Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
simgrid.git
11 months agoAdd odpor_tests_private.hpp to MANIFEST.in
Maxwell Pirtle [Thu, 1 Jun 2023 07:16:30 +0000 (09:16 +0200)]
Add odpor_tests_private.hpp to MANIFEST.in

11 months agoAdd SDPOR backtracking simulation unit test
Maxwell Pirtle [Wed, 31 May 2023 13:18:31 +0000 (15:18 +0200)]
Add SDPOR backtracking simulation unit test

This commit adds a small unit test which
simulates the process of determining which
threads need to be added into the backtrack
set of a given state using SDPOR

11 months agoAdd more independence tests w.r.t a partial execution
Maxwell Pirtle [Wed, 31 May 2023 12:14:12 +0000 (14:14 +0200)]
Add more independence tests w.r.t a partial execution

11 months agoFix MANIFEST.in etc.
Maxwell Pirtle [Wed, 31 May 2023 07:20:55 +0000 (09:20 +0200)]
Fix MANIFEST.in etc.

11 months agoForce the use of sleep sets with ODPOR
Maxwell Pirtle [Tue, 30 May 2023 14:11:45 +0000 (16:11 +0200)]
Force the use of sleep sets with ODPOR

11 months agoAdjust tesh tests according to changes in deps.
Maxwell Pirtle [Tue, 30 May 2023 13:49:26 +0000 (15:49 +0200)]
Adjust tesh tests according to changes in deps.

A number of tests were affected by the previous
additions that labeled transitions executed by
the same actor as dependent. The appropriate
modifications were made to the expected outputs
and a "sanity" check was performed before making
them

11 months agoMark transitions run by the same actor as dependent
Maxwell Pirtle [Tue, 30 May 2023 11:46:37 +0000 (13:46 +0200)]
Mark transitions run by the same actor as dependent

This commit adds a missing component to a number of
transition classes. Transitions that are executed by
the same actor are inherently dependent; but only the
RandomTransition actually checked for this case! This
caused ODPOR to search traces incorrectly under the
assumption that transitions which were actually
dependent were independent

11 months agoFix subtle bug in ~_E computation
Maxwell Pirtle [Tue, 30 May 2023 11:42:32 +0000 (13:42 +0200)]
Fix subtle bug in ~_E computation

The computation for the action ~_E requires
that we compute whether `p` is independent with
some sequence of transitions `w` after an
execution `E`. The algorithm for computing
whether an equivalent trace was found works
to iteratively eliminate events from the
equivalent sequence, leaving behind only
the remaining "bits" that are needed for
insertion into a wakeup tree.

The bug involved a typo of `w` for `w_now`
:( :( which broke everything :/

11 months agoAdd remaining handlers to ReversibleRaceCalculator
Maxwell Pirtle [Fri, 26 May 2023 13:18:04 +0000 (15:18 +0200)]
Add remaining handlers to ReversibleRaceCalculator

This commit adds handlers for each type of transition
to ReversibleRaceCalculator which, as the name suggests,
determines whether a reversible race exists between two
events e1 and e2 based on the type of transition found
at e2. The full semantics have not been added for each
type of transition and will need to be added in the future

11 months agoResolve misconception with ODPOR pseudocode impl.
Maxwell Pirtle [Fri, 26 May 2023 13:12:42 +0000 (15:12 +0200)]
Resolve misconception with ODPOR pseudocode impl.

Like the SDPOR implementation before it, the ODPOR
implementation suffered from a similar fault in its
implementation. Determining whether an insertion was
necessary into a wakeup tree for some point in the
past requires determining whether the sleep set and
the set of weak initials overlap one another. If
their intersection is the empty set, THEN we attempt
to insert an actor. The prior implementation attempted
to find the first such actor that was not contained
in the sleep set, but this is semantically different

11 months agoResolve misconception with SDPOR pseudocode impl.
Maxwell Pirtle [Fri, 26 May 2023 12:45:09 +0000 (14:45 +0200)]
Resolve misconception with SDPOR pseudocode impl.

This commits fixes a semantics mismatch between
the implementation and the pseudocode in SDPOR.
The trouble was that we didn't stop searching
for an initial when we noticed that ANY initial
was contained in the backtrack set; instead, we
simply searched until we found ANY initial contained
OUTSIDE OF the backtrack set. The latter is not
equivalent to the former, so SDPOR was doing
more work than necessary.

A similar implementation issue is also present in
ODPOR which will be resolved in due time.

