-
Notifications
You must be signed in to change notification settings - Fork 542
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
t/porting/bench.t: Failures on Fedora Linux 38 / gcc and g++ builds #21105
Comments
Failures with using gcc-13 for build: If the test were passing, this is what the output of the failing unit tests would look like:
|
I can reproduce the issue on arch with gcc 13 (but I don't know if it failed previously)
|
On 5/19/23 11:54, Leon Timmermans wrote:
I can reproduce the issue on arch with gcc 13 (but I don't know if it
failed previously)
Could you try with gcc-13 and g++-13 with the 5.37.11 release (or tag
v5.37.11)?
I doubt any smoke tester even attempted version 13 until very recently.
|
Also fails. As does 5.36.
The test is skipped if one doesn't have valgrind installed or if it isn't a git checkout, I bet that combination is uncommon. |
I guess this problem begs the question: At what point in the release cycle of major C-compilers to we begin assuring that the Perl core distribution builds and tests properly with that compiler version? I note that GNU has had two major update releases within the past month. GCC 12.3 released [2023-05-08] The release notes for GCC13 state: "The GCC developers are pleased to announce the release of GCC 13.1. This release is a major release, containing new features (as well as many other improvements) relative to GCC 12.x." Moreover, in my experience C-compiler upgrades tend not to break the core distribution test suite. They're more likely to generate new build-time warnings, for which we create issues, label them as such, address them and move on. Here, however, it appears that we've been bitten by a very recent release of a major C-compiler. There's obviously a policy discussion immanent here, and policy discussions are supposed to take place on the P5P list. But what do we do to clear away an obstacle to 5.38.0's release? |
On Fri, May 19, 2023 at 01:41:14PM -0700, James E Keenan wrote:
@rjbs @book @leonerd : Is this a release blocker?
t/porting/bench.t just does some basic sanity checks that the
Porting/bench.pl benchmarking script works. Since that tool is currently
only targeted at core hackers and isn't installed anywhere, I don't
think it matters if that test file, or even bench.pl itself doesn't
work on a particular very bleading edge platform.
So not a blocker.
I have no opinion on whether any other problems with GCC-13 (are there any?)
or whatever should be a blocker.
…--
My get-up-and-go just got up and went.
|
I wouldn't be surprised if the real culprit was a valgrind upgrade instead of a gcc update. Given that it's failing to parse cachegrind output. This issue may need someone who actually understands valgrind. |
It's only really useful for core hackers, but I can imagine the test failing for distributors. It might already help a lot if we extend that |
On Fri, May 19, 2023 at 02:09:44PM -0700, Leon Timmermans wrote:
> Is this a release blocker?
It's only really useful for core hackers, but I can imagine the test failing for distributors. It might already help a lot if we extend that `-d "./.git"` to `-d "./.git" && -e '.mailmap'` (and we may need a better solution for this once blead is open again)
Perhaps the whole test file should skip if its not a devel (odd-numbered)
perl version?
…--
Spock (or Data) is fired from his high-ranking position for not being able
to understand the most basic nuances of about one in three sentences that
anyone says to him.
-- Things That Never Happen in "Star Trek" #19
|
Probably. Though having an explicit opt-out may also be useful as I currently can't successfully test perl until I uninstall valgrind. |
The spacing of some of the output fields changed: https://sourceware.org/git/?p=valgrind.git;a=commit;h=2cccba7cae9f77aa6c2a498bc30a42bdb1ed8829 --cache-sim=no is now the default: https://sourceware.org/git/?p=valgrind.git;a=commit;h=57dbcacfdbff0d4c12dcd52ff56f159631499dc6 Fixes Perl#21105
The spacing of some of the output fields changed: https://sourceware.org/git/?p=valgrind.git;a=commit;h=2cccba7cae9f77aa6c2a498bc30a42bdb1ed8829 --cache-sim=no is now the default: https://sourceware.org/git/?p=valgrind.git;a=commit;h=57dbcacfdbff0d4c12dcd52ff56f159631499dc6 Fixes #21105
The spacing of some of the output fields changed: https://sourceware.org/git/?p=valgrind.git;a=commit;h=2cccba7cae9f77aa6c2a498bc30a42bdb1ed8829 --cache-sim=no is now the default: https://sourceware.org/git/?p=valgrind.git;a=commit;h=57dbcacfdbff0d4c12dcd52ff56f159631499dc6 Fixes Perl#21105
One of our steady smoke-testers has recently begun testing on Fedora Linux 38 (Server Edition).
t/porting/bench.t
is now failing on bothgcc
andg++
builds. Here is the list of failed runs:https://perl5.test-smoke.org/submatrix?test=../t/porting/bench.t&pversion=5.37.12
Sample g++ failure: https://perl5.test-smoke.org/report/5034442
This test was passing on the previous version of this OS, where the build was done with
g++-12
. So I don't know whether this is a Fedora-specific problem or a problem with version 13 of g++ and gcc.I don't have access to version 13 of gcc or g++, so if anyone else has those compilers, please build blead with them and run t/porting/bench.t.
@cjg-cguevara
Thank you very much.
Jim Keenan
The text was updated successfully, but these errors were encountered: