Bug#1025221: abseil: please consider disabling tests on riscv64
control: fixed -1 abseil/20220623.1-3 On 2022-12-06 11:15, Benjamin Barenblat wrote: > Control: owner 1025221 ! > Control: tags 1025221 - patch > > Hi, Manuel, > > Yesterday, I tried a build on a porterbox with your patch. It looks like > disabling parallelism improves the situation but does not completely > solve it; absl_mutex_test is still flaky. I’ll continue investigating > and see if I can get that test fixed. The real cause of the issue has been found and fixed in version 20220623.1-3. I am therefore marking the bug as fixed in this version. Thanks for your help. Regards Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://aurel32.net
Bug#1025221: abseil: please consider disabling tests on riscv64
Hi, Sorry, I didn't notice this reply at the time, I did with the recent activity in this bug report. So, as a belated reply to your earlier efforts, I am writing now to thank you for the effort of looking into this at the time, and sorry for my lack of reply then... and now to the specifics: 2022-12-06 17:15 Benjamin Barenblat: Control: owner 1025221 ! Control: tags 1025221 - patch Hi, Manuel, Yesterday, I tried a build on a porterbox with your patch. It looks like disabling parallelism improves the situation but does not completely solve it; absl_mutex_test is still flaky. I’ll continue investigating and see if I can get that test fixed. In the meantime, it looks like a recent binNMU to Abseil on riscv64 has passed the buildds. Does that unblock your work? And to reply to the question, no, when we were in the debian-ports infra it was not a blocker, as we always had the option to disable tests (if need be) and upload to the special suite "unreleased" which allows to have certain changes to the normal building instructions, like disabling tests. Now in the official infra we cannot do that as a last resort, so we're more limited and the packages have to be built "as-is", without any modifications, so as you have seen by the recent pings from other people it is more or a blocker now. As Aurélien and Stephane said, probably makes more sense to make these non-fatal at the moment, like you do in 20230125.3-2 in experimental for mipsel. Not sure if you'd like to go with the current 20220623.1-2 in unstable & testing for the transition, I suppose that by now the upstream version 20230125.3 must be quite stable. In any case we noticed that you joined #debian-riscv IRC channel, so maybe it's best to discuss it there to find the best option for everyone to go forward. Cheers and thanks. -- Manuel A. Fernandez Montecelo
Bug#1025221: abseil: please consider disabling tests on riscv64
On 2023-08-08 20:46, Aurelien Jarno wrote: > Hi Benjamin, > > On 2022-12-06 11:15, Benjamin Barenblat wrote: > > Control: owner 1025221 ! > > Control: tags 1025221 - patch > > > > Hi, Manuel, > > > > Yesterday, I tried a build on a porterbox with your patch. It looks like > > disabling parallelism improves the situation but does not completely > > solve it; absl_mutex_test is still flaky. I’ll continue investigating > > and see if I can get that test fixed. > > > > In the meantime, it looks like a recent binNMU to Abseil on riscv64 has > > passed the buildds. Does that unblock your work? > > Thanks for your upload fixing #1037567. Unfortunately it appears that we > are now bitten by this bug with a failure in the testsuite. Would that > be possible to use the same patch that you applied to the experimental > branch? [1] While it doesn't fully fix the issue, it increases the > chance for the testsuite to path. I haven't seen the message of Stephane before sending my email. His suggestion to make the test failures non-fatal on riscv64 on make fully sense to avoid blocking the bootstrap of riscv64. There are only 2 builds failures in the testsuite, the majority of it passes. Therefore please go ahead with his patch. We will continue to work on the issue to provide the real fix. Regards Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://aurel32.net
Bug#1025221: abseil: please consider disabling tests on riscv64
Hi Benjamin, On 2022-12-06 11:15, Benjamin Barenblat wrote: > Control: owner 1025221 ! > Control: tags 1025221 - patch > > Hi, Manuel, > > Yesterday, I tried a build on a porterbox with your patch. It looks like > disabling parallelism improves the situation but does not completely > solve it; absl_mutex_test is still flaky. I’ll continue investigating > and see if I can get that test fixed. > > In the meantime, it looks like a recent binNMU to Abseil on riscv64 has > passed the buildds. Does that unblock your work? Thanks for your upload fixing #1037567. Unfortunately it appears that we are now bitten by this bug with a failure in the testsuite. Would that be possible to use the same patch that you applied to the experimental branch? [1] While it doesn't fully fix the issue, it increases the chance for the testsuite to path. BTW, by just looking at the code, the problem with absl_mutex_test could be the use of rdcycle to measure the time. This counter is per CPU, so if the process is rescheduled to another CPU (which is likely in a multithreaded test) the time measurement is totally wrong. In addition rdcycle will likely going to be forbidden from the userland, for security reasons [2]. rdtime would be a better replacement here, but at this stage I haven't looked at how to change that, nor if its the real issue. Thanks, Aurelien [1] https://salsa.debian.org/debian/abseil/-/commit/f4f2c1da90c4e6a0683c4e66c0268baa1b79cdf3 [2] https://lore.kernel.org/lkml/20230802-7c19a712ae071f68030ab5f2@orel/T/ -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://aurel32.net
Bug#1025221: abseil: please consider disabling tests on riscv64
Control: owner 1025221 ! Control: tags 1025221 - patch Hi, Manuel, Yesterday, I tried a build on a porterbox with your patch. It looks like disabling parallelism improves the situation but does not completely solve it; absl_mutex_test is still flaky. I’ll continue investigating and see if I can get that test fixed. In the meantime, it looks like a recent binNMU to Abseil on riscv64 has passed the buildds. Does that unblock your work?
