the translator.  If the translation failed, only the time is
      reported, and the rest of the statistics, which are missing,
      area left empty (in CVS) or null (in JSON).  A new option,
      --omit-missing can be used to remove lines for failed
      translations, and remove these two columns.

    - if ltlcross is used with --products=+5 instead of --products=5
      then the stastics for each of the five products will be output
      separately instead of being averaged.

    - if ltlcross is used with tools that produce deterministic Streett
      or Rabin automata (as specified with %D), then the statistics
      output in CSV or JSON will have some extra columns to report
      the size of these input automata before ltlcross converts them
      into TGBA to perform its regular checks.

    - ltlfilt, ltl2tgba, ltl2tgta, and ltlcross can now read formulas
      from CSV files.  Use option -F FILE/COL to read formulas from
      column COL of FILE.  Use -F FILE/-COL if the first line of
      FILE be ignored.

    - when ltlfilt processes formulas from a CSV file, it will output
      each CSV line whose formula matches the given constraints, with
      the rewriten formula.  The new escape sequence %< (text in
      columns before the formula) and %> (text after) can be used
      with the --format option to alter this output.

    - ltlfilt, genltl, randltl, and ltl2tgba have a --csv-escape option
      to help escape formulas in CSV files.

    - Please check
            http://spot.lrde.epita.fr/csv.html
      for some discussion and examples of the last few features.

  * Bug fixes:

    - ltlcross' CSV output has been changed to be more RFC 4180
      compliant: it no longuer output useless cosmetic spaces, and
      use double-quotes with proper escaping for strings.  The only
      RFC 4180 rule that it does not follow is that it will terminate
      lines with \n instead of \r\n because the latter cause issues
      with a couple of tools.

    - ltlcross failed to report missing input or output escape sequences
      on all but the first configured translator.

New in spot 1.2 (2013-10-01)

  * Changes to command-line tools:

    - ltlcross has a new option --color to color its output.  It is
      enabled by default when the output is a terminal.

    - ltlcross will give an example of infinite word accepted by the
      two automata when the product between a positive automaton and a
      negative automaton is non-empty.

    - ltlcross can now read the Rabin and Streett automata output by
      ltl2dstar.  This type of output should be specified using '%D':

        ltlcross 'ltl2dstar --ltl2nba=spin:path/to/ltl2tgba@-s %L %D'

      However because Spot only supports Büchi acceptance, these Rabin
      and Streett automata are immediately converted to TGBAs before
      further processing by ltlcross.  This is still interesting to
      search for bugs in translators to Rabin or Streett automata, but
      the statistics (of the resulting TGBAs) might not be very relevant.

    - When ltlcross obtains a deterministic automaton from a
      translator it will now complement this automaton to perform
      additional intersection checks.  This is complementation is done
      only for deterministic automata (because that is cheap) and can
      be disabled with --no-complement.

    - To help with debugging problems detected by ltlcross, the
      environment variables SPOT_TMPDIR and SPOT_TMPKEEP control where
      temporary files are created and if they should be erased.  Read
      the man page of ltlcross for details.

    - There is a new command, named dstar2tgba, that converts a
      deterministic Rabin or Streett automaton (expressed in the
      output format of ltl2dstar) into a TGBA, BA or Monitor.

      In the case of Rabin acceptance, the conversion will output a
      deterministic Büchi automaton if one such automaton exist.  Even
      if no such automaton exists, the conversion will actually
      preserves the determinism of any SCC that can be kept
      deterministic.

      In the case of Streett acceptance, the conversion produces
      non-deterministic Büchi automata with Generalized acceptance.
      These are then degeneralized if requested.

      See http://spot.lrde.epita.fr/dstar2tgba.html for some
      examples, and the man page for more reference.

    - The %S escape sequence used by ltl2tgba --stats to display the
      number of SCCs in the output automaton has been renamed to %c.
      This makes it more homogeneous with the --stats option of the
      new dstar2tgba command.

      Additionally, the %p escape can now be used to show whether the
      output automaton is complete, and the %r escape will give the
      number of seconds spent building the output automaton (excluding
      the time spent parsing the input).

    - ltl2tgba, ltl2tgta, and dstar2tgba have a --complete option
      to output complete automata.

    - ltl2tgba, ltl2tgta, and dstar2tgba can use a SAT-solver to
      minimize deterministic automata.  Doing so is only needed on
      properties that are stronger than obligations (for obligations
      our WDBA-minimization procedure will return a minimimal
      deterministic automaton more efficiently) and is disabled by
      default.  See the spot-x(7) man page for documentation about the
      related options: sat-minimize, sat-states, sat-acc, state-based.
      See also http://spot.lrde.epita.fr/satmin.html for some
      examples.

    - ltlfilt, genltl, and randltl now have a --latex option to output
      formulas in a way that its easier to embed in a LaTeX document.
      Each operator is output as a command such as \U, \F, etc.
      doc/tl/spotltl.sty gives one possible definition for each macro.

    - ltlfilt, genltl, and randltl have a new --format option to
      indicate how to present the output formula, possibly with
      information about the input.

    - ltlfilt as a new option, --relabel-bool, to abstract independent
      Boolean subformulae as if they were atomic propositions.
      For instance "a & GF(c | d) & b & X(c | d)" would be rewritten
      as "p0 & GF(p1) & Xp1".

  * New functions and classes in the library:

    - dtba_sat_synthetize(): Use a SAT-solver to build an equivalent
      deterministic TBA with a fixed number of states.

    - dtba_sat_minimize(), dtba_sat_minimize_dichotomy(): Iterate
      dtba_sat_synthetize() to reduce the number of states of a TBA.

    - dtgba_sat_synthetize(), dtgba_sat_minimize(),
      dtgba_sat_minimize_dichotomy(): Likewise, for deterministic TGBA.

    - is_complete(): Check whether a TGBA is complete.

    - tgba_complete(): Complete an automaton by adding a sink state
      if needed.

    - dtgba_complement(): Complement a deterministic TGBA.

    - satsolver(): Run an (external) SAT solver, honoring the
      SPOT_SATSOLVER environment variable if set.

    - tba_determinize(): Run a power-set construction, and attempt
      to fix the acceptance simulation to build a deterministic TBA.

    - dstar_parse(): Read a Streett or Rabin automaton in
      ltl2dstar's format.  Note that this format allows only
      deterministic automata.

    - nra_to_nba(): Convert a (possibly non-deterministic) Rabin
      automaton to a non-deterministic Büchi automaton.

    - dra_to_ba(): Convert a deterministic Rabin automaton to a Büchi
      automaton, preserving acceptance in all SCCs where this is possible.

    - nsa_to_tgba(): Convert a (possibly non-deterministic) Streett
      automaton to a non-deterministic TGBA.

    - dstar_to_tgba(): Convert any automaton returned by dstar_parse()
      into a TGBA.

    - build_tgba_mask_keep(): Build a masked TGBA that shows only
      a subset of states of another TGBA.

    - build_tgba_mask_ignore(): Build a masked TGBA that ignore
      a subset of states of another TGBA.

