pkg/vcs: use committer date than author date

[KumanekoSakura]

The Freshness columns in Instances: table on the dashboard page look outdated, for
these fields are showing when that patch was authored. Showing when that patch was
committed into the tree in question would be more meaningful.

[codecov]

Codecov Report

Merging #2136 into master will increase coverage by 0.0%.
The diff coverage is 44.4%.

Impacted Files	Coverage Δ
pkg/vcs/vcs.go	43.5% <ø> (ø)
syz-ci/manager.go	0.5% <0.0%> (ø)
syz-ci/updater.go	0.0% <0.0%> (ø)
pkg/vcs/git.go	62.4% <57.1%> (-0.2%)	⬇️
pkg/host/machine_info.go	55.6% <0.0%> (+55.6%)	⬆️

[dvyukov]

The date is used in multiple places, not just the Freshness column. It's also used to show commits (in the "git show" style), which needs to use the author date. Longer term we need both dates. So this change needs to obtain both, add CommitDate to the Commit struct, propagate to the dashboard, etc.
Please add Update #1537 to the commit description, that's the issue for this problem.

[dvyukov]

Nice!

By grepping for SyzkallerCommitDate and KernelCommitDate I found few other places that need to be updated as well.

		SyzkallerCommitDate: prog.GitRevisionDate,
./syz-ci/manager.go

This comes from GITREVDATE in Makefile. I think we can just change it to %cd as it's not used for anything else.

			resp.Build.KernelCommitDate = res.Commit.Date
	resp.Build.KernelCommitDate = kernelCommit.Date
./syz-ci/jobs.go

In both of these case we can now use CommitDate. Otherwise Build.KernelCommitDate will have different values for different builds in the database and it will be hard to make sense of these dates.

[KumanekoSakura]

Thank you for the comment. I assumed that

--format=%H%n%s%n%ae%n%an%n%ad%n%P%n%cd%n%b

makes

bytes.Split(output, []byte{'\n'})

in gitParseCommit() to return at least 8 elements (and therefore len(lines) < 8 is the right threshold to test). Am I correct?

[dvyukov]

in gitParseCommit() to return at least 8 elements (and therefore len(lines) < 8 is the right threshold to test). Am I correct?

Yes, I think so.

I see both callers are covered by tests, so we should be good if tests pass.

syzkaller$ go test -coverprofile=coverprofile ./pkg/vcs
ok  	github.com/google/syzkaller/pkg/vcs	1.995s	coverage: 66.3% of statements
syzkaller$ go tool cover -html coverprofile