Version 1.8.4 seems to include some breaking changes. If I start gitlab which uses grpc I see the following error message: ==> /usr/local/www/gitlab/log/puma.stderr.log <== ld-elf.so.1: /usr/local/lib/libgrpc.so.39: Undefined symbol "_ZN4absl12lts_202301254CordC1INSt3__112basic_stringIcNS3_11char_traitsIcEENS3_9allocatorIcEEEETnNS3_9enable_ifIXsr3std7is_sameIT_S9_EE5valueEiE4typeELi0EEEOSB_" Downgrading to benchmark 1.8.3 and rebuild grpc fixes the problem.
Reverted.
A commit in branch main references this bug: URL: https://cgit.FreeBSD.org/ports/commit/?id=6d9df455a863e0618608af3a86d615ff29a7734e commit 6d9df455a863e0618608af3a86d615ff29a7734e Author: Yuri Victorovich <yuri@FreeBSD.org> AuthorDate: 2024-06-25 16:44:57 +0000 Commit: Yuri Victorovich <yuri@FreeBSD.org> CommitDate: 2024-06-25 16:44:57 +0000 devel/benchmark: Revert 1.8.4 → 1.8.3 PR: 279984 Reported by: Matthias Fechner <mfechner@FreeBSD.org> devel/benchmark/Makefile | 3 ++- devel/benchmark/distinfo | 6 +++--- devel/benchmark/pkg-plist | 9 ++++----- devel/grpc/Makefile | 2 +- games/openrct2/Makefile | 2 +- graphics/ospray-studio/Makefile | 2 +- graphics/ospray/Makefile | 2 +- math/e-antic/Makefile | 2 +- math/hpcombi/Makefile | 2 +- net-p2p/ethash/Makefile | 2 +- 10 files changed, 16 insertions(+), 16 deletions(-)
Hi Matthias, Is there a grpc issue to track to make it compatible with benchmark-1.8.4 ? Thanks, Yuri
Dear Yuri, I just tested it and the revert does not work. I did an additional test that succeeded. I reverted the git main branch to the commit before you reverted and did the following (the cherry-pick is necessary as I have to fix some other issues in the port): git checkout -b test 834a7d4a9bc2d9ac2adeb27345208d3a8e1fd5bc git revert -n 5add937f95a55b512758cb75e2f38cd4879d1556 git cherry-pick -n 67b9defc6d9c75bf435f64ef5546e299813e45fc 98487ce4845fedd0e3da8598a53431a61afaab55 f357a3c7cea47e3d38512ffb7591bd3039df049d Then I triggered a new poudriere build to generate the gitlab package with all dependencies: poudriere bulk -t -r -p gitlab -j 141amd64 www/gitlab@ce Rolled back my virtual machine where gitlab 17.0.1 was installed and upgraded all packages. It is working fine. Let me check what happens, if I just increase some PORTREVISION of the ports, maybe there is a hidden bug somewhere else.
I just did now an additional test and just added a PORTREVISION to the devel/grpc port. That break the system. So I have now benchmark 1.8.3 installed with grpc that has PORTREVISION=1. I think we need @sunpoet here to answer this question why increasing the PORTREVISION breaks grpc.
I did now one more test to verify if benchmark 1.8.4 will work or not (and it does). Used the following (used the commit before the revert and included necessary cherry-picks): g checkout -b test 834a7d4a9bc2d9ac2adeb27345208d3a8e1fd5bc g cherry-pick -n 67b9defc6d9c75bf435f64ef5546e299813e45fc 98487ce4845fedd0e3da8598a53431a61afaab55 f357a3c7cea47e3d38512ffb7591bd3039df049d edited devel/grpc/Makefile and removed the PORTREVISION line. Let poudriere build the packages (it detects down/up-grade and removes the packages for rebuild automatically): poudriere bulk -t -r -p gitlab -j 141amd64 www/gitlab@ce Reset my virtual machine again to gitlab 17.0.1 version and executed the following commands: service gitlab stop pkg clean -ay pkg delete gitlab-ce pkg autoremove # (I did this to make sure, we really install the newly build grpc) pkg install gitlab-ce pkg upgrade service gitlab start Then accessing the webpage of gitlab and it *works*. So the source of the problem is somehow related to the PORTREVISION line in devel/grpc which causes this problem.
I think I was now finally able to track the source of the problem (or at least it fixed the problem for me). Gitlab or maybe better grpc needs a more recent abseil: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=275887 Not sure why the problem got only visible by the upgrade of benchmark.