Arnaud Giersch [Thu, 9 Mar 2023 13:18:32 +0000 (14:18 +0100)]
Eager mode is for the first case only (thx adegomme).
[ci-skip]
Maxwell Pirtle [Thu, 9 Mar 2023 13:08:27 +0000 (14:08 +0100)]
Add test to verify topological ordering of EventSet
The topological ordering was moved into EventSet
in a prior commit since it is useful to order
events in an EventSet that reside outside of a
Configuration. This commit adds a stress test
that verifies that the topological orderings
of all possible subsets of a collection of
events obey the topological ordering invariant
after being ordered (in both directions)
Arnaud Giersch [Thu, 9 Mar 2023 10:36:49 +0000 (11:36 +0100)]
Improve doc for smpi/buffering.
Describe the behavior like for smpi/send-is-detached-thresh.
Arnaud Giersch [Wed, 8 Mar 2023 22:15:18 +0000 (23:15 +0100)]
Fix asserts for when received size is 0 (message type is invalid).
Arnaud Giersch [Wed, 8 Mar 2023 15:36:45 +0000 (16:36 +0100)]
Depreciate unused functions s4u::Comm::copy_{buffer,pointer}_callback().
Arnaud Giersch [Wed, 8 Mar 2023 15:36:13 +0000 (16:36 +0100)]
Inline s4u::Comm::copy_pointer_callback which is the default callback.
Maxwell Pirtle [Wed, 8 Mar 2023 14:27:43 +0000 (15:27 +0100)]
Begin filling in computations of ex(C) and en(C)
This commit introduces the first steps towards
computing ex(C) and en(C) using many of the
additions of the previous phases. An important
part of the computation is so far missing, however,
viz. the determination of whether or not an action
`a` is enabled at some point in the past. We'll
need to decipher what's going on in tiny_simgrid
to fully determine what to do here; for now, we
simply ignore adding an event entirely and just
say that `a` was disabled at all points in the past.
Arnaud Giersch [Wed, 8 Mar 2023 14:09:13 +0000 (15:09 +0100)]
Depreciate ForcefulKillException::try_n_catch().
It was used by the Java bindings.
Arnaud Giersch [Wed, 8 Mar 2023 13:23:00 +0000 (14:23 +0100)]
Obvious typo.
Maxwell Pirtle [Wed, 8 Mar 2023 13:11:39 +0000 (14:11 +0100)]
Fix logic for mutual conflict between two events
Prior to this commit, detecting whether two
events were in conflict involved looking at all
of the events with respect to all others. However,
you must instead *only* look at how the event `e`
and `e'` relate to the other event's history:
we wouldn't care if some other event is in conflict
with `e'` if `e` is not that event
Maxwell Pirtle [Wed, 8 Mar 2023 12:12:00 +0000 (13:12 +0100)]
Add conflict-free invariant check to Configuration
When adding events to a configuration, we want to
ensure that the configuration remains both a)
causally closed and b) conflict-free. The latter
required that the tests be modified as they effectively
assumed that the transitions associated with each
event were independent of one another. Since the
`Transition` base class is dependent by default, this
was at odds with the assumptions
Arnaud Giersch [Wed, 8 Mar 2023 10:23:17 +0000 (11:23 +0100)]
Use #include <...> for foreign header files.
Files in teshsuite/smpi/mpich3-test/ were not considered.
Maxwell Pirtle [Wed, 8 Mar 2023 10:18:50 +0000 (11:18 +0100)]
Add preliminary tests for checking event conflicts
Maxwell Pirtle [Wed, 8 Mar 2023 08:13:15 +0000 (09:13 +0100)]
Add conflict detection to EventSet
Detecting whether or not a collection of events
are in conflict important for computing en(C)
and for K-partial alternatives. We simply
follow the definition of conflicts. More
esoteric methods may exist that are faster,
but for now let's keep it simple: we can always
add those in later incrementally
Martin Quinson [Wed, 8 Mar 2023 00:10:15 +0000 (00:10 +0000)]
Merge branch 'udpor-phase4' into 'master'
Phase 4 of UDPOR Integration: Variable For-loop
See merge request simgrid/simgrid!136
Maxwell Pirtle [Tue, 7 Mar 2023 13:36:03 +0000 (14:36 +0100)]
Add ability to restrict maximum subset size
This commit adds the ability to restrict the
size of the maximal subsets that are produced
by the maximal_subsets_iterator. This filtering
significantly reduces the number of subsets that
need to be searched in more complicated graphs.
