ci: Set IS_MERGE_PIPELINE for non-Marge MRs
When IS_MERGE_PIPELINE is set, we will apply a separate set of pre-merge skips to our dEQP and Piglit baselines. But this was only being set for Marge's pipelines, not for non-Marge pipelines, meaning that merge requests would fail when run by a user and then mysteriously go green when Marge applied them.
Fix this for now by also setting $IS_MERGE_PIPELINE for all MR pipelines.
Signed-off-by: Daniel Stone daniels@collabora.com Fixes: 1dc64d06 ("ci: Use merge-skips files during merge pipelines") Fixes: ee0d7822 ("{freedreno,intel}/ci: rename "premerge-skips.txt" to "merge-skips.txt" to accurately reflect what they are")
cc @eric @gallo - this branch is called fdno-slow-fp64
because I was going to apply some skips for !31796 (merged) after seeing a630_gl fail and wonder why, then looked at proximate MRs and saw there was nothing to explain it, and only much later figured out they were in the pre-merge skips.
But I still don't think this is the right thing, because it means that you'll get completely different job results between running a pipeline in a branch (e.g. with ci_run_n_monitor) and running a pipeline in an MR. Not everything is always ready to be an MR, but getting different failure results seems quite weird. Perhaps we should always add the pre-merge skips unless the job is specifically flagged as a 'full' job, or the user has requested it through the environment? And doing that should probably increase the timeout multiplier as well.
In the meantime, I'm pretty sure this is going to break ci-uprev completely, because it will now be getting different results for its jobs vs. what would actually run in an MR.