11 months agoFix subtle bug in Execution regeneration in DFSExplorer
Maxwell Pirtle [Fri, 26 May 2023 08:07:57 +0000 (10:07 +0200)]
Fix subtle bug in Execution regeneration in DFSExplorer

Prior to this commit, during state regeneration,
we had used the state's `get_transition_out()`
method to refill the `Execution` instance bound
to DFSExplorer (`execution_seq_`). The trouble
is that this is an **unstable** source of truth:
it should not be trusted to regenerate an execution.
Instead, the `get_transition_in()` method should
be used on the state which follows. Intuitively,
both methods should be equivalent, which is why
the bug was overlooked for so long. Now that the
execution is correct and that SDPOR/ODPOR are
based on the correct sources of truth, things
are moving along more nicely and the two algorithms
appear to behave more sanely.

In addition to the above bug fix, this commit adds
a return value to the `insert()` function of the
`WakeupTree` class. This extra information is used
in debug logs to get a better idea of what ODPOR
is doing during its insertion. Similar debug information
will be added to SDPOR in subsequent phases.

11 months agoKeep pointers to transitions instead of slices
Maxwell Pirtle [Thu, 25 May 2023 09:18:31 +0000 (11:18 +0200)]
Keep pointers to transitions instead of slices

Prior to this commit, sleep sets contained slices

11 months agoFix two minor bugs in the ODPOR implementation
Maxwell Pirtle [Thu, 25 May 2023 08:25:07 +0000 (10:25 +0200)]
Fix two minor bugs in the ODPOR implementation

This commit fixes the following two bugs:

1. When seeding a wakeup tree for the first
time, we forgot to add a `break` statement
to ensure that only one such enabled thread
were placed into the wakeup tree. Before
all such enabled threads were inserted which
amounted to what would have been a brute-force
search :(.

2. When determining whether a sequence is a
candidate for insertion into a wakeup tree,
we may need to check, for all enabled actors
in a given state, whether those actors'
next steps are independent with the sequence
theretofore constructed. This additional
step peforms the work needed to check whether
those actors are contained in the set of
*weak* initials. The method corresponds
to line 6 of the pseudocode of the ODPOR algorithm

11 months agoAdd reversible race implementations for Comm actions
Maxwell Pirtle [Wed, 24 May 2023 12:04:32 +0000 (14:04 +0200)]
Add reversible race implementations for Comm actions

11 months agoAdd reversible race calculator
Maxwell Pirtle [Wed, 24 May 2023 11:09:05 +0000 (13:09 +0200)]
Add reversible race calculator

Determining whether or not a race between
two events is reversible is a critical component
of ODPOR. It turns out that we need to specialize
the race detection to the action types themselves.
Knowing the types of the transitions will help us
determine whether the race is reversible.

11 months agoAdd non-trivial insertion test
Maxwell Pirtle [Wed, 24 May 2023 09:32:20 +0000 (11:32 +0200)]
Add non-trivial insertion test

11 months agoAdd tests for initials and independence for Execution
Maxwell Pirtle [Tue, 23 May 2023 12:24:13 +0000 (14:24 +0200)]
Add tests for initials and independence for Execution

11 months agoAdd first round of execution independence tests
Maxwell Pirtle [Tue, 23 May 2023 08:18:24 +0000 (10:18 +0200)]
Add first round of execution independence tests

This commit adds the first set of tests for
checking whether a particular transition `t` is
independent with an extension `w` of an execution
sequence `E`. This is needed to determine whether
the transition `t` satisfies condition (2) of the
procedure outlined in the ODPOR paper for testing
whether `v ~_[E] w` holds between two sequences

11 months agoAdd test for removing the first single-process subtree
Maxwell Pirtle [Mon, 22 May 2023 11:56:30 +0000 (13:56 +0200)]
Add test for removing the first single-process subtree

This commit adds the first test checking whether
removing the first single-process node from the
tree produces the correct results, viz. that the
resulting tree is the same as before save for the
missing branch

11 months agoBegin adding tests for subtree rooting
Maxwell Pirtle [Mon, 22 May 2023 11:20:02 +0000 (13:20 +0200)]
Begin adding tests for subtree rooting

11 months agoAdd detailed stress test for WakeupTree
Maxwell Pirtle [Mon, 22 May 2023 09:03:48 +0000 (11:03 +0200)]
Add detailed stress test for WakeupTree

This commit introduces an edge case wherein
a "different looking sequence" ultimately is
already accounted for by a sequence contained
in the tree that "eventually could be made to
look like it" (viz. such that `~_E` holds
between the two sequences). Other stress tests
will be added for the other methods on the
WakeupTree and the Execution to test their
correctness. There is a long way to go

11 months agoBegin first round of in-depth tests for WakeupTree
Maxwell Pirtle [Mon, 22 May 2023 08:52:54 +0000 (10:52 +0200)]
Begin first round of in-depth tests for WakeupTree

11 months agoAdd first "working" version of ODPOR
Maxwell Pirtle [Wed, 17 May 2023 11:46:20 +0000 (13:46 +0200)]
Add first "working" version of ODPOR

This commit introduces the first version of
ODPOR which detects a bug using the example
programs found in SimGrid's test suite! There
is still work to be done (e.g. dealing with
transitions that have multiple possible
paths of execution such as MCRandom() will
be non-trivial to resolve).

