[Runtime] Fix a false metadata cycle diagnostic when threads race to instantiate cyclical metadata. #80505
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@swift-ci please test |
…instantiate cyclical metadata. The metadata creation system detects cycles where metadata depends on other metadata which depends on the first one again and raises a fatal error if the cycle can't be fulfilled. Some cycles can be fulfilled. The cycle may involve a requirement for a metadata state less than full transitive completeness which can be reached without resolving the entire cycle. We only want to raise a fatal error when we detect a cycle that can't be fulfilled. Normally this happens because the cycle checking in `blockOnMetadataDependency` only sees a cycle when it can't be fulfilled. Metadata initialization is advanced as far as it can be at each stage, so a cycle that can be fulfilled will see a fulfilling state and won't generate the dependency in the first place, since we only generate dependencies that haven't yet been met. However, when two threads race to create types in a cycle, we can end up with such a dependency, because the dependency may be generated before another thread fulfilled yet. The cycle checker doesn't account for this and incorrectly raises a fatal error in that case. Fix this by checking the cyclic dependency against the metadata's current state. If we have a dependency that's already been fulfilled, then there isn't really a dependency cycle. In that case, don't raise a fatal error. rdar://135036243
mikeash
force-pushed
the
metadata-cycle-race-fix
branch
from
5feb410 to
9ad534b
Compare
|
@swift-ci please test |
|
@swift-ci please test linux platform |
|
@swift-ci please test windows platform |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The metadata creation system detects cycles where metadata depends on other metadata which depends on the first one again and raises a fatal error if the cycle can't be fulfilled.
Some cycles can be fulfilled. The cycle may involve a requirement for a metadata state less than full transitive completeness which can be reached without resolving the entire cycle. We only want to raise a fatal error when we detect a cycle that can't be fulfilled.
Normally this happens because the cycle checking in
blockOnMetadataDependencyonly sees a cycle when it can't be fulfilled. Metadata initialization is advanced as far as it can be at each stage, so a cycle that can be fulfilled will see a fulfilling state and won't generate the dependency in the first place, since we only generate dependencies that haven't yet been met.However, when two threads race to create types in a cycle, we can end up with such a dependency, because the dependency may be generated before another thread fulfilled yet. The cycle checker doesn't account for this and incorrectly raises a fatal error in that case.
Fix this by checking the cyclic dependency against the metadata's current state. If we have a dependency that's already been fulfilled, then there isn't really a dependency cycle. In that case, don't raise a fatal error.
rdar://135036243