On Wed, 9 Mar 2022 17:00:37 +0100 Hans Verkuil <hverkuil-cisco@xxxxxxxxx> wrote: > Hi Pekka, > > On 3/9/22 16:57, Pekka Paalanen wrote: > > On Wed, 9 Mar 2022 15:45:29 +0100 > > Hans Verkuil <hverkuil-cisco@xxxxxxxxx> wrote: > > ... > >> CI is already done: it's build every day together with the kernel media code > >> and v4l-utils in my daily build. Results of that are posted on the linux-media > >> mailinglist. > > > > Nice, but that is after merging patches, right? I was thinking > > pre-merge. > > Not sure what useful pre-merge testing can be done other than just running 'make' :-) To ensure that patches that are not intended to change the output indeed do not change the output, or do not introduce crashes. Other popular tests are checking for compiler warnings against a specific compiler version, and commit messages e.g. for S-o-b if that's used. For example wayland-scanner (a code generator) has tests in the Wayland test suite that verify the output for certain test inputs does not change. The test inputs and reference outputs are committed into git. If a patch intentionally changes the output, then that patch also includes changes to the reference output files. Simply reading a patch will also show how the output changed. But your test corpus is huge, and this method does not scale up to that directly. You could maybe have a few chosen test EDIDs for this, but running the full corpus needs something quite different. Maybe the full corpus is best the way you do it now. Thanks, pq
Attachment:
pgpiXE3rO6Ngn.pgp
Description: OpenPGP digital signature
