Hi.[Just in case you're not aware, the BTS doesn't forward replies to the bug submitter unless they are subscribed. If you want to have the submitters attention, please add them explicitly.]
On Wed, 29 May 2024 19:55:19 +0200 Philippe SWARTVAGHER <phil.sw...@gmx.fr> wrote:
Well, this issue is "normal". Currently, shaderc 2023.2-1 is in testing. This version suffers from an "undefined symbol" bug (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1070983). This bug is fixed in version 2023.8-1 which is now in unstable.
Can you please elaborate how the fix works. This feels like you're using a symbol that's only available from a new version of glslang, why don't you have the right versioned constrained to prevent migration of shaderc without the right version providing the symbol? And how can the test of shaderc in testing work with the version of glslang in testing? It really looks like src:glslang is still failing to provide the symbol that shaderc (and maybe other packages in testing) are expecting from it.
What I mean to say is that very regularly when an ABI breaks "just rebuilding" reverse dependencies fixes missing symbol issues, but it's the wrong fix. Bumping SONAME when ABI breaks and going through a transition is the right solution if the dropping of the symbol is intended. (I'm not saying that's the case here as I don't fully understand it).
However, this version never migrated to testing due to a blocking bug in glslang (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1062799). This bug in glslang is now fixed with the last upload.
But the new version of glslang still doesn't provide the missing symbol.
I'm a beginner regarding this kind transition / blocking bug / CI, so I don't know how this bug can be solved...
From where I stand this looks like ABI is broken and if it's not going to be reverted, the SONAME must be bumped and we have to go through a transition: https://wiki.debian.org/Teams/ReleaseTeam/Transitions
I don't know if it helps, but I have a RFS (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1072181) for a new version of shaderc (which changes nothing about this bug). You can see on Salsa that autopkgtests pass with the version of glslang available in unstable: https://salsa.debian.org/debian/shaderc/-/pipelines/683648.
Well, shaderc in unstable is build against the new version of glslang. As I mentioned before, rebuilds often paper over missing symbols problems, but is the wrong solution.
Paul
OpenPGP_signature.asc
Description: OpenPGP digital signature