    - class tgba_proxy: Helps writing on-the-fly algorithms that
      delegate most of their methods to the original automaton.

    - class bitvect: A dynamic bit vector implementation.

    - class word: An infinite word, stored as prefix + cycle, with a
      simplify() methods to simplify cycle and prefix in obvious ways.

    - class temporary_file: A temporary file.  Can be instanciated with
      create_tmp_file() or create_open_tmpfile().

    - count_state(): Return the number of states of a TGBA.  Implement
      a couple of specializations for classes where is can be know
      without exploration.

    - to_latex_string(): Output a formula using LaTeX syntax.

    - relabel_bse(): Relabeling of Boolean Sub-Expressions.
      Implements ltlfilt's --relabel-bool option describe above.

  * Noteworthy internal changes:

    - When minimize_obligation() is not given the formula associated
      to the input automaton, but that automaton is deterministic, it
      can still attempt to call minimize_wdba() and check the correcteness
      using dtgba_complement().  This allows dstar2tgba to apply
      WDBA-minimization on deterministic Rabin automata.

    - tgba_reachable_iterator_depth_first has been redesigned to
      effectively perform a DFS.  As a consequence, it does not
      inherit from tgba_reachable_iterator anymore.

    - postproc::set_pref() was used to accept an argument among Any,
      Small or Deterministic.  These can now be combined with Complete
      as Any|Complete, Small|Complete, or Deterministic|Complete.

    - operands of n-ary operators (like & and |) are now ordered so
      that Boolean terms come first.  This speeds up syntactic
      implication checks slightly.  Also, literals are now sorted
      using strverscmp(), so that p5 comes before p12.

    - Syntactic implication checks have been generalized slightly
      (for instance 'a & b & F(a & b)' is now reduced to 'a & b'
      while it was not changed in previous versions).

    - All the parsers implemented in Spot now use the same type to
      store locations.

    - Cleanup of exported symbols

      All symbols in the library now have hidden visibility on ELF systems.
      Public classes and functions have been marked explicitely for export
      with the SPOT_API macro.

      During this massive update, some of functions that should not have
      been made public in the first place have been moved away so that
      they can only be used from the library.  Some old of unused
      functions have been removed.

      removed:
	- class loopless_modular_mixed_radix_gray_code

      hidden:
	- class acc_compl
	- class acceptance_convertor
	- class bdd_allocator
	- class free_list

  * Bug fixes:

    - Degeneralization was not indempotant on automata with an
      accepting initial state that was on a cycle, but without
      self-loop.

    - Configuring with --enable-optimization would reset the value of
      CXXFLAGS.


New in spot 1.1.4 (2013-07-29)

  * Bug fixes:
    - The parser for neverclaim, updated in 1.1.3, would fail to
      parse guards of the form (a) || (b) output by ltl2ba or
      ltl3ba, and would only understand ((a) || (b)).
    - When used from ltlcross, the same parser would fail to
      parse further neverclaims after the first failure.
    - Add a missing newline in some error message of ltlcross.
    - Expressions like {SERE} were wrongly translated and simplified
      for SEREs that accept the empty word: they were wrongly reduced
      to true.  Simplification and translation rules have been fixed,
      and the doc/tl/tl.pdf specifications have been updated to better
      explain that {SERE} has the semantics of a closure operator that
      is not exactly what one could expect after reading the PSL
      standard.
    - Various typos.

New in spot 1.1.3 (2013-07-09)

  * New feature:
    - The neverclaim parser now understands the new style of output
      used by Spin 6.24 and later.
  * Bug fixes:
    - The scc_filter() function could abort with a BDD error.  If all
      the acceptance sets of an SCC but the first one were useless.
    - The script in bench/spin13/ would not work on MacOS X because
      of some non-portable command.
    - A memory corruption in ltlcross.

New in spot 1.1.2 (2013-06-09)

  * Bug fixes:
    - Uninitialized variables in ltlcross (affect the count of terminal
      weak, and strong SCCs).
    - Workaround an old GCC bug to allow compilation with g++ <= 4.5
    - Fix several Doxygen comments so that they display correctly.

New in spot 1.1.1 (2013-05-13):

  * New features:

    - lbtt_reachable(), the function that outputs a TGBA in LBTT's
      format, has a new option to indicate that the TGBA being printed
      is in fact a Büchi automaton.  In this case it outputs an LBTT
      automaton with state-based acceptance.

      The output of the guards has also been changed in two ways:
      1. atomic propositions that do not match p[0-9]+ are always
         double-quoted.  This avoids issues where t or f were used as
         atomic propositions in the formula, output as-is in the
         automaton, and read back as true or false.  Other names that
         correspond to LBT operators would cause problem as well.
      2. formulas that label transitions are now output as
         irredundant-sums-of-products.

    - 'ltl2tgba --ba --lbtt' will now output automata with state-based
      acceptance.  You can use 'ltl2tgba --ba --lbtt=t' to force the
      output of transition-based acceptance like in the previous
      versions.

      Some illustrations of this point and the previous one can be
      found in the man page for ltl2tgba(1).

    - There is a new function scc_filter_states() that removes all
      useless states from a TGBA.  It is actually an abbridged version
      of scc_filter() that does not alter the acceptance conditions of
      the automaton.  scc_filter_state() should be used when
      post-processing TGBAs that actually represent BAs.

    - simulation_sba(), cosimulation_sba(), and
      iterated_simulations_sba() are new functions that apply to TGBAs
      that actually represent BAs.  They preserve the imporant
      property that if a state of the BA is is accepting, the outgoing
      transitions of that state are all accepting in the TGBA that
      represent the BA.  This is something that was not preserved by
      functions cosimultion() and iterated_simulations() as mentionned
      in the bug fixes below.

    - ltlcross has a new option --seed, that makes it possible to
      change the seed used by the random graph generator.

    - ltlcross has a new option --products=N to check the result of
      each translation against N different state spaces, and everage
      the statistics of these N products.  N default to 1; larger
      values increase the chances to detect inconsistencies in the
      translations, and also make the average size of the product
      built against the translated automata a more pertinent
      statistic.

    - bdd_dict::unregister_all_typed_variables() is a new function,
      making it easy to unregister all BDD variables of a given type
      owned by some object.

  * Bug fixes:

    - genltl --gh-r generated the wrong formulas due to a typo.
    - ltlfilt --eventual and --universal were not handled properly.
    - ltlfilt --stutter-invariant would trigger an assert on PSL formulas.
    - ltl2tgba, ltl2tgta, ltlcross, and ltlfilt, would all choke on empty
      lines in a file of formulas.  They now ignore empty lines.
    - The iterated simulation applied on degeneralized TGBA was bogus
      for two reasons: one was that cosimulation was applied using the
      generic cosimulation for TGBA, and the second is that
      SCC-filtering, performed between iterations, was also a
      TGBA-based algorithm.  Both of these algorithms could lose the
      property that if a TGBA represents a BA, all the outgoing
      transitions of a state should be accepting.  As a consequence, some
      formulas where translated to incorrect Büchi automata.

