Bug#951821: gegl: FTBFS on arm*/mips* architectures; testing migration blocked

[Simon McVittie]

On Sat, 22 Feb 2020 at 00:23:13 -0500, Boyuan Yang wrote:
> Looking into the current status, it fails to build on all ARM and MIPS
> architectures. The reasons are mostly timeout in tests. Please make some
> investigation and consider increasing the timeout limit or ignore test errors
> during building.

I recently uploaded gegl_0.4.22-2 which increases the timeout on slower
architectures. This reveals that there were several different failure modes:

- Some tests were genuinely just slow. Previously, they often timed out
  on the slower machines; now they pass.

- gegl:simple / backend-file sometimes times out with no output, especially
  on the arm* family (but an upstream bug report suggests that this can
  also happen on x86_64). I've opened a separate FTBFS bug #954197
  to represent this.

- The "tests/compositions/reference/rgbe-save.hdr" step in
  gegl:compositions / compositions_without_opencl fails on mipsel (only):
  the composed image differs from the reference. I've opened a separate
  FTBFS bug #954196 to represent this.

- gegl:simple / buffer-sharing failing on Hurd appears to be Hurd-specific.
  We already had #892238 (non-RC).

- gegl:ff-load-save / ff-load-save fails on ia64, which appears to be
  because examples/frame-counter segfaults.

I am reluctant to ignore test failures on arm* and mips*, because those
are precisely the sort of architectures where we claim that packages
work, but due to their much smaller number of users, automated tests
are our only actual evidence that those packages really do work.

On Thu, 12 Mar 2020 at 15:36:20 -0400, Boyuan Yang wrote:
> On Sun, 23 Feb 2020 00:05:54 +0100 Andreas Henriksson <andreas at fatal.se>
> wrote:
> > I'm disabling parallel tests via 'dh_auto_test --no-parallel' now
> > but it seems there are more test failures on the archs where
> > the build failed previously.
> 
> Any progress on it?  If it still takes much time, I suggest we force single-
> thread build first and investigate later since the current FTBFS is blocking
> many packages.

Forcing serial testing was already done since 2020-02-22, in 0.4.22-1.

    smcv