CoreFoundation: add a workaround for Linux with the rebranch #3086
Conversation
|
@swift-ci please test |
Avoid optimizations on Linux as LICM, LoadStoreHoisting, and inlining seems to play bad tricks on range merging. This allows the AttributedString tests to pass on Linux after the rebranch. SR-15302 tracks resolving the underlying issue(s). Thanks to @etcwilde for tracking down this issue!
|
@swift-ci please test |
|
Do we want to leave a note that we know that there is an issue with |
|
I think that's better on the JIRA issue; I suspect that once we get a reduced test case, we can talk about this more reliably. It feels like we don't have a strong enough grasp as we couldn't localize it to a single function, and trying to isolate a set of functions seems ... like time better spent on fixing the issue. I would however agree that the change should have a comment referencing the SR .... which I swear I put in. :sigh: I don't want to touch this until the results are back from the associated PR. Worst case, a follow up to add that is something I might argue for, just to avoid the CI penalty 👎 |
That's fine by me. |
This adds an optimization barrier in the loop that is potentially
getting mis-optimized with the rebranch on Linux. This should allow us
to rebranch and follow up with a proper fix subsequently.
Thanks to @etcwilde for tracking down this issue!