[PATCH] D102026: Thread safety analysis: Allow exlusive/shared joins for managed and asserted capabilities

[Delesley Hutchins via Phabricator via cfe-commits]

delesley accepted this revision.
delesley added a comment.
This revision is now accepted and ready to land.

Thanks for taking the time to discuss things with me.  :-)

Wrt. to the TEST_LOCKED_FUNCTION, I agree that you can simulate the behavior using Assert and Lock.  But that pattern is too general/powerful, because it also allows you to write nonsensical and unsound code.  Philosophically, static analysis is concerned with allowing things that are sound, but preventing things that are not, so I would prefer to allow the valid case, but warn on the nonsensical/unsound code.  The goal is not to provide powerful back doors so that you can squeeze anything past the checker -- doing so kind of defeats the point.  :-)  That being said, I'm not certainly no asking you to implement TEST_LOCKED functionality in this particular patch, and I totally understand that it may simply not be worth the effort.

Wrt. to unlocking an Assert, I see no problem.  It makes perfect sense to dynamically assert that a mutex is held, then do something, and then unlock the mutex afterwards; after all, the caller asserted that it was held before unlocking.  You shouldn't disallow that.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D102026/new/

https://reviews.llvm.org/D102026