The key adjustments in this commit are
to the WakeupTree structure itself. Gone
are the days when each node in the tree
maintained a sequence of transitions to
execute; instead, each node contains only
a single transition, and its the paths between
a node and the root that describe the partial
executions that ODPOR needs to explore.
This resulted from a misconception I had about
the structure of wakeup trees. When an insertion
is made, potentially a *chain* of leaf nodes
are added into the tree. This ensures we always
have a single-process branch to pick (which
is what we expected) and greatly improves the
readability of some of the methods found in
State.cpp related to ODPOR

11 months agoAdd tests before changes to WakeupTree structure
Maxwell Pirtle [Tue, 16 May 2023 13:07:31 +0000 (15:07 +0200)]
Add tests before changes to WakeupTree structure

11 months agoAdd first round of extensive docs to ODPOR methods
Maxwell Pirtle [Tue, 16 May 2023 08:55:29 +0000 (10:55 +0200)]
Add first round of extensive docs to ODPOR methods

11 months agoAdd test files for WakeupTree
Maxwell Pirtle [Tue, 16 May 2023 06:48:13 +0000 (08:48 +0200)]
Add test files for WakeupTree

11 months agoAdd explicit ODPOR clean-up phase to DFSExplorer
Maxwell Pirtle [Mon, 15 May 2023 13:31:31 +0000 (15:31 +0200)]
Add explicit ODPOR clean-up phase to DFSExplorer

ODPOR requires that we remove elements from the
wakeup tree and add elements to the sleep sets
at state in the execution only AFTER we've
completed fully the recursive exploration of the
space pointed towards by the wakeup trees; i.e.,
we only remove subtrees and add to the sleep sets
when we're walking back UP the state stack looking
for a point to recontinue the execution.

11 months agoAdd ODPOR "backtracking" logic
Maxwell Pirtle [Mon, 15 May 2023 08:50:25 +0000 (10:50 +0200)]
Add ODPOR "backtracking" logic

11 months agoAdd ODPOR extension computation (lines 4-6)
Maxwell Pirtle [Fri, 12 May 2023 11:19:50 +0000 (13:19 +0200)]
Add ODPOR extension computation (lines 4-6)

ODPOR asks us to compute, for each reversible
race in the current maximal execution `E`,
whether some subsequence `v` of events
extending a prefix `E'` of `E` should be
inserted into a wakeup tree.

There's a few subtle details that should
be noted in this computation. ODPOR asks
us to compute `notdep(e, E)`, which means
possibly looking at events that occur
*after* `e'` and THEN adding `proc(e')`.
The subtlety lies in the fact that the
actual transition associated with `e'`
must be added AFTER all of the others.
We got lucky with SDPOR since next_E_p
was already in the last position and
thus required no special treatment. With
ODPOR, we have to be more careful.

Even more subtle is the observation that
any events in `notdep(e, E)` can neither
affect the enabledness of `e'` nor can
`e'` affect the enabledness of any of
those events, or else if they do
affect the enabledness of `e'` they
will necessarily be contained "between"
`e` and `e'`

11 months agoUse `std::shared_ptr<Transition>` for Execution
Maxwell Pirtle [Fri, 12 May 2023 08:11:33 +0000 (10:11 +0200)]
Use `std::shared_ptr<Transition>` for Execution

Prior to this commit, the `Execution` class
maintained raw pointers to the transitions
maintained by the stack used in DFSExplorer.
Instead, we opt for the execution to contain
a collection of shared pointers to facilitate
the creation of partial executions

11 months agoAdd ODPOR race detection phase rough-draft
Maxwell Pirtle [Fri, 12 May 2023 07:55:10 +0000 (09:55 +0200)]
Add ODPOR race detection phase rough-draft

ODPOR is decoposed into two parts: the "passive"
observer phase, where ODPOR selects search points
based on the work it did before to fill in its
wakeup trees, and the active race-detection phase,
where ODPOR looks at all reversible races in an
execution and performs the work to determine if
new traces need be searched. This commit adds
in an outline for the latter phase. Note that
there is still much work to be done with respect
to the race detection -- we still need to implement
the partial execution function for example.

Likewise, handling the issue of multiple execution
possibilities for a given process will be particularly
tricky.

11 months agoAdd boolean check for wakeup tree initialization
Maxwell Pirtle [Thu, 11 May 2023 13:36:28 +0000 (15:36 +0200)]
Add boolean check for wakeup tree initialization

While not great, one way to prevent the wakeup
tree from continuing to add elements after it
has already done so when empty is to add a
boolean flag that's checked to prevent the
addition. For now, we'll live with it

11 months agoAutomatically remove nodes from parents
Maxwell Pirtle [Thu, 11 May 2023 12:18:56 +0000 (14:18 +0200)]
Automatically remove nodes from parents

After removing a node from a WakeupTree,
we need to be careful to also have any references
to the node removed, viz. that held onto the
parent of the node in the list of children.
This commit adds a back reference to the
parent node of a given WakeupTreeNode to
allow removal from the parent when destroyed.

Additionally, some bugs were fixed for creating
subtrees and removing subtrees (we forgot to remove
the root for example)

11 months agoAdd tree pruning/subtree methods to State
Maxwell Pirtle [Thu, 11 May 2023 10:38:46 +0000 (12:38 +0200)]
Add tree pruning/subtree methods to State

ODPOR requires that we can remove a
subtree from a wakeup tree as we continue
the exploration. Further, we must also be
able to create a subtree from a given
wakeup tree. The functionality for those
methods has already been added; this commit
instead introduces the management of wakeup
trees specifically in the context of ODPOR
exploration.

11 months agoAdd skeleton for expansion phase of ODPOR
Maxwell Pirtle [Thu, 11 May 2023 09:04:23 +0000 (11:04 +0200)]
Add skeleton for expansion phase of ODPOR

11 months agoAdd logic for subtree node removal
Maxwell Pirtle [Wed, 10 May 2023 13:55:18 +0000 (15:55 +0200)]
Add logic for subtree node removal

ODPOR also mandates that subtrees be removed
from a wakeup tree after having finished searching
"in the direction" of the subtree. This commit
adds a method to WakeupTree which, given a handle
to a node in the tree, removes the subtree rooted at
that node.

11 months agoFinish post-order travesal with WakeupTreeIterator
Maxwell Pirtle [Wed, 10 May 2023 13:03:08 +0000 (15:03 +0200)]
Finish post-order travesal with WakeupTreeIterator

The WakeupTreeIterator is now full implemented:
we now properly add nodes to the top of the
stack in reverse order, resulting in a post-order
traversal as desired.

One subtlety arose with the implementation: the
root node is not contained in the list of any
other node in the tree. To handle this issue,
a "fake" list managed by the iterator was added
into which the root is placed at construction-time.
This prevents us from needing to change any of the
logic, and we can simply treat the root as any other
node in the traversal

11 months agoAdd logic for subtree extraction from wakeup trees
Maxwell Pirtle [Wed, 10 May 2023 12:34:02 +0000 (14:34 +0200)]
Add logic for subtree extraction from wakeup trees

Extracting a subtree from a wakeup tree is required
for ODPOR to continue its search and to pass information
down to the other paths that ODPOR decides to visit.
Extracting a copy of a subtree is difficult as it
requires a deep copy of the tree, making sure
that references to any nodes in the tree that is copied
are copied appropriately. One additional subtlety is that
the subtree can only be taken with respect to single-process
branches and, further, that that process is removed from
each descendant of the node that contains the single process.

11 months agoAdd complicated computation of v ~_E w to Execution
Maxwell Pirtle [Wed, 10 May 2023 09:38:07 +0000 (11:38 +0200)]
Add complicated computation of v ~_E w to Execution

The relation `v ~_E w` is used when traversing
a WakeupTree to determine is an "equivalent"
execution is already noted in the tree. This commit
adds the first "working" implementation of
several components to this process, viz.

1. Assuming that iteration has been implemented
(which is to come in future commits), the WakeupTree
now implements the `insert_E(v, B)` operation, where
`B` refers to the tree instance itself.