The advantage of adding the filtering directly
to the maximal_subsets_iterator over using a
higher-order iterator such as boost::filter_iterator
is that branches can be pruned immediately. Since
the maximal_subsets_iterator performs its search
in depth-first order, something like
boost::filter_iterator, while technically equivalent,
would have to actually perform the iteration over
entire branches that are otherwise useless due to
their size.
Searching maximal subsets of a restricted size is
useful when determining conflicts between sets of
events: we want to look at pairs of maximal events
Maxwell Pirtle [Tue, 7 Mar 2023 13:05:01 +0000 (14:05 +0100)]
Add asserts for Configuration vs EventSet
This commit adds two asserts to ensure that
traversing a configuration is equivalent to
traversing the events of that configuration.
That is, the assertions check that a configuration
can effectively be treated as a set of events
when it comes to traversal of maximal sets and
computing a topological ordering
Maxwell Pirtle [Tue, 7 Mar 2023 12:48:29 +0000 (13:48 +0100)]
Allow iteration over maximal subsets of an EventSet
The maximal_subsets_iterator was updated to allow
traversal over the subsets of events of some set `E`
which are maximal (i.e. causally-free). It is a
simple addition which extends the power of the iterator
to EventSets more generally (as opposed to strictly
for Configurations)
Maxwell Pirtle [Tue, 7 Mar 2023 12:41:41 +0000 (13:41 +0100)]
Move topological ordering to EventSet
The events of an EventSet can be ordered
topologically by considering them as part
of the configuration containing the histories
of all events in the set. The trick is to
ignore adding in an event into the ordering
if the event is not contained in the original
set itself
Maxwell Pirtle [Tue, 7 Mar 2023 09:05:45 +0000 (10:05 +0100)]
Rename `EventSet::is_maximal_event_set()`
The method `is_maximal_event_set()` was renamed
to simply `is_maximal()`, since the question is
already being asked about some event set (i.e.
the `_event_set` bit is redundant)
Maxwell Pirtle [Tue, 7 Mar 2023 08:10:58 +0000 (09:10 +0100)]
Add some basic tests for variable_for_loop
Some tests were added to confirm that the
variable_for_loop iterator behaved as expected.
Arnaud Giersch [Mon, 6 Mar 2023 15:16:49 +0000 (16:16 +0100)]
Fixes in docs/source/tuto_* (untested).
Maxwell Pirtle [Mon, 6 Mar 2023 15:13:50 +0000 (16:13 +0100)]
Write maximal_subsets_iterator_wrapper with xbt utils
Maxwell Pirtle [Mon, 6 Mar 2023 14:28:05 +0000 (15:28 +0100)]
Fix edge cases in variable_for_loop
Some tests were added which detected an off-by-one
error in the implementation logic of the variable_for_loop.
Subsequent commits will add more tests to the iterator
for further verification
Maxwell Pirtle [Mon, 6 Mar 2023 13:41:24 +0000 (14:41 +0100)]
Write LazyPowerSet in terms of iterator_wrapper
Maxwell Pirtle [Mon, 6 Mar 2023 13:01:48 +0000 (14:01 +0100)]
Add iterator_wrapping convenience container
The iterator_wrapping structure is a light-weight
container whose contents are traversed according
to the behavior of the iterator that it creates
when asked. The wrapping can only construct
a container around an iterator whose termination
condition can be succinctly represented with a
default-constructed argument. It is not always
the case that such a default-constructed iterator
represents the end, so this should be kept in made
(although this is generally the convention).
Creating an instance of a wrapping allows, e.g.,
for the traversal over a collection without explicit
need to construct the iterators/loop manually.
For example, a struct `Iterable` may define a method
`make_complicated_forward_iterator_over_me()`
that can be called and used inside e.g. an auto-based
for-loop. This helps with readability immensely.