New in spot 1.1 (2013-04-28):

  Several of the new features described below are discribed in

    Tomáš Babiak, Thomas Badie, Alexandre Duret-Lutz, Mojmír
    Křetínský, Jan Strejček: Compositional Approach to Suspension and
    Other Improvements to LTL Translation.  To appear in the
    proceedings of SPIN'13.

  * New features in the library:

    - The postprocessor class now takes an optional option_map
      argument that can be used to specify fine-tuning options, making
      it easier to benchmark different scenarios while developing new
      postprocessings.

    - A new translator class implements a complete translation chain,
      from LTL/PSL to TGBA/BA/Monitor.  It performs pre- and
      post-processings in addition to the core translation, and offers
      an interface similar to that used in the postprocessor class, to
      specify the intent of the translation.

    - The degeneralization algorithm has learned three new tricks:
      level reset, level caching, and SCC-based ordering.  The former
      two are enabled by default.  Benchmarking has shown that the
      latter one does not always have a positive effect, so it is
      disabled by default.  (See SPIN'13 paper.)

    - The scc_filter() function, which removes dead SCCs and also
      simplify acceptance conditions, has learnt how to simplify
      acceptance conditions in a few tricky situations that were not
      simplified previously.  (See SPIN'13 paper.)

    - A new translation, called compsusp(), for "Compositional
      Suspension" is implemented on top of ltl_to_tgba_fm().
      (See SPIN'13 paper.)

    - Some experimental LTL rewriting rules that trie to gather
      suspendable formulas are implemented and can be activated
      with the favor_event_univ option of ltl_simplifier.  As
      always please check doc/tl/tl.tex for the list of rules.

    - An experimental "don't care" (direct) simulation has been
      implemented.  This simulations consider the acceptance
      of out-of-SCC transitions as "don't care".  It is not
      enabled by default because it currently is very slow.

    - remove_x() is a function that take a formula, and rewrite it
      without the X operator.  The rewriting is only correct for
      stutter-insensitive LTL formulas (See K. Etessami's paper in IFP
      vol. 75(6). 2000) This algorithm is accessible from the
      command-line using ltlfilt's --remove-x option.

    - is_stutter_insensitive() takes any LTL formula, and check
      whether it is stutter-insensitive.  This algorithm is accessible
      from the command-line using ltlfilt's --stutter-insensitive
      option.

    - Several functions have been introduced to check the
      strength of an SCC.
        is_inherently_weak_scc()
	is_weak_scc()
	is_syntactic_weak_scc()
	is_complete_scc()
	is_terminal_scc()
	is_syntactic_terminal_scc()

      Beware that the costly is_weak_scc() function introduced in Spot
      1.0, which is based on a cycle enumeration, has been renammed to
      is_inherently_weak_scc() to match established vocabulary.

  * Command-line tools:

    - ltl2tgba and ltl2tgta now honor a new --extra-options (or -x)
      flag to fine-tune the algorithms used.  The available options
      are documented in the spot-x (7) manpage.  For instance use '-x
      comp-susp' to use the afore-mentioned compositional suspension.

    - The output format of 'ltlcross --json' has been changed slightly.
      In a future version we will offer some reporting script that turn
      such JSON output into various tables and graphs, and these change
      are required to make the format usable for other benchmarks (not
      just ltlcross).

    - ltlcross will now count the number of non-accepting, terminal,
      weak, and strong SCCs, as well as the number of terminal, weak,
      and strong automata produced by each tool.

  * Documentation:

    - org-mode files used to generate the documentation about
      command-line tools (shown at http://spot.lrde.epita.fr/tools.html)
      is distributed in doc/org/.  The resulting html files are also
      in doc/userdoc/.

  * Web interface:

    - A new "Compositional Suspension" tab has been added to experiment
      with compositional suspension.

  * Benchmarks:

    - See bench/spin13/README for instructions to reproduce our Spin'13
      benchmark for the compositional suspension.

  * Bug fixes:

    - There was a memory leak in the LTL simplification code, that could
      only be triggered when disabling advanced simplifications.

    - The translation of the PSL formula !{xxx} was incorrect when xxx
      simplified to false.

    - Various warnings triggered by new compilers.

New in spot 1.0.2 (2013-03-06):

  * New features:
    - the on-line ltl2tgba.html interface can output deterministic or
      non-deterministic monitors.  However, and unlike the ltl2tgba
      command-line tool, it doesn't different output formats.
    - the class ltl::ltl_simplifier now has an option to rewrite Boolean
      subformulaes as irredundante-sum-of-product during the simplification
      of any LTL/PSL formula.  The service is also available as a method
      ltl_simplifier::boolean_to_isop() that applies this rewriting
      to a Boolean formula and implements a cache.
      ltlfilt as a new option --boolean-to-isop to try to apply the
      above rewriting from the command-line:
      	    % ltlfilt --boolean-to-isop -f 'GF((a->b)&(b->c))'
	    GF((!a & !b) | (b & c))
      This is currently not used anywhere else in the library.
  * Bug fixes:
    - 'ltl2tgba --high' is documented to be the same as 'ltl2tgba',
      but by default ltl2tgba forgot to enable LTL simplifications based
      on language containment, which --high do enable.  There are now
      enabled by default.
    - the on-line ltl2tgba.html interface failed to output monitors,
      testing automata, and generalized testing automata due to two
      issues with the Python bindings.  It also used to display
      Testing Automaton Options when the desired output was set to Monitor.
    - bench/ltl2tgba would not work in a VPATH build.
    - a typo caused some .dir-locals.el configuration parameters to be
      silently ignored by emacs
    - improved Doxygen comments for formula_to_bdd, bdd_to_formula,
      and bdd_dict.
    - src/tgbatest/ltl2tgba (not to be confused with src/bin/ltl2tgba)
      would have a memory leak when passed the conflicting option -M
      and -O.  It probably has many other problems.  Do not use
      src/tgbatest/ltl2tgba if you are not writing a test case for
      Spot.  Use src/bin/ltl2tgba instead.

New in spot 1.0.1 (2013-01-23):