Bug#1025221: abseil: please consider disabling tests on riscv64
Source: abseil Version: 20220623.1-1 Followup-For: Bug #1025221 User: debian-ri...@lists.debian.org Usertags: riscv64 X-Debbugs-Cc: m...@debian.org, 1025221-submit...@bugs.debian.org Control: tags -1 ftbfs patch Control: retitle -1 abseil: please increase timeout or do not run tests in parallel on riscv64 Hello, I was building this package disabling parallelism and it seems to solve the issue. I solved it with the attached patch, but feel free to do it in some other way. The issue is that, due to limitations of the current building machines for the riscv64 arch, it seems that some of the tests reached timeouts and thus fail. Running serially seems to help by giving them more time to run before timing out, or something to that effect. I did not find an obvious way to increase the timeout, at least not without having more complicate patches, if you know how to do it in a simple way I would encourage you to go that way. And maybe by setting parallelism to lower levels, like half of the computing cores would be enough. But at the moment I think that this is an acceptable solution which keeps running the tests and not causing much extra maintenance effort, even if it takes a bit longer to build the package. Thanks and cheers. -- Manuel A. Fernandez Montecelo diff -Nru abseil-20220623.1/debian/changelog abseil-20220623.1/debian/changelog --- abseil-20220623.1/debian/changelog 2022-10-18 14:02:49.0 + +++ abseil-20220623.1/debian/changelog 2022-11-30 21:22:35.0 + @@ -1,3 +1,10 @@ +abseil (20220623.1-1+0.riscv64.1) unreleased; urgency=medium + + * Non-maintainer upload. + * riscv64: disable parallel checks + + -- Manuel A. Fernandez Montecelo Wed, 30 Nov 2022 21:22:35 + + abseil (20220623.1-1) unstable; urgency=medium * New upstream release. diff -Nru abseil-20220623.1/debian/rules abseil-20220623.1/debian/rules --- abseil-20220623.1/debian/rules 2022-10-18 12:37:10.0 + +++ abseil-20220623.1/debian/rules 2022-11-30 21:22:35.0 + @@ -50,8 +50,11 @@ dh_auto_build -Bshared ifeq ($(ABSL_RUN_TESTS),ON) +ifneq ($(filter $(DEB_HOST_ARCH),riscv64),) +TEST_PARALLEL=--no-parallel +endif override_dh_auto_test: - dh_auto_test -Bshared + dh_auto_test -Bshared $(TEST_PARALLEL) endif override_dh_auto_install:
Bug#1025221: abseil: please consider disabling tests on riscv64
Source: abseil Version: 20220623.1-1 Hello, looks like riscv64 is timeouting on some tests, due to non-fast-enough hardware. Can you please consider disabling the tests also on that architecture? Also powerpc might be disabled I think log on riscv64: 99% tests passed, 2 tests failed out of 182 Total Test time (real) = 152.50 sec The following tests FAILED: 165 - absl_mutex_test (Subprocess aborted) 167 - absl_per_thread_sem_test (Failed) Errors while running CTest log on powerpc: 99% tests passed, 2 tests failed out of 182 Total Test time (real) = 1500.51 sec The following tests FAILED: 49 - absl_symbolize_test (Subprocess aborted) 50 - absl_failure_signal_handler_test (Timeout) Errors while running CTest thanks, Gianfranco OpenPGP_signature Description: OpenPGP digital signature