2. The `Execution` class can now determine what
is needed to complete the `insert()` operation,
specifically determining first if a) two sequences
`v` and `w` are related by `v ~_E w` and b)
what the smallest sequence `w'` is such that
`w [= v.w'`. The latter computation uses the neat
observation that (w / v) is the smallest such
sequence (viz. removing anything in `w` that's in
`v` "one at a time") and computes this value as
it is in the process of determining whether the
`~_E` relation noted holds

11 months agoAdd skeleton of implementation for tree insertion
Maxwell Pirtle [Wed, 10 May 2023 07:08:27 +0000 (09:08 +0200)]
Add skeleton of implementation for tree insertion

The `insert` method on a Wakeup tree is an
essential component of the ODPOR algorithm.
Tree insertion guides the ODPOR algorithm
towards the unique, unsearched traces of the
state space. Insertion is based heavily on the
correctness of computing the `~` relation over
a given execution. Subsequent commits will fill
in this method as well as several others that
will be needed to implement that functionality

11 months agoAdd WakeupTreeIterator and WakeupTree skeletons
Maxwell Pirtle [Tue, 9 May 2023 14:34:04 +0000 (16:34 +0200)]
Add WakeupTreeIterator and WakeupTree skeletons

11 months agoAdd more docmentation for get_first_sdpor_initial()
Maxwell Pirtle [Tue, 16 May 2023 07:39:15 +0000 (09:39 +0200)]
Add more docmentation for get_first_sdpor_initial()

A very important function at the center of SDPOR
was missing significant context that would help
a reader understand what exactly was going on
with the function. What makes the situation tricky
is the fact that the pseudocode's implementation
is blended into several lines and computed "on the
fly" rather than fully.

11 months agoAdd more documentation to ClockVector/Execution
Maxwell Pirtle [Mon, 15 May 2023 07:37:22 +0000 (09:37 +0200)]
Add more documentation to ClockVector/Execution

11 months agoAdd unit tests for ClockVector
Maxwell Pirtle [Mon, 15 May 2023 07:04:11 +0000 (09:04 +0200)]
Add unit tests for ClockVector

11 months agoFix MANIFEST.in
Maxwell Pirtle [Tue, 9 May 2023 11:57:07 +0000 (13:57 +0200)]
Fix MANIFEST.in

We should note that the Execution that's a part of
both ODPOR and SDPOR is now contained in DFSExplorer
even in stateless execution. This will likely eventually
be true also for ODPOR; hence, the ODPOR sources
have been added to the SRC_MC_STATELESS portion as
needed

11 months agoIntegrate SDPOR into `model-check/reduction` flag
Maxwell Pirtle [Tue, 9 May 2023 08:36:13 +0000 (10:36 +0200)]
Integrate SDPOR into `model-check/reduction` flag

SDPOR and ODPOR can be used as part of the
DFSExplorer (ODPOR should integrate with some
minor details in the DFS implementation, but that's
for a later PR). This commit replaces the simple
`with_dpor` boolean flag used to control the
reduction with the more precise `ReductionMode`
that is newly extracted from within `DFSExplorer`
and is instead made "public"

11 months agoAdd more documentation to essential SDPOR methods
Maxwell Pirtle [Fri, 5 May 2023 12:47:07 +0000 (14:47 +0200)]
Add more documentation to essential SDPOR methods

11 months agoModify tesh suite expectations with changes to dependencies
Maxwell Pirtle [Tue, 9 May 2023 07:47:27 +0000 (09:47 +0200)]
Modify tesh suite expectations with changes to dependencies

11 months agoFix ObjectAccess dependency check
Maxwell Pirtle [Fri, 5 May 2023 12:21:17 +0000 (14:21 +0200)]
Fix ObjectAccess dependency check

The ObjectAccess transition should defer
dependency computation to the ActorJoin action.
Previously, it only decided dependency by
simply checking if the `other` transition
were simply another object access on the
same object

11 months agoFix subtlety with SDPOR/ODPOR initials
Maxwell Pirtle [Fri, 5 May 2023 09:16:07 +0000 (11:16 +0200)]
Fix subtlety with SDPOR/ODPOR initials

Computing initials for SDPOR/ODPOR requires
extending a search with respect to
a prefix of an execution. Previously, events
added to the extension were referred to by
their location within the larger execution
from which the prefix was taken. The problem
is that the new events need to be referred
to by their position relative to the
*extension* of the prefix; otherwise,
asking about "happens-before" in the extension
while using their handles in the full execution
will give incorrect results

11 months agoFix two off-by-one errors with clock vectors
Maxwell Pirtle [Fri, 5 May 2023 08:29:02 +0000 (10:29 +0200)]
Fix two off-by-one errors with clock vectors

Two errors were identified with managing clock
vectors in the Execution class. The index
assigned to the new transition was previously
`size() + 1` when it should have read `size()`.
Furthermore, the loop for checking "happens-between"
events excluded event 0. Both issues are now resolved
and a test with a race at the beginning was added
as verification

11 months agoMark `ActorJoin` dependent with `target_`
Maxwell Pirtle [Fri, 5 May 2023 06:53:47 +0000 (08:53 +0200)]
Mark `ActorJoin` dependent with `target_`

The `ActorJoin` transition was previously
labeled as independent with all other
transitions, including those of the actor
upon which the join acted. However, this
assumption was faulty: the join action is
instead dependent with ANY transition executed
by the actor joined upon, as the action
BECOMES enabled only as soon as the last action
run by the actor is executed.

