[PATCH] D43219: [PartialInlining] Fix Crash from holding a reference to a destructed ORE

[Sean Fertile via Phabricator via llvm-commits]

sfertile added a comment.

In https://reviews.llvm.org/D43219#1043101, @modocache wrote:

> Thanks for tracking down this bug! You may wish to ask @anemet for a review, I think he would have a good idea of what a proper fix would look like. As-is this LGTM, although in my opinion the test case could benefit from being reduced further.


Thanks, asking @anemet is a good idea. I actually have a new patch where I roll back to just creating the OREs directly at the point of use rather then trying to get it 2 different ways from the legacy and new passes.  This makes the code nice and straightforward, but I don't know if there are any drawbacks to doing it that way.  I got sidetracked testing that though when using the new pass manager and --pgo in the test-suite caused a number of failures. (Ended up being a know issue https://bugs.llvm.org/show_bug.cgi?id=33773) .

The test I included here is so big because it had to fail being inlined to trigger the crash, I've realized now I can probably manually set the inline threshold to something lower and reduce it down to something reasonable.


Repository:
  rL LLVM

https://reviews.llvm.org/D43219