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

Cheng Li shiptux at gmail.com
Sat Nov 7 11:40:07 GMT 2026


Hi Andreas,

> 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.

Done, as a pull request since that is what upstream's CONTRIBUTING.md asks
for:

  https://github.com/bredelings/BAli-Phy/pull/33

It tags the long-running tests into a meson test suite, so a slow architecture
could skip them with "meson test --no-suite long-running".

But I think we should not use it, and I may close the PR. Upstream is already
fixing the cause:

  "I have on-going work to address this by just making all the tests run
  faster. [...] The following tests are now much faster -- maybe 10 times as
  fast.  I'm also avoiding running multiple tests in parallel until the caches
  are created."

riscv64 is about 11x slower than armhf on this package, taking the median over
the 784 tests that pass on both, so a 10x on the tests themselves puts it at
roughly armhf speed, where nothing is near its limit. A skip-list in
debian/rules would then be dead weight that hides a real regression later. I
would rather wait for his work than patch anything on our side.

Two things for later. The slowest riscv64 tests are not the three that fail:
prob_prog/infer_tree/yule/1 takes 195.98s against its 300s limit and 29 tests
use more than a third of theirs, so that is where the next timeout will come
from if the speedup does not reach them. And upstream offered to continue the
discussion on a Debian list rather than in the PR -- would you rather have that
on debian-med at lists.debian.org, or here?

Kind regards,
Cheng Li



More information about the Debian-med-packaging mailing list