  * Bug fixes:
    - Two executions of the simulation reductions could produce
      two isomorphic automata, but with transitions in a different
      order.
    - ltlcross did not diagnose write errors to temporary files,
      and certain versions of g++ would warn about it.
    - "P0.init" is parsed as an atomic even without the double quotes,
      but it was always output with double quotes.  This version will
      not quote this atomic proposition anymore.
    - "U", "W", "M", "R" were correctly parsed as atomic propositions
      (instead of binary operators) when placed in double quotes, but
      on output they were output without quotes, making the result
      unparsable.
    - the to_lbt_string() functions would always output a trailing space.
      This is not the case anymore.
    - tgba_product::transition_annotation() would segfault when
      called in a product against a Kripke structure.
  * Minor improvements:
    - Four new LTL simplifications rules:
        GF(a|Xb) = GF(a|b)
        GF(a|Fb) = GF(a|b)
        FG(a&Xb) = FG(a&b)
        FG(a&Gb) = FG(a&b)
    - The on-line version of ltl2tgba now displays edge and
      transition counts, just as the ltlcross tool.
    - ltlcross will display the number of timeouts at the end
      of its execution.
    - ltlcross will diagnose tools with missing input or
      output %-sequence before attempting to run any of them.
    - The parser for LBT's prefix-style LTL formulas will now
      read atomic propositions that are not of the form p1, p2...
      This makes it possible to process formulas written in
      ltl2dstar's syntax.
  * Pruning:
    - lbtt has been removed from the distribution.  A copy of the last
      version we distributed is still available at
        http://spot.lip6.fr/dl/lbtt-1.2.1a.tar.gz
      and our test suite will use it if it is installed, but the same
      tests are already performed by ltlcross.
    - the bench/ltl2tgba/ benchmark, that used lbtt to compare various
      LTL-to-Büchi translators, has been updated to use ltlcross.  It
      now output summary tables in LaTeX.  Support for Modella (no
      longer available online), and Wring (requires a too old Perl
      version) have been dropped.
    - the half-baked and underdocumented "Event TGBA" support in
      src/evtgba*/ has been removed, as it was last worked on in 2004.

New in spot 1.0 (2012-10-27):

  * License change: Spot is now distributed using GPL v3+ instead
    of GPL v2+.  This is because we started using some third-party
    files distributed under GPL v3+.

  * Command-line tools

    Useful command-line tools are now installed in addition to the
    library.  Some of these tools were originally written for our test
    suite and had evolved organically into useful programs with crappy
    interfaces: they have now been rewritten with better argument
    parsing, saner defaults, and they come with man pages.

    - genltl:   Generate LTL formulas from scalable patterns.
                This offers 20 patterns so far.

    - randltl:  Generate random LTL/PSL formulas.

    - ltlfilt:  Filter lists of formulas according to several criteria
                (e.g., match only safety formulas that are larger than
		some given size).  Besides being used as a "grep" tool
		for formulas, this can also be used to convert
		files of formulas between different syntaxes, apply
		some simplifications, check whether to formulas are
		equivalent, ...

    - ltl2tgba: Translate LTL/PSL formulas into Büchi automata (TGBA,
                BA, or Monitor).  A fundamental change to the
                interface is that you may now specify the goal of the
                translation: do you you favor deterministic or smaller
                automata?

    - ltl2tgta: Translate LTL/PSL formulas into Testing Automata.

    - ltlcross: Compare the output of translators from LTL/PSL to
                Büchi automata, to find bug or for benchmarking.  This
                is essentially a Spot-based reimplementation of LBTT
                that supports PSL in addition to LTL, and that can
                output more statistics.

    An introduction to these tools can be found on-line at
      http://spot.lrde.epita.fr/tools.html

    The former test versions of genltl and randltl have been removed
    from the source tree.  The old version of ltl2tgba with its
    gazillion options is still in src/tgbatest/ and is meant to be
    used for testing only.  Although ltlcross is meant to replace
    LBTT, we are still using both tools in this release; however this
    is likely to be the last release of Spot that redistributes LBTT.

  * New features in the Spot library:

    - Support for various flavors of Testing Automata.

      The flavors are:
      + "classical" Testing Automata, as used for instance by
	Geldenhuys and Hansen (Spin'06), using Büchi and
	livelock acceptance conditions.
      + Generalized Testing Automata, extending the previous
	with multiple Büchi acceptance sets.
      + Transition-based Generalized Testing Automata moving Büchi
        acceptance to transitions, and getting rid of livelock
        acceptance conditions by expliciting stuttering self-loops.

      Supporting algorithms include anything required to run
      the automata-theoretic approach using testing automata:

      + dedicated synchronized product
      + dedicated emptiness-check for TA and GTA, as these
        may require two passes because of the two kinds of
	acceptance, while a TGTA can be checked for emptiness
	with the same one-pass algorithm as a TGBA.
      + conversion from a TGBA to any of the above kind, with
        options to reduce these automata with bisimulation,
	and to produce a BA/GBA that require a single pass
	(at the expense of determinism).
      + output in dot format for display

      A discussion of these automata, part of Ala Eddine BEN SALEM's
      PhD work, should appear in ToPNoC VI (LNCS 7400).  The web-based
      interface and the aforementioned ltl2tgta tool can be used
      to build testing automata.

    - TGBA can now be reduced by Reverse Simulation (in addition to
      the Direct Simulation introduced in 0.9).  A function called
      iterated_simulations() will alternate direct and reverse
      simulations in a loop as long as it diminishes the size of the
      automaton.

    - The enumerate_cycles class implements the Loizou-Thanisch
      algorithm to enumerate elementary cycles in a SCC.  As an
      example of use, is_weak_scc() will tell whether an SCC is
      inherently weak (all its cycles are accepting, or none of them
      are).

    - parse_lbt() will parse an LTL formula expressed in the prefix
      syntax used (at least) by LBT, LBTT and Scheck.
      to_lbt_string() can be used to print an LTL formula using this
      syntax.

    - to_wring_string() can be used to print an LTL formula into
      Wring's syntax.

    - The LTL/PSL parser now has a lenient mode that can be useful
      to interpret atomic proposition with language-specific constructs.
      In lenient mode, any (...) or {...} block that cannot be parsed
      as formula will be assumed to be an atomic proposition.
      For instance the input (a < b) U (process[2]@ok), normally
      flagged as a syntax error, is read as "a < b" U "process[2]@ok"
      in lenient mode.

    - minimize_obligation() has a new option to disable WDBA
      minimization it cases it would produce a deterministic automaton
      that is bigger than the original TGBA.  This can help
      choosing between less states or more determinism.

    - new functions is_deterministic() and count_nondet_states()
      (The count of nondeterministic states is now displayed on
      automata generated with the web interface.)

    - A new class, "postprocessor", makes it easier to apply
      all available simplification algorithms on a TGBA/BA/Monitors.

  * Minor changes:

    - The '*' operator can (again) be used as an AND in LTL formulas.
      This is for compatibility with formula written in Wring's
      syntax.  However inside SERE it is interpreted as the Kleen
      star.

    - When printing a formula using Spin's LTL syntax, we don't
      double-quote complex atomic propositions (that was not valid
      Spin input anyway).   For instance F"foo == 2" used to be
      output as <>"foo == 2".  We now output <>(foo == 2) instead.
      The latter syntax is understood by Spin 6.  It can be read
      back by Spot in lenient mode (see above).

    - The gspn-ssp benchmark has been removed.