Maxwell Pirtle [Mon, 6 Mar 2023 09:32:54 +0000 (10:32 +0100)]
Add first implementation of variable for-loop
Martin Quinson [Mon, 6 Mar 2023 09:50:07 +0000 (10:50 +0100)]
Fix clang build and make output more portable/predictible
Maxwell Pirtle [Fri, 3 Mar 2023 12:34:02 +0000 (13:34 +0100)]
Add computation for minimally reproducible sets
A minimally reproducible set of a configuration
is simply the smallest set of events whose
collective histories joint together cover every
event in the configuration. Effecively, it is
the maximal set of C whose history equals C.
The computation builds upon the maximal_subsets_iterator
introduced in a prior phase and exploits all of
the work that already went into computing the
subsets
Martin Quinson [Mon, 6 Mar 2023 08:05:55 +0000 (09:05 +0100)]
sthread: Add a way to verify accesses to non-reentrant data structures
The example is very fragile so far, segfaulting 2 times over 3 in
general, so don't activate it.
Martin Quinson [Mon, 6 Mar 2023 09:17:42 +0000 (10:17 +0100)]
Cosmetics in sthread output when not using SMPI
Arnaud Giersch [Mon, 6 Mar 2023 07:49:26 +0000 (08:49 +0100)]
Enable compile optimizations when tests will be run.
[ci-skip]
Arnaud Giersch [Sun, 5 Mar 2023 21:48:50 +0000 (22:48 +0100)]
Missing ']'
[ci-skip]
Arnaud Giersch [Sun, 5 Mar 2023 20:56:34 +0000 (21:56 +0100)]
Fix compilation error with clang 11.0.1.
Building CXX object CMakeFiles/simgrid.dir/src/mc/explo/udpor/maximal_subsets_iterator.cpp.o
In file included from .../src/mc/explo/udpor/maximal_subsets_iterator.cpp:1:
.../src/mc/explo/udpor/maximal_subsets_iterator.hpp:38:3: error: explicitly defaulted default constructor is implicitly deleted [-Werror,-Wdefaulted-function-deleted]
maximal_subsets_iterator() = default;
^
.../src/mc/explo/udpor/maximal_subsets_iterator.hpp:43:68: note: default constructor of 'maximal_subsets_iterator' is implicitly deleted because field 'config' of const-qualified type 'const std::optional<std::reference_wrapper<const Configuration>>' would not be initialized
const std::optional<std::reference_wrapper<const Configuration>> config;
^
1 error generated.
Arnaud Giersch [Sun, 5 Mar 2023 20:31:19 +0000 (21:31 +0100)]
Enable _GLIBCXX_DEBUG for one of the builds.
Also run the tests in this case.
Arnaud Giersch [Sun, 5 Mar 2023 16:35:16 +0000 (17:35 +0100)]
Sonar smells.
Martin Quinson [Sun, 5 Mar 2023 20:13:24 +0000 (20:13 +0000)]
Merge branch 'maximal-subset-search' into 'master'
Phase 3 of UDPOR Integration: Iteration Over Maximal Sets of a Configuration
See merge request simgrid/simgrid!135
Arnaud Giersch [Sun, 5 Mar 2023 14:32:32 +0000 (15:32 +0100)]
Don't clear flops_amount on ExecImpl::reset.
Fix failures where flops_amount is not defined anymore in ExecImpl::start
(found by building with CXXFLAGS=-D_GLIBCXX_DEBUG)
Arnaud Giersch [Sun, 5 Mar 2023 14:01:22 +0000 (15:01 +0100)]
Quick tests show that currentContainer is always empty at this point.
Arnaud Giersch [Sat, 4 Mar 2023 09:35:59 +0000 (10:35 +0100)]
Walking the stack in the forward direction gives the same result, but makes the code simpler.
Arnaud Giersch [Thu, 2 Mar 2023 21:14:38 +0000 (22:14 +0100)]
Simplify assertions.
