Release testers, if you still have cycles (perhaps you didn't even
have time to start rc4 yet), please run the test script, share your
results, and upload binaries.
As mentioned above, this rc is very similar to the previous one. There
are no open blockers, so it could be the last release candidate.
Just a heads up - I won’t be able to do any macOS builds for a bit - my mac pro has died and needs to be replaced. I hope to get a replacement next week, but meanwhile I won’t be able to do any testing or binary uploads.
Let me know if anyone else with a mac have some CPU cycles to spare for this release.
Thanks that would be great - llvm comes with a script that builds and runs the tests in llvm/utils/release/test-release.sh
I usually create a release directory and then invoke it like this
…/llvm/utils/release/test-release.sh -release 11.0.0 -rc 5 -triple x86_64-apple-darwin -use-ninja -j
That takes two hours or so then I inspect the log file for testing errors:
grep “FAIL:” logs/testing-rc5.txt (I think don’t remember exactly the name of the file).
make sha256 sum: shasum -a 256 clang+llvm-…tar.xz
And then upload to the sftp server, but in this case - feel free to put it on gdrive or something like that and I can upload it to the sftp server for Hans to add to the release.
It sounds like you are saying there are two errors and that may appear in the totals counts in an Unexpected Failures line after the Expectedly Failed line.
The detail of the errors will commonly be immediately above the totals counts area and can be seen by reducing the start line number enough in the above sed command. You can then copy the errors to the llvm-dev list or create a blocking bug tied to as noted below.
I looked into the libunwind failures. There was a misconfiguration resulting in the compiled libunwind not actually being linked against the tests (they were linking against the system unwinder instead), and fixing that fixed libunwind_01.pass.cpp. signal_frame.pass.cpp is failing because the unwinder doesn't recognize signal frames; I don't know if that's expected with the compact unwind encoding format that's used for Mach-O, and I'm CCing Saleem in case he knows. libunwind doesn't appear to have a Bugzilla category, so I'm not sure how best to proceed here and if this test failure should be considered launch-blocking. The test also fails for me with 10.0.1, so it's not a regression in that sense.
Release testing for llvm-11.0.0-rc5 on Power PC 64bit Little Endian for Ubuntu 18.04 and Red Hat 7.4 is completed, did not see any regressions.
I have uploaded the respective binaries and have attached the sha1 files.
I have uploaded clang+llvm-11.0.0-rc5-x86_64-apple-darwin.tar.xz to
the sftp for you.
Shoaib - thanks for doing this build and checking out the test
failures. I didn't see any failures on my machine before as far as I
can remember - unfortunately I can't double check now with my machine
fried. I wonder if it's a difference in setup or me just missing it?
Anyway since it's not a regression compared to 10.0.1 - I don't see it
as a blocker for the 11 release.