New in spot 0.9.2 (2012-07-02):

  * New features to the web interface.
    - It can run ltl3ba (Babiak et al., TACAS'12) where available.
    - "a loading logo" is displayed when result is not instantaneous.
  * Speed improvements:
    - The unicity hash table of BuDDy has been separated separated
      node table for better cache-friendliness.  The resulting speedup
      is around 5% on BDD-intensive algorithms.
    - A new BDD operation, called bdd_implies() has been added to
      BuDDy to check whether one BDD implies another.  This benefits
      mostly the simulation and degeneralization algorithms of Spot.
    - A new offline implementation of the degeneralization (which
      had always been performed on-the-fly so far) available.  This
      especially helps the Safra complementation.
  * Bug fixes:
    - The CGI script running for ltl2tgba.html will correctly timeout
      after 30s when Spot's translation takes more time.
    - Applying WDBA-minimization on an automaton generated by the
      Couvreur/LaCIM translator could lead to an incorrect automaton
      due to a bug in the definition of product with symbolic
      automata.
    - The Makefile.am of BuDDy, LBTT, and Spot have been adjusted to
      accomodate Automake 1.12 (while still working with 1.11).
    - Better error recovery when parsing broken LTL formulae.
    - Fix errors and warnings reported by clang 3.1 and the
      upcoming g++ 4.8.

New in spot 0.9.1 (2012-05-23):

  * The version of LBTT we distribute includes a patch from Tomáš
    Babiak to count the number of non-deterministic states, and the
    number of deterministic automata produced.
    See lbtt/NEWS for the list of other differences with the original
    version of LBTT 1.2.1.
  * The Couvreur/FM translator has learned two new tricks.  These only
    help to speedup the translation by not issuing states or
    acceptance conditions that would be latter suppresed by other
    optimizations.
    - The translation rules used to translate subformulae of the G
      operator have been adjusted not to produce useless loops
      already implied by G.  This generalizes the "GF" trick
      presented in Couvreur's original FM'99 paper.
    - Promises generated for formula of the form P(a U (b U c))
      are reduced into P(c), avoiding the introduction of many
      promises that imply each other.
  * The tgba_parse() function is now available via the Python
    bindings.
  * Bug fixes:
    - The random SERE generator was using the wrong operators
      for "and" and "or", mistaking And/Or with AndRat/OrRat.
    - The translation of !{r} was incorrect when this subformula
      was recurring (e.g. in G!{r}) and r had loops.
    - Correctly recognize ltl2tgba's option -rL.
    - Using LTL simplification rules based on syntactic implication,
      or based on language containment checks, caused BDD variables
      to be allocated in an "unnatural" order, resulting in a slower
      translation and a less optimal degeneralization.
    - When ltl2tgba reads a neverclaim, it now considers the resulting
      TGBA as a Büchi automaton, and will display double circles in
      the dotty output.

New in spot 0.9 (2012-05-09):

  * New features:
    - Operators from the linear fragment of PSL are supported.  This
      basically extends LTL with Sequential Extended Regulat
      Expressions (SERE), and a couple of operators to bridge SERE and
      LTL.  See doc/tl/tl.pdf for the list of operators and their
      semantics.
    - Formula rewritings have been completely revamped, and augmented
      with rules for PSL operators (and some new LTL rules as well).
      See doc/tl/tl.pdf for the list of the rewritings implemented.
    - Some of these rewritings that may produce larger formulas
      (for instance to rewrite "{a;b;c}" into "a & X(b & Xc)")
      may be explicitely disabled with a new option.
    - The src/ltltest/randltl tool can now generate random SEREs
      and random PSL formulae.
    - Only one translator (ltl2tgba_fm) has been augmented to
      translate the new SERE and PSL operators.  The internal
      translation from SERE to DFA is likely to be rewriten in a
      future version.
    - A new function, length_boolone(), computes the size of an
      LTL/PSL formula while considering that any Boolean term has
      length 1.
    - The LTL/PSL parser recognizes some UTF-8 characters (like ◇ or
      ∧) as operators, and some output routines now have an UTF-8
      output mode.  Tools like randltl and ltl2tgba have gained an -8
      option to enable such output.  See doc/tl/tl.pdf for the list
      of recognized codepoints.
    - A new direct simulation reduction has been implemented.  It
      works directly on TGBAs.  It is in src/tgbaalgos/simlation.hh,
      and it can be tested via ltl2tgba's -RDS option.
    - unabbreviate_wm() is a function that rewrites the W and M operators
      of LTL formulae using R and U.  This is called whenever we output
      a formula in Spin syntax.  By combining this with the aforementioned
      PSL rewriting rules, many PSL formulae that use simple SERE can be
      converted into LTL formulae that can be feed to tools that only
      understand U and R.  The web interface will let you do this.
    - changes to the on-line translator:
      + SVG output is available
      + can display some properties of a formula
      + new options for direct simulation, larger rewritings, and
        utf-8 output
    - configure --without-included-lbtt will prevent LBTT from being
      configured and built.  This helps on systems (such as MinGW)
      where LBTT cannot be built.  The test-suite will skip any
      LBTT-based test if LBTT is missing.
  * Interface changes:
    - Operators ->, <->, U, W, R, and M are now parsed as
      right-associative to better match the PSL standard.
    - The constructors for temporal formulae will perform some trivial
      simplifications based on associativity, commutativity,
      idempotence, and neutral elements.  See doc/tl/tl.pdf for the
      list of such simplifications.
    - Formula instances now have many methods to inspect their
      properties (membership to syntactic classes, absence of X
      operator, etc...) in constant time.
    - LTL/PSL formulae are now handled everywhere as 'const formula*'
      and not just 'formula*'.  This reflects the true nature of these
      (immutable) formula objects, and cleanups a lot of code.
      Unfortunately, it is a backward incompatible change: you may have
      to add 'const' to a couple of lines in your code, and change
      'ltl::const_vistitor' into 'ltl::visitor' if you have written a
      custom visitor.
    - The new entry point for LTL/PSL simplifications is the function
      ltl_simplifier::simplify() declared in src/ltlvisit/simplify.hh.
      The ltl_simplifier class implements a cache.
      Functions such as reduce() or reduce_tau03() are deprecated.
    - The old game-theory-based implementations for direct and delayed
      simulation reductions have been removed.  The old direct
      simulation would only work on degeneralized automata, and yet
      produce results inferior to the new direct simulation introduced
      in this release.  The implementation of delayed simulation was
      unreliable.  The function reduc_tgba_sim() has been kept
      for compatibility (it calls the new direct simulation whatever
      the type of simulation requested) and marked as deprecated.
      ltl2tgba's options -Rd, -RD are gone.  Options -R1t, -R1s,
      -R2s, and -R2t are deprecated and all made equivalent to -RDS.
    - The tgba_explicit hierarchy has been reorganized in order to
      make room for sba_explicit classes that share most of the code.
      The main consequence is that the tgba_explicit type no longuer
      exists.  However the tgba_explicit_number,
      tgba_explicit_formula, and tgba_explicit_string still do.

New in spot 0.8.3 (2012-03-09):

  * Support for both Python 2.x and Python 3.x.
    (Previous versions would only work with Python 2.x.)
  * The online ltl2tgba.html now stores its state in the URL so that
    history is preserved, and links to particular setups can be sent.
  * Bug fixes:
    - Fix a segfault in the compression code used by the -Z
      option of dve2check.
    - Fix a race condition in the CGI script.
    - Fix a segfault in the CGI script when computing a Büchi run.

