| Commit message (Collapse) | Author | Age |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We rely on ErrorTransienceValue being different across evaluations by
way of java reference inequality, instead make it a singleton and make
it so that ErrorTransienceValue compares equal to nothing, including itself.
Alternatively we could attach the corresponding Version to ErrorTransienceValue
and use that to force inequality across evaluations. However this gets into
the different possible Version implementations and their varying semantics, so
go with the simplest approach first and see how it works out.
--
MOS_MIGRATED_REVID=106001826
|
|
|
|
|
|
|
| |
Nothing can extend SkyFunctionName. It's final.
--
MOS_MIGRATED_REVID=105989976
|
|
|
|
|
|
|
|
| |
Refactors SkyFunction map creation into the tester from tests.
Encapsulates modifiedValues better.
--
MOS_MIGRATED_REVID=105982046
|
|
|
|
|
|
|
| |
RuntimeExceptions that were thrown after an InterruptedException was thrown would be swallowed.
--
MOS_MIGRATED_REVID=105902192
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=105889244
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=105539869
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=104791799
|
|
|
|
|
|
|
|
|
|
|
| |
The headers were modified with
`find . -type f -exec 'sed' '-Ei' 's|Copyright 201([45]) Google|Copyright 201\1 The Bazel Authors|' '{}' ';'`
And manual edit for not Google owned copyright. Because of the nature of ijar, I did not modified the header of file owned by Alan Donovan.
The list of authors were extracted from the git log. It is missing older Google contributors that can be added on-demand.
--
MOS_MIGRATED_REVID=103938715
|
|
|
|
|
|
|
| |
avoid mutating the deps of nodes that are still going to be deps after evaluation is finished.
--
MOS_MIGRATED_REVID=103659429
|
|
|
|
|
|
|
| |
since ErrorInfo doesn't override equality, there's no issue with preserving equality for now.
--
MOS_MIGRATED_REVID=103414237
|
|
|
|
|
|
|
|
|
| |
to interrupt evaluation in constant time.
Some ParallelEvaluator tests that implicitly relied on cleaning happening before the next evaluation were moved into MemoizingEvaluatorTest as a result.
--
MOS_MIGRATED_REVID=102696653
|
|
|
|
|
|
|
| |
NotifyingInMemoryGraph in tests.
--
MOS_MIGRATED_REVID=102616906
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=102265945
|
|
|
|
|
|
|
| |
method wasn't guaranteed to catch all bugs. Also convert to a singleton since there's no reason to have multiple instances.
--
MOS_MIGRATED_REVID=102158719
|
|
|
|
|
|
|
| |
deflaking I just did.
--
MOS_MIGRATED_REVID=102149874
|
|
|
|
|
|
|
|
|
| |
ensuring that the shutdown signaler starts before we do other things.
Also store assertion errors in the unexpected exception variable, as opposed to allowing them to get swallowed. Note that there's still some flakiness, hence the especially verbose error messages.
--
MOS_MIGRATED_REVID=102089689
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=101818063
|
|
|
|
|
|
|
|
|
| |
Parents should depend on their children's transience to force their re-evaluation, not say directly that they are transient. This can lead to an inconsistency between whether an ErrorInfo says it is transient and whether the node depends on the ErrorTransienceValue.
This should reduce the number of edges in the graph in case of transient error without sacrificing correctness.
--
MOS_MIGRATED_REVID=101537029
|
|
|
|
|
|
|
|
|
| |
if it isn't newer than the current node.
ErrorTransienceValue has to be treated specially during pruning because normally deps added during pruning remain if the node is actually rebuilt. The whole point of ErrorTransienceValue is that on a re-evaluation, the node may not depend on it anymore. So we have special logic to force a rebuild if the ErrorTransienceValue has changed. If it hasn't, however, it is ok for pruning to add it back as a dep -- we'll end up marking this value clean, since ErrorTransienceValue is always the last dep.
--
MOS_MIGRATED_REVID=101522197
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=101245691
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=101150250
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The primary user of invalidation tracking is the SkyframeBuildView,
which monitored which ConfiguredTargetValues were invalidated. It
did that so the SkyframeExecutor could limit its search for artifact
conflicts to when the set of configured targets in the build changed.
For the newly introduced set of dirtied keys,
"dirtiedConfiguredTargetKeys" in SkyframeBuildView, to be as useful as
the "dirtyConfiguredTargets" set it replaced, the
ConfiguredTargetValueInvalidationReceiver must only remove a key from
the set if it was found to be clean when it was re-evaluated. If it
was rebuilt, then the key must stay in the set, to represent that the
set of configured target values has truly changed.
This CL introduces a semantic change that hopefully has a small effect,
if any. Previously, the informInvalidationReceiver method in
InvalidatingNodeVisitor only informed the invalidationReceiver when a
non-null value was invalidated. (This is usually, but not always,
equivalent to a non-error value. The uncommon exception is that in
keep-going builds, some nodes with errors may also have values, and
the invalidator would inform the receiver when such a node was
invalidated.) Now, the receiver is informed that a node is invalidated
regardless of whether its value is null. Because the receiver uses this
information to determine whether artifact conflict search has to be
rerun, and that search is expensive, it's possible this change will
negatively impact performance. However, the only way an extra search
could be invoked is if the invalidated configured target nodes are
all in error. That seems like it would happen rarely, if at all.
Further cleanup of unused SkyValues returned by markDirty to come in
a subsequent CL.
--
MOS_MIGRATED_REVID=100304744
|
|
|
|
|
| |
--
MOS_MIGRATED_REVID=98914195
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We do this by implementing equality for TaggedEvents (and all objects
it transitively includes). Before this change, if a Skyframe node
re-evaluated to the same value as in the previous build, but had
(transitive) events, change pruning would not cut off the evaluation
of its parents. This is not a big issue in practice because most nodes
that would re-evaluate to the same value (like FileValues or
GlobValues) never emit events, and others (like ActionExecutionValues)
have secondary caches that mask this effect.
Also do a drive-by fix where we were using the hash code of a nested
set instead of the shallow hash code (didn't have any bad effects in
practice because we never hash these values).
(Minor formatting clean-ups from https://bazel-review.googlesource.com/1610 )
--
Change-Id: I751a8479627f0456993c5ec8834528aeb593d736
Reviewed-on: https://bazel-review.googlesource.com/1610
MOS_MIGRATED_REVID=98115908
|
|
|
|
|
|
|
|
|
| |
If otherTop did not enqueue its dep for evaluation before errorKey threw, its dep would never be enqueued and therefore never signal. This used to be ok because we didn't shut the threadpool down as aggressively.
The fix is just to delay the actual throwing of errorKey until otherTop is signaled.
--
MOS_MIGRATED_REVID=93349168
|
|
|
|
|
|
|
|
|
| |
encounters a node with an error.
We were doing this in most cases, but not if the error was in a node that was already done or was revalidated during change pruning.
--
MOS_MIGRATED_REVID=92521223
|
|
|
|
|
|
|
| |
dependency group, were checked for dirtiness.
--
MOS_MIGRATED_REVID=89658366
|
|
|
|
|
|
|
|
|
| |
This allows the pending action counter to be decremented correctly when in --keep_going mode.
As part of this, there's a small refactor in ParallelEvaluator that also fixes a potential bug, that nodes in error were signaling their parents with the graph version as opposed to their actual version. Because errors never compare equal (ErrorInfo doesn't override equality), this isn't an issue in practice. But it could be in the future.
--
MOS_MIGRATED_REVID=88395500
|
|
--
MOE_MIGRATED_REVID=85702957
|