11 months agoPrevent adding outgoing transition for top-most state
Maxwell Pirtle [Fri, 5 May 2023 06:46:09 +0000 (08:46 +0200)]
Prevent adding outgoing transition for top-most state

The outgoing transition for the top-most
state of the state stack in the DFSExplorer refers
to that which was taken as part of the last
trace explored by the algorithm. Thus, only the
sequence of transitions leading up to, but
not including, the last state must be included
when reconstructing the Exploration for SDPOR.

11 months agoAdd workaround for subtlety with state regeneration
Maxwell Pirtle [Thu, 4 May 2023 12:29:44 +0000 (14:29 +0200)]
Add workaround for subtlety with state regeneration

Regeneration of the Execution inside DFSExplorer
should be as simple as choosing the prefix relative
to the appropriate backtracking point. However,
it did not appear to function so simple. After
a good amount of debugging, it appears that
the stack contents can change unexpectedly so
(e.g. a transition "in the middle" of the stack
seemingly switches arbitrarily). Until we can
pinpoint the true cause here, we simple resort
to rebuilding the execution based off the new
stack each time we decide to backtrack. The
downside is that all clock vectors have to be
recomputed after each backtrack. For now,
this is something that's OK to live with,
especially considering that this certainly
won't be a bottle neck in performance

11 months agoAdd first tests for "happens-before"
Maxwell Pirtle [Thu, 4 May 2023 07:33:25 +0000 (09:33 +0200)]
Add first tests for "happens-before"

The happens-before relation is used
extensively by SDPOR and ODPOR during
their computations for determining what
executions should be searched next. This
commit introduces the first round of tests
for the happens-before relation on
artificially constructed executions

11 months agoMove SDPOR core computation into a method
Maxwell Pirtle [Wed, 3 May 2023 09:35:01 +0000 (11:35 +0200)]
Move SDPOR core computation into a method

The core component of SDPOR involves selecting
an "initial" for a hypothetical computation
extending from a prefix of the current execution
that results in the reversal of the reversible
race SDPOR detected. This comomit moves that
computation as much as possible into the
`Execution` class itself to make the SDPOR
implementation easier to read

11 months agoAdd tentatively-working SDPOR implementation
Maxwell Pirtle [Wed, 3 May 2023 07:34:12 +0000 (09:34 +0200)]
Add tentatively-working SDPOR implementation

This commit introduces SDPOR into SimGrid
(well, mostly). The current implementation
will be improved upon by cleaning up some
of the code that was added to improve
its testability

11 months agoAdd initial outline of SDPOR implementation
Maxwell Pirtle [Tue, 2 May 2023 10:38:43 +0000 (12:38 +0200)]
Add initial outline of SDPOR implementation

The SDPOR algorithm is nearing an implementation
into SimGrid (sans a few technical details and
code clean ups that will be nice to add). The
algorithm largely builds off the existing
infrastructure of DFSExplorer fortunately.

11 months agoAdd Execution concept without ExecutionViews
Maxwell Pirtle [Tue, 2 May 2023 06:58:21 +0000 (08:58 +0200)]
Add Execution concept without ExecutionViews

The Execution is the key concept in ODPOR
and SDPOR for generating new traces. Executions
for the basis off of which new search paths are
created

11 months agoRemove sdpor folder in favor of odpor only
Maxwell Pirtle [Thu, 27 Apr 2023 08:43:37 +0000 (10:43 +0200)]
Remove sdpor folder in favor of odpor only

11 months agoAdd `Execution` to represent series of transitions
Maxwell Pirtle [Wed, 26 Apr 2023 13:57:28 +0000 (15:57 +0200)]
Add `Execution` to represent series of transitions

The Execution class represents a series of transitions
taken by a process. An execution importantly also encodes
the happens-before relation among the series of steps
represented. The eventual implementation will leverage
clock vectors to compute the relation between events.
A more refined happens-before relation is left for
as a future (complicated) exercise...

11 months agoAdd class for eventual "happens-before" computations
Maxwell Pirtle [Mon, 24 Apr 2023 12:55:12 +0000 (14:55 +0200)]
Add class for eventual "happens-before" computations

Computing the "happens-before" relation over
a sequence of executions is an important step
in determining whether or not DPOR (and SDPOR
alike) needs to add a backtracking point.
This commit introduces the "ClockVector"
class which acts effectively as a light
wrapping around an `unordered_map` mapping
actor ids to integer values.

