[PATCH 2 of 2] setup: call identify directly from mercurial.commands

[Gilles Moris]

On Fri February 13 2009 10:03:47 Martin Geisler wrote:
> > It fails as well, with the same error:
> > ValueError: source or destination out of range.
> 
> Hmm, could you try with different source and destination offsets to
> see if anything works at all?
> 

Nothing at all: even move(1,0,1) traceback the same error.

> > However, you will found the following even more interresting.
> > I had recompiled several versions of python in addition of the standard
> > python 2.4.3 installed by default. I have repeated the same commands:
> > - python 2.3.6 recompiled: fails
> > - python 2.4.3 FC5: fails
> > - python 2.4.4 recompiled: succeeds
> > - python 2.5.2 recompiled: succeeds
> >
> > Note that this is 64 bits python binaries. It could be worth testing
> > on 32 bits version prior 2.4.4 if it's working.
> 
> I've now tested it succesfully on Python 2.4.3 on 32 bit Red Hat
> Enterprise Linux Client release 5.3 (Tikanga).
> 
> I then tested it succesfully on Python 2.2.3 and Python 2.3.5, both
> running on 32 bit Debian 3.1.
> 
> So I believe something is wacky about your system :-/
> 

I really think this is 64 bits before 2.4.4. I recompiled the same 2.3.6
in 32 bits, and it works. They might have fixed a 64 bits bug in 2.4.4, but
I see nothing in the release notes.

Now I don't know if it's worth pushing the investigation. The problem is in
the pure fallback code, so I will patch the version file by hand after the
build. We can look to get a fallback on stringIO, but that's not a priority
for me for 1.2.

Thanks.
Gilles.