Incremental: ensure we add post-compile jobs if no new compile jobs are found

[nkcsgexi]

rdar://74447605

[nkcsgexi]

@swift-ci please test

[artemcm]

LGTM, @davidungar what do you think?

[nkcsgexi]

@swift-ci please test

[davidungar]

Thank you for looking at this issue

I suspect that, with this fix, the post-compile jobs would get to run before the compile-pch job finished.

I'm starting to wonder if a better fix is to remove the early return from addRuleBeyondMandatoryCompiles and instead to put an early return in collectJobsDiscoveredToBeNeededAfterFinishing , returning empty if a non-compile job?

[davidungar]

Hate to say it, but I'm now waffling.
Here's what I'm thinking... in a non-incremental build, maybe there would be nothing to sequence the jobs other than llbuild input-output matching. (Is that right?) So, could all the pre-compile jobs just get run in parallel with the post-compile jobs modulo inputs and outputs? If so, your fix is better: unfinishedJobs -> unfinishedCompileJobs.

[nkcsgexi]

Ok, let's merge this for now. @davidungar any advice for adding test for this change?

[davidungar]

Ok, let's merge this for now. @davidungar any advice for adding test for this change?

Good question, @nkcsgexi . I guess we need a test that creates a precompiled header. I would look at testIncremental to see what that does. Much of that machinery should be reusable.

[davidungar]

One last thought: with this change are we guaranteed that the whole driver process won't terminate before the pch job?

[artemcm]

One last thought: with this change are we guaranteed that the whole driver process won't terminate before the pch job?

I think this job is always going to be a part of the initial set of jobs we submit to llbuild, (not one discovered when other jobs finish), so the ExecuteAllJobsRule should ensure it actually gets run before we return from MultiJobExecutor::execute.

[davidungar]

One last thought: with this change are we guaranteed that the whole driver process won't terminate before the pch job?

I think this job is always going to be a part of the initial set of jobs we submit to llbuild, (not one discovered when other jobs finish), so the ExecuteAllJobsRule should ensure it actually gets run before we return from MultiJobExecutor::execute.

Thanks! Sounds good to me.