New in spot 0.8.2 (2012-01-19):

  * configure now has a --disable-python option to disable
    the compilation of Python bindings.
  * Minor speedups in the Safra complementation.
  * Better memory management for the on-the-fly degeneralization
    algorithm.  This mostly benefits to the Safra complementation.
  * Bug fixes:
    - spot::ltl::length() forgot to count the '&' and '|' operators
      in an LTL formula.
    - minimize_wdba() could fail to mark some transiant SCCs as accepting,
      producing an automaton that was not fully minimized.
    - minimize_dfa() could produce incorrect automata, but it is not
      clear whether this could have had an inpact on WDBA minimization
      (the worse case is that some TGBA would not have been minimized
      when they could).
    - Fix a Python syntax error in the CGI script.
    - Fix compilation with g++ 4.0.
    - Fix a make check failure when valgrind is missing.

New in spot 0.8.1 (2011-12-18):

  * Only bug fixes:
    - When ltl2tgba is set to perform both WDBA minimization and
      degeneralization, do the latter only if the former failed.
      In previous version, automata were (uselessly) degeneralized
      before WDBA minimization, causing important slowdowns.
    - Fix compilation with Clang 3.0.
    - Fix a Makefile setup causing a "make check" failure on MacOS X.
    - Fix an mkdir error in the CGI script.

New in spot 0.8 (2011-11-28):

  * Major new features:
    - Spot can read DiVinE models.  See iface/dve2/README for details.
    - The genltl tool can now output 20 different LTL formula families.
      It also replaces the LTLcounter Perl scripts.
    - There is a printer and parser for Kripke structures in text format.
  * Major interface changes:
    - The destructor of all states is now private.  Any code that looks like
      "delete some_state;" will cause an compile error and should be
      updated to "some_state->destroy();".  This new syntax is supported
      since version 0.7.
    - The experimental Nips interface has been removed.
  * Minor changes:
    - The dotty_reachable() function has a new option "assume_sba" that
      can be used for rendering automata with state-based acceptance.
      In that case, acceptance states are displayed with a double
      circle. ltl2tgba (both command line and on-line) Use it to display
      degeneralized automata.
    - The dotty_reachable() function will also display transition
      annotations (as returned by the tgba::transitition_annotation()).
      This can be useful when displaying (small) state spaces.
    - Identifiers used to name atomic proposition can contain dots.
      E.g.: X.Y is now an atomic proposition, while it was understood
      as X&Y in previous versions.
    - The Doxygen documentation is no longer built as a PDF file.
  * Internal improvements:
    - The on-line ltl2tgba CGI script uses a cache to produce faster
      answers.
    - Better memory management for the states of explicit automata.
      Thanks to the aforementioned ->destroy() change, we can avoid
      cloning explicit states.
    - tgba_product has learned how to be faster when one of the operands
      is a Kripke structure (15% speedup).
    - The reduction rule for "a M b" has been improved: it can be
      reduced to "a & b" if "a" is a pure eventuallity.
    - More useless acceptance conditions are removed by SCC simplifications.
  * Bug fixes:
    - Safra complementation has been fixed in cases where more than
      one acceptance conditions where needed to convert the
      deterministic Streett automaton as a TGBA.
    - The degeneralization is now idempotent.  Previously, degeneralizing
      an already degeneralized automaton could add some states.
    - The degeneralization now has a deterministic behavior.  Previously
      it was possible to obtain different output depending on the
      memory layout.
    - Spot now outputs neverclaims with fully parenthesized guards.
      I.e., instead of
         (!x && y) -> goto S1
      it now outputs
         ((!(x)) && (y)) -> goto S1
      This prevents problems when the model defines `x' as
         #define x flag==0
      because !x then evaluated to (!flag)==0 instead of !(flag==0).

New in spot 0.7.1 (2011-02-07):

  * The LTL parser will accept operator ~ (for not) as well
    as --> and <--> (for implication and equivalence), allowing
    formulae from the Büchi Store to be read directly.
  * The neverclaim parser will accept guards of the form
        :: !(...) -> goto ...
    instead of the more commonly used
        :: (!(...)) -> goto ...
    This makes it possible to read neverclaims provided by the Büchi Store.
  * A new ltl2tgba option, -kt, will count the number of "sub-transitions".
    I.e., a transition labelled by "true" counts for 4 "sub-transitions"
    if the automaton uses 2 atomic propositions.
  * Bugs fixed:
    - Fix segfault during WDBA minimization on automata with useless states.
    - Use the included BuDDy library if the one already installed
      is older than the one distributed with Spot 0.7.
    - Fix two typos in the code of the CGI scripts.

New in spot 0.7 (2011-02-01):

  * Spot is now able to read an automaton expressed as a Spin neverclaim.
  * The "experimental" Kripke structure introduced in Spot 0.5 has
    been rewritten, and is no longer experimental.  We have a
    developement version of checkpn using it, and it should be
    released shortly after Spot 0.7.
  * The function to_spin_string(), that outputs an LTL formula using
    Spin's syntax, now takes an optional argument to request
    parentheses at all levels.
  * src/ltltest/genltl is a new tool that generates some interesting
    families of LTL formulae, for testing purpose.
  * bench/ltlclasses/ uses the above tool to conduct the same benchmark
    as in the DepCoS'09 paper by Cichoń et al.  The resulting benchmark
    completes in 12min, while it tooks days (or exhausted the memory)
    when the paper was written (they used Spot 0.4).
  * Degeneralization has again been improved in two ways:
    - It will merge degeneralized transitions that can be merged.
    - It uses a cache to speed up the improvement introduced in 0.6.
  * An implementation of Dax et al.'s paper for minimizing obligation
    formulae has been integrated.  Use ltl2tgba -Rm to enable this
    optimization from the command-line; it will have no effect if the
    property is not an obligation.
  * bench/wdba/ conducts a benchmark similar to the one on Dax's
    webpage, comparing the size of the automata expressing obligation
    formula before and after minimization.  See bench/wdba/README for
    results.
  * Using similar code, Spot can now construct deterministic monitors.
  * New ltl2tgba options:
    -XN: read an input automaton as a neverclaim.
    -C, -CR: Compute (and display) a counterexample after running the
             emptiness check.  With -CR, the counterexample will be
             replayed on the automaton to ensure it is correct
	     (previous version would always compute a replay a
	     counterexample when emptiness-check was enabled)
    -ks: traverse the automaton to compute its number of states and
         transitions (this is faster than -k which will also count
         SCCs and paths).
    -M: Build a deterministic monitor.
    -O: Tell whether a formula represents a safety, guarantee, or
        obligation property.
    -Rm: Minimize automata representing obligation properties.
  * The on-line tool to translate LTL formulae into automata
    has been rewritten and is now at http://spot.lip6.fr/ltl2tgba.html
    It requires a javascript-enabled browser.
  * Bug fixes:
    - Location of the errors messages in the TGBA parser where inaccurate.
    - Various warning fixes for different versions of GCC and Clang.
    - The neverclaim output with ltl2tgba -N or -NN used to ignore any
      automaton simplification performed after degeneralization.
    - The formula simplification based on universality and eventuality
      had a quadratic run-time.

