2026-11-15  Alexandre Duret-Lutz  <adl@lre.epita.fr>

	Release Spot 2.16.3

	* NEWS, configure.ac, doc/org/setup.org: Adjust for release.

2026-11-14  Alexandre Duret-Lutz  <adl@lrde.epita.fr>

	ltl2tgba_fm: diagnose out-of-range marks with many promises

	The FM translation allocates one acceptance color per promise without
	any bound, but bdd_to_mark() and split_cube_as_cond_dest_and_acc()
	converted these colors with mark_t::set(), which assumes that its index
	is smaller than mark_t::max_accsets().  With a formula requiring more
	promises than that (e.g., the 64 clients of simple_arbiter_enc64.tlsf),
	debug builds reported "bit index is out of bounds", while other builds
	invoked undefined behavior.

	Restricting the allocation of colors would be too strong, since the OTF
	interface can explore such formulas symbolically (via succ_as_bdd())
	as long as no mark_t is built, so the limit is checked when a mark is
	constructed instead.

	* spot/twa/acc.hh (mark_t::set_checked): New method, similar to set()
	but reporting the excess of acceptance sets instead of assuming that
	the index is in range.
	* spot/twaalgos/ltl2tgba_fm.cc (translate_dict::bdd_to_mark)
	(translate_dict::split_cube_as_cond_dest_and_acc): Use it to turn
	colors into marks.
	* tests/python/acc.py: Test mark_t::set_checked().

2026-11-14  Alexandre Duret-Lutz  <adl@lrde.epita.fr>

	game: fix a spurious assertion in fix_strat_acc()

	Commit 034fa6a2b ("game: add heuristic for shorten the strategy")
	moved the assertion that the priority of an outgoing edge does not
	exceed max_priority out of the "subgame_[e_fix.dst] >= rd" guard, so
	it was also evaluated on edges leading to older subgames, where a
	priority above max_priority with the same parity is legitimate: those
	edges are outside the current subgame, which is the only place where
	max_priority bounds the priorities.  Such edges were already excluded
	from the repair by the min_winning_priority/max_priority test, so the
	solver's behavior is unchanged.  This aborted ltlsynt on the
	ltl2dba_R4 and ltl2dpa12 specifications of the SyntComp'23 benchmark.

	* spot/twaalgos/game.cc (parity_game::fix_strat_acc): Restrict the
	assertion to edges that stay in the current subgame.
	* tests/core/ltlsynt.test: Add the LTL versions of those two
	specifications.
	* NEWS: Mention the bug fix.

2026-11-14  Alexandre Duret-Lutz  <adl@lrde.epita.fr>

	relabel_bse: reimplement to rename without overlap

	Fixes #667.

	* spot/tl/relabel.cc: Rewrite relabel_bse() to compute a minimal
	renaming.  A Boolean subformula is now replaced by a fresh atomic
	proposition iff every occurrence of its atomic propositions is inside
	it, and the subformulas hiding the most atomic propositions are
	selected first, so that two renamed subformulas never share an atomic
	proposition.  Additionally, the BDD check used to detect tautological
	or contradictory pieces is skipped for pieces in which each atomic
	proposition occurs at most once.
	* spot/misc/bitvect.hh (bitvect): Add a constructor taking a size, make
	intersects() const, and initialize local_storage_ in the copy
	constructor.
	* tests/core/ltlrel.test, tests/core/ltlfilt.test,
	tests/python/relabel.py: Adjust expectations, and check that the
	renaming is legal (restores the formula, never increases the number of
	atomic propositions, and keeps the renamed pieces AP-disjoint).
	* tests/python/_mealy.ipynb: Update a label whose operands are ordered
	by formula id().
	* NEWS: Mention the change.

2026-11-14  Alexandre Duret-Lutz  <adl@lrde.epita.fr>

	mtdswa: label α->ff, α<->ff, and α xor tt by !α during propeq

	obligation_to_mtdswa() left labels such as Gi->(Xp & X!p & G(...))
	unsimplified.  Propositional equivalence knew the encoding was !x, but
	had no formula attached to it, so the residual kept its implication
	and obligation_is_accepting() called it accepting.  This caused to be
	G(i)->G(i<->X(Xp & X!p)) incorrectly reported as realizable.  Now such
	a label is simplified to !Gi which is correctly classified as
	rejecting.

	Fixes #666, reported by Guillermo Perez.

	* spot/twaalgos/mtdswa.cc (simple_ltl_translator::propeq_encode):
	Register the complement of the surviving operand when Implies, Equiv,
	or Xor erases all its others (level 0 only, never for constants), and
	reuse the cached encoding of an atom for its negation at deeper levels,
	so that Xp & X!p cancels.
	* tests/python/mtdswa.py: Test X(f) constants, and propositionally
	false sub-formulas under X.
	* tests/python/ltlf2dfa.ipynb: Refresh the BDD node ids in the saved
	outputs.
	* NEWS: Mention the fix.

2026-11-14  Alexandre Duret-Lutz  <adl@lre.epita.fr>

	Clang should use -fno-assume-unique-vtables for shared libraries

	The previous fix for issue #609 did not work because enable_shared was
	tested before being set.

	* configure.ac: Move LT_INIT abouve the enable_shared test.
	* NEWS: Mention the fix.

2026-11-03  Alexandre Duret-Lutz  <adl@lre.epita.fr>

	* NEWS, configure.ac: Bump version to 2.16.2.dev.

2026-11-03  Alexandre Duret-Lutz  <adl@lre.epita.fr>

	Release Spot 2.16.2

	* NEWS, configure.ac, doc/org/setup.org: Bump version.

2026-11-02  Alexandre Duret-Lutz  <adl@lrde.epita.fr>

	ltl2tgba_fm: deprecate the unsound simplifier argument

	The simplifier argument of ltl_to_tgba_fm() (and of the newer
	ltl_to_tgba_fm_otf()) used to simplify the LTL formula representing
	each state of the automaton before computing its successors.  This is
	unsound: a simplifier may rewrite a destination in a way that moves a
	promise under a temporal operator (Xd & FG!d becomes X(d & FG!d)),
	while the acceptance marks put on a transition are collected from the
	unsimplified destination.  The promise is then never charged in the
	state where it should be, and the automaton accepts too much.

	This option sat unused for years, until version 2.16 changed
	ltl_satisfiable() to explore the automaton on-the-fly, and this caused
	G(b -> XXd) & Gb & FG!d to be reported as satisfiable.

	Fixes #663, reported by Ben Andrew.

	* spot/twaalgos/ltl2tgba_fm.cc: Remove impl::dest_simplifier, and stop
	simplifying the destination formulas in succ_as_edges() and
	succ_as_acc_and_dest(); a caller-supplied simplifier is now only
	applied to the input formula.
	* spot/twaalgos/ltl2tgba_fm.hh: Document the deprecation of the
	simplifier argument of ltl_to_tgba_fm() and ltl_to_tgba_fm_otf(),
	with the Xd & FG!d counterexample.
	* spot/tl/sat.cc (otf_satisfiable): Do not pass a simplifier to the
	explorer.
	* tests/python/ltlsimple.py: Check that G(b -> XXd) & Gb & FG!d is
	unsatisfiable.
	* tests/core/ikwiad.cc: Drop the -fr option, now equivalent to the
	default.
	* tests/core/spotlbtt.test: Drop the two disabled algorithms using -fr.
	* NEWS: Mention the deprecation and the bug fix.
	* THANKS: Add Ben.

2026-11-02  Alexandre Duret-Lutz  <adl@lre.epita.fr>

	* bin/spot-x.cc: Fix "acd" doc.
