> By "flag" I mean _cache_generated_egg ("flag" as in Boolean value)
Ah, I had forgotten this earlier message:

> I get the opposite failure to Nadeem as far as InstallDataTestCase.test_resources: it works on
> Ubuntu 64-bit, but fails on 32-bit. Digging into it a bit further, I find that _generate_cache in
> Lib/packaging/database.py returns prematurely in the failing case, because _cache_generated_egg is
> True in the failing case but not in the test run which succeeds.

That it depends on the architecture currently baffles me.

> Ordinarily I'd submit a cpython patch, but in this case it's a one liner as Paul has suggested,
> so there's not much to review / comment on.
A patch in the list of files is much easier to find that a one-liner in a message.  Anyway, it’s not your fault.  Antoine was kind enough to get my attention on the buildbot failures; to make this manageable, I will open separate reports with specific names for each different bug.

> Here's the change to packaging.manifest: [snip]
> and the change to distutils.filelist: [snip]
Thanks.  Luckily, these modules recently gained full test coverage, so I will be able to commit the fixes and feel safe.

> You'll see I used a different idiom to Paul in my fix :-)
I’m a big fan of format strings other string concatenation, too.

> Can the distutils/packaging duplication not be avoided?
No.  They are independent modules.  distutils will die; packaging will be improved and cleaned up.  However, contributors can work on packaging only and leave the gruesome backporting work to me.

> IMO the correct cpython fix would address this.
I don’t understand.