The essential component of the ClockVector
class is that its contents are "implicit"
in the sense that actors for which no
value is explicitly contained in the map
are assigned to the value `0` by default.
This allows clock vectors to be flexible
enough to support the creation of new
actors and the enabling/disabling of old
actors as they come and go.

11 months agoadd two more utility functions to the Operation plugin
Fred Suter [Thu, 11 May 2023 14:29:20 +0000 (10:29 -0400)]
add two more utility functions to the Operation plugin

11 months agoFix make distcheck
Martin Quinson [Wed, 10 May 2023 21:35:25 +0000 (23:35 +0200)]
Fix make distcheck

11 months agoMC: Kill the now useless code State::get_recipe
Martin Quinson [Wed, 10 May 2023 21:27:34 +0000 (23:27 +0200)]
MC: Kill the now useless code State::get_recipe

11 months agoMerge branch 'master' of framagit.org:simgrid/simgrid
Martin Quinson [Wed, 10 May 2023 21:22:28 +0000 (23:22 +0200)]
Merge branch 'master' of framagit.org:simgrid/simgrid

11 months agoDFS MC: Restore from system states even if we have to run some transitions out of it
Martin Quinson [Wed, 10 May 2023 21:21:10 +0000 (23:21 +0200)]
DFS MC: Restore from system states even if we have to run some transitions out of it

This makes the model-check/checkpoint a bit more efficient by
increasing the amount of cases where we can reuse taken checkpoints.

It is still awfully slow because checkpoints are. The next step is to
implement model-check/checkpoint with forks.

11 months agoMerge branch 'dag-lab' into 'master'
Martin Quinson [Wed, 10 May 2023 11:08:09 +0000 (11:08 +0000)]
Merge branch 'dag-lab' into 'master'

Add dag scheduling lab to docs

See merge request simgrid/simgrid!150

11 months agoMerge branch 'changelog-operation-plugin' into 'master'
Martin Quinson [Wed, 10 May 2023 11:07:36 +0000 (11:07 +0000)]
Merge branch 'changelog-operation-plugin' into 'master'

Add plugins operation and battery to changelog

See merge request simgrid/simgrid!149

12 months agoUse forwarding references with std::forward (sonar).
Arnaud Giersch [Tue, 9 May 2023 14:46:08 +0000 (16:46 +0200)]
Use forwarding references with std::forward (sonar).

(hope I got it right this time)

12 months agoUse a std::vector as an underlying container for backtrack_points, so that maximal_su...
Arnaud Giersch [Tue, 9 May 2023 14:27:35 +0000 (16:27 +0200)]
Use a std::vector as an underlying container for backtrack_points, so that maximal_subsets_iterator can have a noexcept move constructor by default.

12 months agoUse a static variable for empty_set, and allow default noexcept move constructor.
Arnaud Giersch [Tue, 9 May 2023 13:57:50 +0000 (15:57 +0200)]
Use a static variable for empty_set, and allow default noexcept move constructor.

12 months agoadd dag scheduling lab
Adrien Gougeon [Tue, 9 May 2023 12:45:25 +0000 (14:45 +0200)]
add dag scheduling lab

12 months agoPlug memory leak.
Arnaud Giersch [Mon, 8 May 2023 19:27:14 +0000 (21:27 +0200)]
Plug memory leak.

12 months agoUse pointer-to-const for parameter (sonar).
Arnaud Giersch [Sat, 6 May 2023 11:48:33 +0000 (13:48 +0200)]
Use pointer-to-const for parameter (sonar).

12 months agoFix sonar bug: attribute access on a value that can be 'None'.
Arnaud Giersch [Fri, 5 May 2023 08:38:39 +0000 (10:38 +0200)]
Fix sonar bug: attribute access on a value that can be 'None'.

12 months agoMisc Sonar issues.
Arnaud Giersch [Thu, 4 May 2023 13:13:45 +0000 (15:13 +0200)]
Misc Sonar issues.

