Zielonka algorithm used to iterate over an std::unordered_set, thus
producing different strategies depending on compiler...
* spot/misc/game.cc: replace std::unordered_set with std::set
Related to issue #351.
* spot/twaalgos/sccfilter.cc: When handling weak automata, we know
they are very-weak if the SCC count is equal to the number of states.
* tests/core/dca2.test, tests/core/monitor.test,
tests/core/parity2.test, tests/core/randomize.test,
tests/core/readsave.test, tests/core/remfin.test,
tests/core/sccsimpl.test, tests/core/wdba2.test,
tests/python/dualize.py, tests/python/remfin.py: Adjust output.
... instead of inherently-weak. The reason they were tagged
as inherently-weak is historical: this property was introduced
1.5 years before the weak propery.
Fixes#351.
* spot/twaalgos/ltl2tgba_fm.cc: Use prop_weak() instead of
prop_inherently_weak(). Also be more conservative about the use of
single_acc when unambiguous automata are generated.
This improves gf_guarantee_to_ba() on formulas GF(φ) where the
automaton for F(φ) as several leading transiant SCCs. E.g.,
GF(a <-> XXXa) where we know get results that are as good as
those of delag without loosing on the cases where delag's technique
would actually produce two big automata.
* spot/twaalgos/gfguarantee.cc: Implement this.
* spot/twaalgos/gfguarantee.hh, NEWS: Document it.
* tests/core/ltl2tgba2.test, tests/core/ltl3ba.test: Add test cases.
Fixes#347.
* spot/twaalgos/minimize.cc, spot/twaalgos/minimize.hh: An SCC of the
determinized automaton is marked as accepting if it intersects any
accepting SCC of the original automaton. This can be checked by
making a product. As a consequence the output of minimize_wdba() now
always includes the original language, and minimize_obligation() only
needs a single containment check.
* NEWS: Mention this.
Together with the previous patch, this Fixes#353.
Implementing this required to extend our interface two support
two-parameter patterns.
* spot/gen/formulas.cc, spot/gen/formulas.hh: Implement it.
* bin/genltl.cc: Add --sejk-f.
* bin/common_output.cc, bin/common_output.hh: Adjust to handle
"line numbers" that are not integers (e.g., "3,2"), since those
are used to display pattern parameters.
* bin/ltlfilt.cc: Adjust.
* python/spot/gen.i: Add support for two-parameters patterns.
* tests/core/genltl.test, tests/python/gen.ipynb: Augment.
* NEWS: Mention it.
These correspond to the first three blocks of table 1 in S. Sickert,
J. Esparza, S. Jaax, and J. Křetínský: Limit-Deterministic Büchi
Automata for Linear Temporal Logic. CAV'16. LNCS 9780.
For #353.
* spot/gen/formulas.cc, spot/gen/formulas.hh, bin/genltl.cc: Implement
the new families.
* tests/core/genltl.test: Test it.
* bin/man/genltl.x, NEWS: Document it.
When dealternating the VWAA for GFa, our result had two states that
could not be fused by simulation because of unmatched acceptance mark.
With this change, the result can be simplified.
* spot/twaalgos/alternation.cc: Here.
* tests/core/alternating.test, tests/python/alternation.ipynb: Update
test case.
* NEWS: Mention it.
* spot/misc/bitset.cc, spot/misc/bitset.hh (set, clear):
Turn asserts into exceptions.
* spot/twa/acc.hh (mark_t): As a consequence, the
constructor is not noexcept anymore.
* tests/core/acc.cc, tests/python/except.py: More tests.
The exception raised by << and >> when shifting mark_t by too many
bits are only enabled in SPOT_DEBUG, as those operations are quite
low-level. However we were always testing them, and although we
wanted them to be active in Python, it was not always the case.
* spot/twa/acc.hh: introduce max_accsets() as
a static constexpr method, so we can see it in Python.
* spot/misc/bitset.hh: Fix preprocessing directive
so the check is actually enabled when compiling the Python
bindings.
* bin/autcross.cc, bin/autfilt.cc, bin/ltlcross.cc: Use max_accsets().
* tests/core/acc.cc: Comment out the shifting exception when
SPOT_DEBUG is unset.
* tests/python/except.py: Make sure the exception is always raised in
Python.
* spot/misc/bitset.hh: add methods set() and clear()
* spot/twa/acc.hh: deprecate comparison of mark_t with unsigned, and
rely more on biset for efficiency
Fixes#348, reported by Jeroen Meijer.
* bin/autfilt.cc: If -B is used with many --product,
degeneralize intermediate products as needed.
* NEWS: Mention the change.
* tests/core/prodchain.test: New file.
* tests/Makefile.am: Add it.
* spot/twa/acc.cc, spot/twa/acc.hh: Fix reporting of
overflow.
* tests/core/acc.cc: Adjust.
This longstanding issue caused some emptiness checks options to be
ignored in the test suite, reducing coverage.
* spot/misc/optionmap.cc (set): Insert the other keys without
erasing the existing ones.
This allows to represent more than 32 acceptance marks.
* configure.ac: add an option to specify the number of marks
* spot/twa/acc.hh: implement it
* tests/python/acc_cond.ipynb, tests/core/acc.cc,
tests/core/ltlcross3.test: update tests
* NEWS: document it
* bin/randltl.cc: fix an include
