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

Gilles Moris gilles.moris at free.fr
Thu Feb 12 16:40:51 CST 2009


On Thu February 12 2009 21:21:49 Martin Geisler wrote:
> Gilles Moris <gilles.moris at free.fr> writes:
> > Why not:
> >     sys.path.insert(0, 'mercurial')
> >     sys.path.insert(0, os.path.join('mercurial', 'pure'))
> > to have [ 'mercurial/pure', 'mercurial' ] in front.
> > Otherwise, it takes the site-wise mpatch instead of the local pure.
> 
> Does that make a difference? The old code spawned 'hg id' and so it
> would also have used the site-wise mpatch.
> 

Just that internal API are supposed to be less stable than CLI across versions.

[...]
> But the pure code should definitely work! The other patch I sent makes
> the code more robust on Windows and Mac OS X, but your error occurs on
> Linux and it is different...
> 
> Can you run the test suite with the pure code? Go to tests and do
> 
>   ./run-tests.py --pure
> 

Wouah !

./run-tests.py
0 test failed
./run-tests.py --pure
187 tests failed

Most of them seems located in the mmap.move().
I think my python 2.4.3 on FC5 has a serious flaw in mmap. 0 test failed
on Fedora 8 with --pure.

I was not counting using pure implementation anyway.

Thanks for your help, Martin.
Gilles.


More information about the Mercurial-devel mailing list