From a discussion with Petr Rockai: "the possibly non-intuitive bit
that you probably didn't notice is that the hashset is supposed to be
passed to each thread by value. The copy semantics of the entire
hashset are that of a shared pointer: multiple copies share the same
underlying data. Each thread *must* have its own private copy of the
hashset object (there are bits of state that each thread needs for
bookkeeping and those must not be shared)."
* spot/ltsmin/ltsmin.cc: here.
Swarming implies that a single instance of the kripke
structure (or product) will be explored by diffrent threads
with their own exploration order. Most of the modification
aims to have a thread safe kripke structure.
* spot/kripke/kripke.hh, spot/ltsmin/ltsmin.cc,
spot/ltsmin/ltsmin.hh, spot/mc/ec.hh,
spot/mc/intersect.hh, spot/mc/reachability.hh,
spot/misc/hash.hh, spot/twacube/twacube.hh,
tests/core/twacube.test, tests/ltsmin/modelcheck.cc: here.
Passing atomic propositions by reference allows to
save very little memory so it doesn't worth complexifying
memory management.
* spot/twacube/twacube.cc, spot/twacube/twacube.hh: here.
From the working draft: "The macro ATOMIC_FLAG_INIT shall be defined
in such a way that it can be used to initialize an object of type
atomic_flag to the clear state. The macro can be used in the form:
atomic_flag guard = ATOMIC_FLAG_INIT; It is unspecified whether the
macro can be used in other initialization contexts."
* bricks/brick-shmem.h: here.
Warning! this patch introduces redundancy (or difference)
between wether you ask for a kripkecube or a kripke.
* spot/ltsmin/ltsmin.cc, spot/ltsmin/ltsmin.hh: here.
In order to reuse the computation of the
intersection between kripke and twa efficiently,
we use template inheritance through the
"mixin templates" technique.
* spot/Makefile.am, spot/mc/Makefile.am,
spot/mc/ec.hh, spot/mc/unionfind.cc,
spot/mc/unionfind.hh: here.
Use ltldo:... instead of error:... and warning:... and also improve
the diagnostic displayed after a translation failure to mention the
tool and formula.
Incidentally, this fixes a spurious test case failure observed by
Philipp Schlehuber on CentOS7.7 where glibc 2.17 is installed. With
this system, when posix_spawn() starts a binary that does not exist,
it returns success and let the child die with exit code 127. On more
recent glibc, posix_spawn() manages to return execve()'s errno, as if
the child had not been created. We handle those two different ways to
fail, but before this patch one used to print "error:..." and the
other "ltldo:...".
* bin/ltldo.cc: Display the program_name in error message. Display
the command name and formula on translation failure.
* tests/core/ltldo.test: Adjust test case.
* NEWS: Mention the fix.
Bison <3.6 used to complain about "$undefined", while Bison >=3.6 now
write "invalid token".
* tests/core/parseaut.test, tests/core/parseerr.test,
tests/core/sugar.test: Adjust expected diagnostics to match Bison pre
and post 3.6.
Closes#405. This shows no difference on the test suite,
but that is thanks to the previous patch: without it, an
example in automata.ipynb would have an extra edge.
* spot/twaalgos/cleanacc.cc (simplify_acceptance): Call
unit_propagation() before simplify_complementary_marks_here() and
fuse_marks_here(), because that is simpler to perform.
