[Debian-med-packaging] Please check maybe-failed of bali-phy

Andreas Tille tille at debian.org
Tue Nov 3 17:57:49 GMT 2026


Hi Cheng Li,

Am Tue, Nov 03, 2026 at 11:03:16PM +0900 schrieb Cheng Li:
> I looked at the build log. This does not look like an architecture
> problem: three tests simply do not finish within their timeout on the
> riscv64 buildd and get killed.
> 
> >From the test summary in the log:
> 
> Ok:                785
> Fail:              0
> Timeout:           3
> 
> So 3 out of 788, and no failures. All three were killed by signal 15
> (SIGTERM), and there are no architecture-specific errors and no wrong
> results anywhere in the log.
> 
> The timeouts for these three tests are hardcoded in src/meson.build:
> 
> 176-181  generated alignment program standalone   timeout: 600   needs ~720s
> 184-185  bali-phy 5d test                         timeout: 300   needs ~370s
> 186-187  bali-phy 5d +A 50                        timeout: 300   needs ~460s

Thank you for your analysis.
 
> The estimates come from comparing the 161 tests that pass on both
> riscv64 and armhf and take more than 5s: riscv64 is roughly 11x (median)
> slower than armhf. All three estimates are above the current limits, and
> close enough to them that the tests were most likely still progressing
> rather than stuck.
> 
> Note that bali-phy 48 +A 3 took 487.89s against a 600s limit, only 19%
> headroom, so it is likely to be the next one to time out.
> 
> As for a fix, we could raise the timeouts via dh_auto_test:
> 
>   dh_auto_test -- --timeout-multiplier 3

While I think this could finally work to build the package on all
architectures I wonder whether we want to stress test Debian's slow
architecture with this kind of things.
 
> If that sounds reasonable, I am happy to report this as a bug and
> provide a patch.

I think the optimal solution would be some flag HEAVY_TESTS (or so) in
upstream test which we can switch off from debian/rules for selected
architectures.

I'd welcome if you would create a MR upstream implementing this.  As
far as I know upstream is responsible and friendly to Debian (was
even directly involved into the packaging.

Thanks again for your help
   Andreas.

-- 
https://fam-tille.de



More information about the Debian-med-packaging mailing list