New in spot 0.6 (2010-04-16):

  * Several optimizations to improve some auxiliary steps
    of the LTL translation (not the core of the translation):
    - Better degeneralization
    - SCC simplifications has been tuned for degeneralization
      (ltl2tgba now has two options -R3 and -R3f: the latter will
      remove every acceptance condition that used to be removed
      in Spot 0.5 while the former will leave useless acceptance conditions
      going to accepting SCC.  Experience shows that -R3 is more
      favorable to degeneralization).
    - ltl2tgba will perform SCC optimizations before degeneralization
      and not the converse
    - We added a syntactic simplification rule to rewrite F(a)|F(b) as F(a|b).
      We only had a rule for the more specific FG(a)|FG(b) = F(Ga|Gb).
    - The syntactic simplification rule for F(a&GF(b)) = F(a)&GF(b) has
      be disabled because the latter formula is in fact harder to translate
      efficiently.
  * New LTL operators: W (weak until) and its dual M (strong release)
    - Weak until allows many LTL specification to be specified more
      compactly.
    - All LTL translation algorithms have been updated to
      support these operators.
    - Although they do not add any expressive power, translating
      "a W b" is more efficient (read smaller output automaton) than
      translating the equivalent form using the U operator.
    - Basic syntactic rewriting rules will automatically rewrite "a U
      (b | G(a))" and "(a U b)|G(a)" as "a W b", so you will benefit
      from the new operators even if you do not use them.  Similar
      rewriting rules exist for R and M, although they are less used.
  * New options have been added to the CGI script for
    - SVG output
    - SCC simplifications
  * Bug fixes:
    - The precedence of the "->" and "<->" Boolean operators has been
      adjusted to better match other tools.
      Spot <= 0.5 used to parse "a & b -> c & d" as "a & (b -> c) & d";
      Spot >= 0.6 will parse it as "(a & b) -> (c & d)".
    - The random graph generator was fixed (again!) not to produce
      dead states as documented.
    - Locations in the error messages of the LTL parser were off by one.

New in spot 0.5 (2010-02-01):

  * We have setup two mailing lists:
    - <spot-announce@lrde.epita.fr> is read-only and will be used to
      announce new releases.  You may subscribe at
      https://www.lrde.epita.fr/mailman/listinfo/spot-announce
    - <spot@lrde.epita.fr> can be used to discuss anything related
      to Spot.  You may subscribe at
      https://www.lrde.epita.fr/mailman/listinfo/spot-announce
  * Two new LTL translations have been implemented:
    - eltl_to_tgba_lacim() is a symbolic translation for ELTL based on
      Couvreur's LaCIM'00 paper.  For this translation (available with
      ltl2tgba's option -le), all operators are described as finite
      automata.  A default set of operators is provided for LTL
      (option -lo) and user may add more automaton operators.
    - ltl_to_taa() is a translation based on Tauriainen's PhD thesis.
      LTL is translated to "self-loop" alternating automata
      and then to Transition-based Generalized Automata.  (ltl2tgba's
      option -taa).
    The "Couvreur/FM" translation remains the best LTL translation
    available in Spot.
  * The data structures used to represent LTL formulae have been
    overhauled, and it resulted in a big performence improvement
    (in time and memory consumption) for the LTL translation.
  * Two complementation algorithms for state-based Büchi automata
    have been implemented:
    - tgba_kv_complement is an on-the-fly implementation of the
      Kupferman-Vardi construction (TCS'05) for generalized acceptance
      conditions.
    - tgba_safra_complement is an implementation of Safra's
      complementation.  This algorithm takes a degeneralized Büchi
      automaton as input, but our implementation for the Streett->Büchi
      step will produce a generalized automaton in the end.
  * ltl2tgba has gained several options and the help text has been
    reorganized.  Please run src/tgbatest/ltl2tgba without arguments
    for details.  Couvreur/FM is now the default translation.
  * The ltl2tgba.py CGI script can now run standalone.  It also offers
    the Tauriainen/TAA translation, and some options for SCC-based
    reductions.
  * Automata using BDD-encoded transitions relation can now be pruned
    for useless states symbolically using the delete_unaccepting_scc()
    function.  This is ltl2tgba's -R3b option.
  * The SCC-based simplification (ltl2tgba's -R3 option) has been
    rewritten and improved.
  * The "*" symbol, previously parsed as a synonym for "&" is no
    longer recognized.  This makes room for an upcoming support of
    rational operators.
  * More benchmarks in the bench/ directory:
    - gspn-ssp/    some benchmarks published at ACSD'07,
    - ltlcounter/  translation of a class of LTL formulae used by
                     Rozier & Vardi at SPIN'07
    - scc-stats/   SCC statistics after translation of LTL formulae
    - split-product/ parallelizing gain after splitting LTL automata
  * An experimental Kripke interface has been developed to simplify
    the integration of third party tools that do not use acceptance
    conditions and that have label on states instead of transitions.
    This interface has not been used yet.
  * Experimental interface with the Nips virtual machine.
    It is not very useful as Spot isn't able to retrieve any property
    information from the model.  This will just check assertions.
  * Distribution:
    - The Boost C++ library is now required.
    - Update to Autoconf 2.65, Automake 1.11.1, Libtool 2.2.6b,
      Bison 2.4.1, and Swig 1.3.40.
    - Thanks to the newest Automake, "make check" will now
      run in parallel if you use "make -j2 check" or more.
  * Bug fixes:
    - Disable warnings from the garbage collection of BuDDy, it
      could break the standard output of ltl2tgba.
    - Fix several C++ constructs to ensure Spot will build with
      GCC 4.3, 4.4, and older 3.x releases, as well as with Intel's
      ICC compiler.
    - A very old bug in the hash function for LTL formulae caused Spot
      to sometimes (but very rarely) consider two different LTL formulae
      as equal.

New in spot 0.4 (2007-07-17):

  * Upgrade to Autoconf 2.61, Automake 1.10, Bison 2.3, and Swig 1.3.31.
  * Better LTL simplifications.
  * Don't initialize Buddy if it has already been initialized (in case
    the client software is already using Buddy).
  * Lots of work in the greatspn interface for our ACSD'05 paper.
  * Bug fixes:
    - Fix the random graph generator not to produce dead states as documented.
    - Fix synchronized product in case both side use acceptance conditions.
    - Fix some syntax errors with newer versions of GCC.

New in spot 0.3 (2006-01-25):

  * lbtt 1.2.0
  * The CGI script for LTL translation also offers emptiness check algorithms.
  * tau03_opt_search implements the "ordering heuristic".
    (Submitted by Heikki Tauriainen.)
  * A couple of bugs were fixed into the LTL or automata simplifications.

New in spot 0.2 (2005-04-08):

  * Emptiness checks:
    - the new spot::option_map class is used to pass options to
      emptiness-check algorithms.
    - the new emptiness_check_instantiator class is used to turn a
      string such as `algorithm(option1, option2)' into an actual
      instance of this emptiness-check algorithm with the given
      options.  All tools use this.
    - tau03_opt_search implements the "condition heuristic".
      (Suggested by Heikki Tauriainen.)
  * Minor bug fixes.