Augustin Degomme [Sat, 4 Mar 2023 13:13:06 +0000 (14:13 +0100)]
try to use another mattermost action, as the main one is linux only (uses docker)
Augustin Degomme [Sat, 4 Mar 2023 12:10:51 +0000 (13:10 +0100)]
[ci-skpi] fix sending of messages, and do parallel testing on github action
Augustin Degomme [Sat, 4 Mar 2023 12:09:16 +0000 (13:09 +0100)]
assume mmap is not functional for us on macos>=11 as memory accesses are restricted by default
Augustin Degomme [Sat, 4 Mar 2023 10:51:11 +0000 (11:51 +0100)]
[ci skip] remove spurious line
Augustin Degomme [Sat, 4 Mar 2023 10:43:20 +0000 (11:43 +0100)]
add macos build to github action, untested
Augustin Degomme [Sat, 4 Mar 2023 10:42:56 +0000 (11:42 +0100)]
remove ns3 from ubuntu stable and github actions, as 3.35 is broken on 22.04
Martin Quinson [Fri, 3 Mar 2023 23:08:16 +0000 (00:08 +0100)]
Fix a useless change -- and go get some sleep
Martin Quinson [Fri, 3 Mar 2023 22:57:15 +0000 (23:57 +0100)]
Bump cmake to at least 3.12
- It's difficult to search for Python Development component before that version
- This version is available in Debian buster (oldstable)
Martin Quinson [Fri, 3 Mar 2023 22:50:05 +0000 (23:50 +0100)]
python3-dev is another dependency of the Python bindings
Martin Quinson [Fri, 3 Mar 2023 20:21:07 +0000 (21:21 +0100)]
Reindent -- almost no real change
Simply don't traverse an array if debug logs are disabled.
Arnaud Giersch [Fri, 3 Mar 2023 16:32:08 +0000 (17:32 +0100)]
Forgot to add c++17g to the "ignore" list.
[ci-ckip]
Martin Quinson [Fri, 3 Mar 2023 14:31:22 +0000 (15:31 +0100)]
fix MC + gcc builds
Martin Quinson [Fri, 3 Mar 2023 10:45:34 +0000 (11:45 +0100)]
Move the implem of a function out of the header to reduce issues
Martin Quinson [Thu, 2 Mar 2023 23:30:29 +0000 (00:30 +0100)]
Move the PageStore from ModelChecker to RemoteApp
Maxwell Pirtle [Fri, 3 Mar 2023 10:06:54 +0000 (11:06 +0100)]
Correct MANIFEST.in according to `make distcheck`
Maxwell Pirtle [Fri, 3 Mar 2023 08:00:25 +0000 (09:00 +0100)]
Filter events before performing iteration
A filter can be supplied to the maximal subset
iterator to remove events from consideration in
maximal subsets. Instead of needlessly processing
events that will never be considered, we can
instead remove those events at the beginning while
perserving the relative ordering among events that
we do actually care about. This both simplifies the
bookkeeper's code slightly and reduces the number
of times we need to invoke the filter function drastically
(once per event instead of multiple times per combination).
Martin Quinson [Thu, 2 Mar 2023 22:59:14 +0000 (23:59 +0100)]
Move the stats of visited_states count from ModelChecker to DFSExplorer where it belongs
Martin Quinson [Thu, 2 Mar 2023 22:48:01 +0000 (23:48 +0100)]
Move the dot_output from ModelChecker to Exploration
Martin Quinson [Thu, 2 Mar 2023 22:31:04 +0000 (23:31 +0100)]
One reference to the global singleton mc_model_checker that I'd like to kill
Martin Quinson [Thu, 2 Mar 2023 22:22:13 +0000 (23:22 +0100)]
Rename some fields to obey our coding standard
Martin Quinson [Thu, 2 Mar 2023 20:53:31 +0000 (21:53 +0100)]
Allow to call SMPI_app_instance_start() once the simulation started
The main trick was to set the needed property before the
Actor::start() using init()->set_property()->start(), so that the
property is here soon enough.
Arnaud Giersch [Thu, 2 Mar 2023 17:27:15 +0000 (18:27 +0100)]
Zero-initialize more messages.
Maxwell Pirtle [Thu, 2 Mar 2023 15:24:22 +0000 (16:24 +0100)]
Fix subtle implementation bug with maximal set filtering
There was a subtle bug that was resolved involving
traversal over the maximal event set tree of a configuration.
