.buildbot-sourcedata sometimes contains unicode strings, causes full resync during builds

[dberger]

we don't have a perfect repro yet, but we noticed that some of our upgraded builders (0.7.11p3 -> 0.8.4p1) like to do full resyncs on force builds and after build failures. a newly created builder hasn't exhibited this problem.

we're using perforce.

after much tracking down, it looks like the common thread is that the misbehaving builders have u'/perforce/branch/path' in their buildbot-sourcedata file, rather than '/perforce/branch/path' - which suggests that argsbranch? is coming in unicode, and being written to the file that way.

If/as we learn more, I'll update the ticket.

[dberger]

we manually fixed a couple buildbot-sourceinfo files on windows builders, and they didn't force-sync on a force build. performing the same transformation on linux/mac builders didn't fix it, so there's more going on here.

[dustin]

Any further progress? Python will compare u'...' as equal to '...' (by casting the latter to unicode).

[dberger]

It won't when the comparison is against str([list, of, values]) and some of the values are unicode - you end up comparing '["foo","bar"]' against '["foo",u"bar"]' - and they're different. (this may be/is specific to the p4 code)

I believe we've gotten all the builders happy by manually modifying their on disk soucedata files - I'll confirm when I'm back in the office Monday.

[dustin]

Ah, yes, the u'..' gets encoded there, then. Hrm. Can you suggest a test and subsequent code fix for this? I don't think Buildbot has too many P4 users, but I'd hate for all of them to have to make this change manually.

[Dustin J. Mitchell]

Be consistent about string handling in Perforce's sourcedata string

Fixes #2040.

Changeset: 2fc90f01ba845db05f804d3d7e969a2b72a95da4