[SR-11287] memory leak in code completion about swift-syntax HOT 8 CLOSED

Interestingly, a lot of the other completion tests (`test/IDE/complete_*`) have leaks detected by LSAN even at commit 482d062. So maybe there were already some completion-related leaks, and then PR26403 added even more?

from swift-syntax.

@rintaro have you made any progress on this? It's currently blocking some work on a Google-internal SourceKit deployment that I'm working on. Would it maybe be possible to roll back the PR?

from swift-syntax.

@marcrasi Sorry for being late.
swiftlang/swift#26664 fixes leak in IDE/complete_repl_identifier_prefix_1.swift which is introduced by swiftlang/swift#26403

Interestingly, a lot of the other completion tests (`test/IDE/complete_*`) have leaks detected by LSAN even at commit 482d062. So maybe there were already some completion-related leaks, and then PR26403 added even more?

Right. I'm still fixing existing leaks.

from swift-syntax.

Awesome, I just verified that swiftlang/swift#26664 fixes the issue in the Google-internal SourceKit too! Thanks 🙂

from swift-syntax.

Do you think it would be possible to configure the LSAN CI job to run some SourceKit completion tests, to catch regressions earlier? Looks like that job currently doesn't run any tests (https://github.com/apple/swift/blob/1afd0f7cf757ba6b12d2af0d151eafb022acf439/utils/build-presets.ini#L1084), I guess because a lot of them leak so the job would just fail. Maybe we could configure it to run a whitelisted subset of tests under LSAN and slowly whitelist more tests as more leaks get fixed.

I'd be happy to look into this if you think it's a good idea.

from swift-syntax.

I think that's great idea. Could you file another JIRA issue for that?
CC: @shahmishal

from swift-syntax.

Filed https://bugs.swift.org/browse/SR-11313 !

from swift-syntax.

Merged swiftlang/swift#26664

from swift-syntax.