Specifically, it's important to
1. Only pick candidates that satisfy the predicate. This
means that the `bookkeeper` needs to own the reference
to the filter so that is can accurately pick events
correctly
2. At the start, pick the first *viable* candidate, and not simply
the *first* node in the ordering, as the first node may
not satisfy the predicate. Effectively, you pick the
"most recent" node that satisfies the predicate
A possible future addition would be to pre-compute the
predicate for all of the events at the start of the
iterator. This would reduce the number of calls made
to the filter function dramatically, but for now we'll
stick with something simple (the change wouldn't involve
too much intervention if desired)
Arnaud Giersch [Thu, 2 Mar 2023 14:53:35 +0000 (15:53 +0100)]
Message is retrieved as an unsigned long (according to the manual).
Maxwell Pirtle [Thu, 2 Mar 2023 13:38:47 +0000 (14:38 +0100)]
Add working implementation of maximal_subsets_iterator
There were a few subtleties in computing traversing
the graph of maximal subsets iteratively. The most
important was the distinction between the case of
expanding the current maximal event set and attempting
to find the newest branch point (i.e. walking back up
the tree after performing the depth-first order).
Martin Quinson [Thu, 2 Mar 2023 10:51:50 +0000 (11:51 +0100)]
Fix a typo found by clang in MC builds
Maxwell Pirtle [Thu, 2 Mar 2023 10:41:00 +0000 (11:41 +0100)]
Clean up maximum_subsets_iterator
A couple of smaller cosmetic changes were made
to maximum_subsets_iterator, in particular moving
the somewhat complicated while-loop into a dedicated
method (before it had several `break`s contained within
it which acted more like return values from a function
which suggested a dedicated function was called for).
Subsequent commits will add test cases for the iterator,
after which a MR should be ready.
Arnaud Giersch [Wed, 1 Mar 2023 21:00:05 +0000 (22:00 +0100)]
Replace memset(..., 0, ...) with zero-initialization.
Arnaud Giersch [Wed, 1 Mar 2023 20:55:31 +0000 (21:55 +0100)]
Zero-initialize messages (essentially their padding bits).
This is mainly to clear valgrind errors about uninitialised bytes.
Arnaud Giersch [Wed, 1 Mar 2023 10:04:24 +0000 (11:04 +0100)]
Help to identify null values.
Arnaud Giersch [Wed, 1 Mar 2023 10:03:20 +0000 (11:03 +0100)]
Lookup only once in map.
Arnaud Giersch [Tue, 28 Feb 2023 15:01:44 +0000 (16:01 +0100)]
Useless assignments to local variable.
Martin Quinson [Thu, 2 Mar 2023 10:04:35 +0000 (11:04 +0100)]
GH action: build out of sources
Martin Quinson [Thu, 2 Mar 2023 09:13:12 +0000 (10:13 +0100)]
Install python deps on GHaction + better documentation of build dependencies
Maxwell Pirtle [Thu, 2 Mar 2023 09:00:42 +0000 (10:00 +0100)]
Pass references to `const Unfolding*` in most places
Passing references to `const Unfolding*` is better
style as no methods actually modify events directly:
instead, they perform computations over sets of events,
compute histories, etc, but leave the event instances
themselves otherwise unaffected
Martin Quinson [Thu, 2 Mar 2023 08:55:34 +0000 (09:55 +0100)]
Mais quel débile, punaise
Martin Quinson [Thu, 2 Mar 2023 08:48:45 +0000 (09:48 +0100)]
MC builds have specific dependencies
Maxwell Pirtle [Thu, 2 Mar 2023 08:28:34 +0000 (09:28 +0100)]
Add first implementation of maximal_subsets_iterator
The maximal_subsets_iterator traverses the subsets
of events of a configuration `C` which are maximal,
viz. those subsets of `C` which are causally-free.
