The 1.7 freeze has started!

[Steve Borho]

On Thu, Oct 21, 2010 at 8:10 AM, Adrian Buehlmann <adrian at cadifra.com> wrote:
> On 21.10.2010 15:00, Dan Villiom Podlaski Christiansen wrote:
>> On 21 Oct 2010, at 06:57, Steve Borho wrote:
>>
>>> Ok, after a rather painful bisect session of:
>>>
>>> * get a revision from bisect
>>> * local tag it, build an installer, install it, smoke test
>>> * if good, bisect -g and repeat
>>> * if bad, install known good package, bisect -b, repeat
>>>
>>> bisect says:
>>>
>>> The first bad revision is:
>>> changeset:   12686:fe31f834a9ff
>>> user:        Dan Villiom Podlaski Christiansen <danchr at gmail.com>
>>> date:        Tue Sep 14 23:00:39 2010 +0200
>>> summary:     demandimport: fix an obscure corner-case.
>>>
>>> And indeed if I back out this change on the tip of stable, I can make
>>> a working installer package again.
>>
>> Although I'll readily admit to not completely understanding how
>> demandimport & __import__ work, I still think the change was correct.
>> Is it possible we're running into some odd bug with the zipimporter?
>> Could you try to build TortoiseHg without zip archiving?
>
> It would be a royal PITA to maintain the wix installer sources for a
> Windows installer (*.msi) which does that.
>
> There are potentially also issues with path funnies by tons of dll's
> suddenly being exposed on PATH (which are now nicely hidden inside the zip).
>
> BTW, this most likely affects the plain mercurial msi installer as well
> (not just TortoiseHg).

Yes, it did.  It make tracking down the problem almost sane.

>> Anyway, it's probably best to back the changeset out for now…
>
> If this is possible, yes please.

This bug is even odder than that.  I have been building functional MSI
packages on the tip of the default branch this whole time (for the Qt
nightlies).  It is only on that one machine, that I use to generate
the official GTK installers, that this dies.  So I think it's just the
zipimporter from that specific version of Python (and perhaps earlier)
that does not like the keyword argument.  I don't have access to the
machine now, but I'm fairly certain it has the oldest 2.6.X revision
of any of the machines I build installers on.

If danchr doesn't feel too strongly about keeping it, I'll just back it out.

-- 
Steve Borho