Closed uenoku closed 2 days ago
Thanks for reminding.
This smells kinda like a caching issue... I'm gonna try clearing the relevant caches.
Nope. Clearing short integration tests cache didn't work. Same issue. I'm gonna revert the clang version change in the image.
Oh shit... the clang change was for the regular build image, not the integration test image.
Here's the 13.1 -> 15.0 image build diff: https://github.com/circt/images/compare/v13.1...v15.0#diff-6a1298b7c0a74ef094b033382b6e61d38a5213fcff54392b62b4d7880d5b6934 . I don't see any changes in there (in the integration test image) which could cause this issue...
I think absl::lts_20230125::string_view
is the problem here. That is not the abseil version or-tools builds with. Its the one shipped by gRPC. I think the CIRCT scheduling lib is pulling in the wrong header during compilation, so the signature doesn't line up with the or-tools library.
Yeah, that's what I was just investigating.
An interesting question is why the PR which contained the image change passed the short integration tests build but now the short integration tests build is failing.
I feel like I'm pretty close to a solution here: build Abseil on its own (and install it), then have gRPC and or-tools use the system-installed version. I'm just working through some "details" now but probably won't have time to finish today.
My current work-in-progress is here: https://github.com/circt/images/tree/teqdruid/fix-int-image
Well, I wasn't even close. Why is this so damn difficult to get right? or-tools seems very sensitive to the way Abseil is compiled.
I gave up, reverted the image, and disabled the ESI runtime build #7236. That PR should fix the nightlies and short integration test.
Last night's nightlies didn't fail last night. Closing this issue.
https://github.com/llvm/circt/actions/runs/9633232858/job/26567402157
Probably due to docker image bump v13.1 -> v15 in https://github.com/llvm/circt/commit/88eeb265f479ab9cd7800d9859dd623f4d90c891?