Martin Quinson [Thu, 2 Mar 2023 08:31:52 +0000 (09:31 +0100)]
GH action: ninja is verbose on error anyway
Martin Quinson [Thu, 2 Mar 2023 08:27:40 +0000 (09:27 +0100)]
GH action: use the correct package name
Martin Quinson [Thu, 2 Mar 2023 08:20:40 +0000 (09:20 +0100)]
gh action: we need 'sudo' here
Martin Quinson [Thu, 2 Mar 2023 08:16:13 +0000 (09:16 +0100)]
GH action: install deps
Martin Quinson [Thu, 2 Mar 2023 08:04:52 +0000 (09:04 +0100)]
GH action: use existing runners
Martin Quinson [Thu, 2 Mar 2023 07:44:09 +0000 (08:44 +0100)]
messing with git actions
Martin Quinson [Thu, 2 Mar 2023 07:42:02 +0000 (08:42 +0100)]
Fix yaml syntax :(
Martin Quinson [Thu, 2 Mar 2023 07:37:31 +0000 (08:37 +0100)]
Add a github action to test our git on Debian11, regular and MC
This is triggered manually, when jenkins is off as right now
Henri Casanova [Thu, 2 Mar 2023 02:38:48 +0000 (16:38 -1000)]
Pass on documentation's Design Goals section
Henri Casanova [Wed, 1 Mar 2023 23:38:37 +0000 (13:38 -1000)]
More work on the documentation's Models section
Martin Quinson [Wed, 1 Mar 2023 20:44:56 +0000 (21:44 +0100)]
New function SMPI_app_instance_start() to easily start a MPI instance in your S4U simulation
Martin Quinson [Wed, 1 Mar 2023 19:56:56 +0000 (20:56 +0100)]
Try to fix a fixme blindly, just in case it was easy
Henri Casanova [Wed, 1 Mar 2023 19:34:00 +0000 (09:34 -1000)]
Pass on the documentation's Introduction page. Minor grammatical fixes,
some rewrites and word changes, minor content removal due to redundancy.
Maxwell Pirtle [Wed, 1 Mar 2023 08:22:43 +0000 (09:22 +0100)]
Use boost::iterator_facade for History::Iterator
The iterator that is used in the `History` class
was adjusted to use boost::iterator_facade, as it
simplifies life considerably, makes the code more
readable, and provides better conformance to being
a forward iterator (i.e. it probably handles some
subtleties with forward iterators etc. and implements
many methods automatically)
Martin Quinson [Tue, 28 Feb 2023 18:12:26 +0000 (19:12 +0100)]
Fix make distcheck
Augustin Degomme [Tue, 28 Feb 2023 15:59:29 +0000 (15:59 +0000)]
Merge branch 'udpor-phase3' into 'master'
Phase 2.5 of UDPOR Integration: Add Iterators for Subsets
See merge request simgrid/simgrid!134
Maxwell Pirtle [Tue, 28 Feb 2023 15:36:36 +0000 (16:36 +0100)]
Add src/xbt/.../subsets_tests.cpp to manifest file
Maxwell Pirtle [Tue, 28 Feb 2023 14:52:20 +0000 (15:52 +0100)]
Remove CompatibilityGraph and friends
The CompatibilityGraph and friends that were
added in order to compute maximal sets of events
needlessly complicated the algorithm. A much
better approach was suggested which will be
implemented in the next phase.
While perhaps a rebase could also have been
performed, it would have required quite careful
precision. Since the end goal is the same and
there are better uses of our time, I simply opted
to tag on another commit
Maxwell Pirtle [Tue, 28 Feb 2023 14:31:54 +0000 (15:31 +0100)]
Add tests for LazyKSubsets and LazyPowerset
The LazyKSubsets and LazyPowerset are simply
wrappers around an iterator type which can
enumerate all subsets of a given iteration.
While this won't be used anymore to enumerate
all possible sets of events of the configuration
in the compatibility graph, which will soon be
removed, it will still be used (or a variant
thereof) when computing K-partial alternatives
later on
Maxwell Pirtle [Tue, 28 Feb 2023 10:30:28 +0000 (11:30 +0100)]
Add LazyPowerSet and LazyKSubsets
Two new classes were added that act effectively
as convenience methods around creating subset
iterators. The classes themselves act mostly
like iterators but are instead light-weight
"containers" which can be iterated over. Effectively,
they are contains which "contain" the entire
collection of subsets (of size k respectively) of
some collection without actually storing that collection
in memory.
Maxwell Pirtle [Tue, 28 Feb 2023 10:14:43 +0000 (11:14 +0100)]
Add more comments to subsets_iterator implementation