Andreas Tille, on 2022-11-05: > Am Sat, Nov 05, 2022 at 05:36:16PM +0100 schrieb Étienne Mollier: > > So I have been spending some time to try to put salmon back in > > working shape. The code currently available in Salsa[1] builds > > through, but fails it's test suite. Said test is fortunately > > straightforward to run, and from further analysis with the > > debugger and test rebuilds with the pristine upstream source > > code, I believe I isolated that the root cause of the test > > failure is that salmon crashes with a segmentation fault when > > built against libcereal 1.3.2, while it works properly when > > built against libcereal 1.3.0, which is the version vendored by > > upstream if a local copy is not available. > > Thanks a lot for this detailed analysis. As a quick (hopefully > temporary) solution I could imagine to keep the code copy of libcereal > for the moment. At the same time it makes sense to inform salmon > upstream. Further inspection of the backtrace shows the linker was confused between the embedded and system version of libcereal. Everything goes back to normal when the libcereal copy embedded in pufferfish embedded in salmon is deleted (one more reason to avoid vendoring). On upstream code, this situation can occur when their external/ directory is not properly cleaned up and there is a system libcereal installed in the meantime. Not sure what to do about this, this looks like quite the corner case to make some noise about it. > > [1]: https://salsa.debian.org/med-team/salmon Have a good night, :) -- Étienne Mollier <emollier@emlwks999.eu> Fingerprint: 8f91 b227 c7d6 f2b1 948c 8236 793c f67e 8f0d 11da Sent from /dev/pts/2, please excuse my verbosity. On air: Lanfear - Traces Of Infinity
Attachment:
signature.asc
Description: PGP signature