Re: CMake 4 upload to unstable

[Lucas Nussbaum <lucas@debian.org>]

On 25/09/25 at 12:14 +0300, Adrian Bunk wrote:
> On Thu, Sep 25, 2025 at 10:33:31AM +0200, Lucas Nussbaum wrote:
> > On 25/09/25 at 10:00 +0200, Paul Gevers wrote:
> > > I'm wondering if we
> > > should hold off migrating cmake 4 while we're waiting a bit for autoremoval
> > > to do its job (including sending notifications). Or should we be more
> > > aggressive with manual removal? What do you think?
> > 
> > I wonder if we should decrease the auto-removal timeouts to make it
> > easier to get broken packages out of testing (and new packages in
> > testing) and increase it again later in the release cycle.
> > 
> > There's some value to faster transitions even if that means temporarily
> > removing non-key packages from testing. There's plenty of time to get
> > them back in testing.
> 
> Your suggestion might not result in faster transitions.
> 
> As long as the autopkgtest of at least one key package gets broken
> (as is the case for CMake) it's either a manual hint to ignore
> autopkgtest failures or no migration.

Ah, right

> Of the ~ 800 CMake 4 bugs only ~ 30 might have relevance for
> migration.[1]
> 
> BTW: The first key package with an autopkgtest broken by CMake 4
>      I looked at does not even have a bug (libzstd).

And there are others. I came up with a UDD query to combine the
pseudo-excuses data with bugs[1], and could not identify corresponding
bugs for the following autopkgtest regressions:

 double-conversion
 fastcdr (ppc64el only)
 ffc
 foonathan-memory
 khronos-opencl-headers
 libsfml
 libthread-pool
 libzstd
 msgpack-c
 msgpack-cxx
 mshr (ppc64el only)
 nvidia-cuda-samples (ppc64el only)

Also, capnproto has #1095942 but I'm not sure it's relevant
and clfft has #1112789 but it is marked as done, so it's worth
re-checking.

Lucas

[1]
with pseudo_exc_cmake as (
  select policy_info['autopkgtest'] as test_results from migration_excuses_experimental where source='cmake'
),
res1 as (
  select 
  split_part(d.key, '/', 1) as source,
  split_part(d.key, '/', 2) as version,
  d.value as result
  from pseudo_exc_cmake, jsonb_each(pseudo_exc_cmake.test_results) as d
  where d.key != 'verdict'
),
res2 as (
  select source,
  version,
  d.key as architecture,
  d.value->>0 as result,
  d.value as details
  from res1, jsonb_each(res1.result) as d
),
regressions as (
  select distinct source, version, architecture from res2 where result = 'REGRESSION'
),
cmake_bugs as (
  select * from bugs
  where title ~* 'cmake'
)
select distinct r.source, r.architecture, id, severity, status, title
from regressions r
left join cmake_bugs using (source);