* declare functions "const"
* use pointer-to-const or reference-to-const
* use structured bindings
* extract assignment from expression
* mark constructors "explicit"
* pass large object parameters by reference-to-const
* use default implementation constructor
* explicitly capture required variables in lambdas

12 months agomess up with operations
Fred Suter [Fri, 5 May 2023 01:20:01 +0000 (21:20 -0400)]
mess up with operations
+ chainable setters
+ getters
+ boost instrusive pointers

12 months agoRename Link::get_usage() to Link::get_load() for consistency with Host::
Martin Quinson [Thu, 4 May 2023 14:14:54 +0000 (16:14 +0200)]
Rename Link::get_usage() to Link::get_load() for consistency with Host::

12 months agoadd plugins operation and battery
Adrien Gougeon [Thu, 4 May 2023 09:22:11 +0000 (11:22 +0200)]
add plugins operation and battery

12 months agofurther simplify this example
Fred Suter [Wed, 3 May 2023 14:39:56 +0000 (10:39 -0400)]
further simplify this example

12 months agoOne use less of get_recipe() that will soon die
Martin Quinson [Fri, 28 Apr 2023 00:01:20 +0000 (02:01 +0200)]
One use less of get_recipe() that will soon die

12 months agoFactorize more code between DFSExplo and LivenessExplo, and UDPOR
Martin Quinson [Thu, 27 Apr 2023 23:25:04 +0000 (01:25 +0200)]
Factorize more code between DFSExplo and LivenessExplo, and UDPOR

12 months agoKill useless empty function.
Arnaud Giersch [Tue, 2 May 2023 14:32:11 +0000 (16:32 +0200)]
Kill useless empty function.

12 months agoMissing "override".
Arnaud Giersch [Tue, 2 May 2023 14:22:02 +0000 (16:22 +0200)]
Missing "override".

12 months agoSimplify expression.
Arnaud Giersch [Tue, 2 May 2023 14:14:43 +0000 (16:14 +0200)]
Simplify expression.

12 months agoFix use-after-free observed with the s4u-operation examples.
Arnaud Giersch [Tue, 2 May 2023 14:12:18 +0000 (16:12 +0200)]
Fix use-after-free observed with the s4u-operation examples.

12 months agoCosmetics (sonar).
Arnaud Giersch [Tue, 2 May 2023 14:24:29 +0000 (16:24 +0200)]
Cosmetics (sonar).

12 months agoDefine each identifier in a dedicated statement (sonar).
Arnaud Giersch [Tue, 11 Apr 2023 09:13:59 +0000 (11:13 +0200)]
Define each identifier in a dedicated statement (sonar).

12 months agoRevert "Run mc-*-liveness tests serial, and hope to pass on CI."
Arnaud Giersch [Sat, 29 Apr 2023 08:03:02 +0000 (10:03 +0200)]
Revert "Run mc-*-liveness tests serial, and hope to pass on CI."

This reverts commit 881507dc9fbeba0c63112dd0058d9c591b67419e.

In fact, it was just the CI infrastructure that was unusually slow these days.

12 months agoWhitespace cleanup (codefactor.io).
Arnaud Giersch [Tue, 18 Apr 2023 09:39:48 +0000 (11:39 +0200)]
Whitespace cleanup (codefactor.io).

12 months agoLazily compute the recipe of a state, on need only
Martin Quinson [Thu, 27 Apr 2023 13:41:03 +0000 (15:41 +0200)]
Lazily compute the recipe of a state, on need only

This will be useful when leveraging intermediate states during
backtracks. In that case, we need to compute partial recipes, up to
the advanced restaure point.

12 months agoMore automatic memory mgmt in MC
Martin Quinson [Thu, 27 Apr 2023 12:09:03 +0000 (14:09 +0200)]
More automatic memory mgmt in MC

12 months agoMC: give each state an incoming transition too
Martin Quinson [Thu, 27 Apr 2023 11:48:19 +0000 (13:48 +0200)]
MC: give each state an incoming transition too

Using the parent's outgoing transition is not reliable anymore now
that we don't always explore the tree in order, but rather following
the exploration strategies.

12 months agoFix comment [no-ci]
Martin Quinson [Tue, 18 Apr 2023 07:01:51 +0000 (09:01 +0200)]
Fix comment [no-ci]

12 months agoRun mc-*-liveness tests serial, and hope to pass on CI.
Arnaud Giersch [Thu, 27 Apr 2023 12:57:19 +0000 (14:57 +0200)]
Run mc-*-liveness tests serial, and hope to pass on CI.

12 months agoFix comment.
Arnaud Giersch [Thu, 27 Apr 2023 09:28:08 +0000 (11:28 +0200)]
Fix comment.

12 months agoThere is no need to declare the default constructors here.
Arnaud Giersch [Thu, 27 Apr 2023 09:27:02 +0000 (11:27 +0200)]
There is no need to declare the default constructors here.

12 months agoFix build error: exception specification of explicitly
Arnaud Giersch [Thu, 27 Apr 2023 09:26:20 +0000 (11:26 +0200)]
Fix build error: exception specification of explicitly
defaulted move constructor does not match the calculated one.