New in spot 0.1 (2005-01-31):

  * Emptiness checks:
    - They all follow the same interface, and gather statistical data.
    - New algorithms: gv04.hh, se05.hh, tau03.hh, tau03opt.hh
    - New options for couvreur99: poprem and group.
    - reduce_run() try to reduce accepting runs produced by emptiness checks.
    - replay_run() ensure accepting runs are actually accepting runs.
  * New testing tools:
    - ltltest/randltl: Generate random LTL formulae.
    - tgbatest/randtgba: Generate random TGBAs.  Optionally multiply them
        against random LTL formulae.  Optionally check them for emptiness
        with all available algorithms.  Optionally gather statistics.
  * bench/emptchk/: Contains scripts that benchmark emptiness-checks.
  * Split the degeneralization proxy in two:
    - tgba_tba_proxy  uses at most max(N,1) copies
    - tgba_sba_proxy  uses at most 1+max(N,1) copies and has a
                      state_is_accepting() method
  * tgba::transition_annotation annotate a transition with some string.
    This comes handy to associate that transition to its high-level name.
  * Preliminary support for Event-based GBA (the evtgba*/ directories).
    This might as well disappear in a future release.
  * LTL formulae are now sorting using their string representation, instead
    of their memory address (which is still unique).  This makes the output
    of the various functions more deterministic.
  * The Doxygen documentation is now organized using modules.

New in spot 0.0x (2004-08-13):

  * New atomic_prop_collect() function: collect atomic propositions
    in an LTL formula.
  * Fix several typos in documentation, and some warnings in the code.
  * Now compiles on Darwin and Cygwin.
  * Upgrade to Automake 1.9.1, and lbtt 1.1.2.
    (And drop support for older lbtt versions.)
  * Support newer versions of Valgrind (>= 2.1.0).

New in spot 0.0v (2004-06-29):

  * LTL formula simplifications using basic rewriting rules,
    a-la Wring syntactic approximations, and Etessami's universal
    and existential classes.
     - Function reduce() in ltlvisit/reduce.hh is the main interface.
     - Can be tested with the CGI script.
  * TGBA simplifications using direct simulation, delayed simulation,
    and SCC-based simplifications.  This is still experimental.
  * The LTL parser will now read LTL formulae written using Wring's syntax.
  * ltl2tgba_fm() now has options for on-the-fly fair-loop approximations,
    and Modella-like branching-postponement.
  * GreatSPN interface:
     - The `declarative_environment' is now part of Spot itself rather than
       part of the interface with GreatSPN.
     - the RG and SRG interface can deal with dead markings in three
       ways (omit deadlocks from the state graph, stutter on the deadlock
       and consider as a regular behavior, or stutter and distinguish the
       deadlock with a property).
     - update SSP interface to Soheib Baarir latest work.
  * Preliminary Python bindings for BuDDy's FDD and BVEC.
  * Upgrade to BuDDy 2.3.

New in spot 0.0t (2004-04-23):

  * `emptiness_check':
      - fix two bugs in the computation of the counter example,
      - revamp the interface for better customization.
  * `never_claim_reachable': new function.
  * Introduce annonymous BDD variables in `bdd_dict', and start
    to use it in `ltl_to_tgba_fm'.
  * Offer never claim in the CGI script.
  * Rename EESRG as SSP, and offer specialized variants of the
    emptiness_check.

New in spot 0.0r (2004-03-08):

  * In ltl_to_tgba_fm:
    - New option `exprop' to optimize determinism.
    - Make the `symbolic indentification' from 0.0p optional.
  * `nonacceptant_lbtt_reachable' new function to help getting
    accurate statistics from LBTT.
  * Revamp the cgi script's user interface.
  * Upgrade to lbtt 1.0.3, swig 1.3.21, automake 1.8.3

New in spot 0.0p (2004-02-03):

  * In ltl_to_tgba_fm:
    - identify states with identical symbolic expansions
      (i.e., identical continuations)
    - use Acc[b] as acceptance condition for Fb, not Acc[Fb].
  * Update and speed-up the cgi script.
  * Improve degeneralization.

New in spot 0.0n (2004-01-13):

  * emptiness_check::check2() is a variant of Couvreur's emptiness check that
    explores visited states first.
  * Build the EESRG supporting code condinally, as the associated
    GreatSPN changes have not yet been contributed to GreatSPN.
  * Add a powerset algorithm (determinize TGBA ignoring acceptance
    conditions, i.e., as if they were used to recognize finite languages).
  * tgba_explicit::merge_transitions: merge transitions with same source,
    destination, and acceptance condition.
  * Run test cases within valgrind.
  * Various bug fixes.

New in spot 0.0l (2003-12-01):

  * Computation of prime implicants.  This simplify the output of
    ltl_to_tgba_fm, and allows conditions to be output as some of
    product in dot output.
  * Optimize translation of GFy in ltl_to_tgba_fm.
  * tgba_explicit supports arbitrary binary formulae on transitions
    (only conjunctions were allowed).

New in spot 0.0j (2003-11-03):

  * Use hash_map's instead of map's almost everywhere.
  * New emptiness check, based on Couvreur's algorithm.
  * LTL propositions can be put inside doublequotes to disambiguate
    some constructions.
  * Preliminary support for GreatSPN's EESRG.
  * Various bug fixes.

New in spot 0.0h (2003-08-18):

  * More python bindings:
    - "import buddy" works (see wrap/python/tests/bddnqueen.py for an example),
    - almost all the Spot API is now available via "import spot".
  * wrap/python/cgi/ltl2tgba.py is an LTL-to-Büchi translator that
    work as as a cgi script.
  * Couvreur's FM'99 ltl-to-tgba translation.

New in spot 0.0f (2003-08-01):

  * More python bindings, still only for the spot::ltl:: namespace.
  * Functional GSPN interface.  (Enable with --with-gspn=directory.)
  * The LTL scanner recognizes /\, \/, and xor.
  * Upgrade to lbtt 1.0.2.
  * tgba_tba_proxy is an on-the-fly degeneralizer.
  * Implements the "magic search" algorithm.
    (Works only on a tgba_tba_proxy.)
  * Tgba's output algorithms (save(), dotty()) now non-recursive.
  * During products, succ_iter will optimize its set of successors
    using information computed from the current product state.
  * BDD dictionaries are now shared between automata and.  This
    gets rid of all the BDD-variable translating machinery.

New in spot 0.0d (2003-07-13):

  * Optimize translation of G operators occurring at the root
    of a formula (or its immediate children when the root is a
    conjunction).  This saves two BDD variables per G operator.
  * Distribute lbtt, and run it during `make check'.
  * First sketch of GSPN interface.
  * succ_iter_concreate::next() completely rewritten.
  * Transitions are now labelled by boolean formulae (not only
    conjunctions).
  * Documentation:
    - Output collaboration diagrams.
    - Build and distribute PDF manual.
  * Many bug fixes.

New in spot 0.0b (2003-06-26):

  * Everything.
