0.9.5 release plans

Jim Hague jim.hague at acm.org
Thu Oct 25 04:06:15 CDT 2007


On Wednesday 24 October 2007 16:22:54 Matt Mackall wrote:
> On Wed, Oct 24, 2007 at 02:43:50PM +0100, Jim Hague wrote:
> > > I haven't got to the bottom of the 'hg serve' failure yet; still
> > > looking.
> >
> > OK, found the problem. It's the venerable Python version 2.3.4, still the
> > latest distributed by IBM. There's a likely-looking bz2 flushing bug in
> > there marked as fixed in 2.3.5. I've build Python 2.3.6 on AIX and hg's
> > test-archive works fine with that Python version.
>
> Thanks! If there's a simple work-around for 2.3.4, we'd be interested.

I'm no Python whiz, but I'm afraid I don't think there is. The fix in 2.3.5 is 
correcting the flushing on an internal class in the Python tarfile.py module; 
I can't see any obviously way of prodding it into happening from outside 
without fiddling directly with the guts of the tarfile module.

Having said that, it turns out that the bug only manifests itself on very 
small repos - like the test-archive generated repo of 3 files of a few bytes 
each. On more realistic size archives, like the Mercurial archive, everything 
works fine. In other words, requests for a .bz2 archive will work most of the 
time, and if they don't then you always have the workaround of asking for .gz 
instead; it is definitely only .bz2 that has the problem.

I am much more inclined to put my energies into using the IBM source RPM info 
(they publish .spec and patches) to package 2.3.6 than perpetrating outrages 
on standard Python modules :-) As time permits I'll publish an AIX Mercurial 
RPM, though sadly it probably won't have the Subversion importer; Subversion 
isn't packaged for AIX, it would be a big job packaging it, and it won't 
currently build either.
-- 
Jim Hague - jim.hague at acm.org          Never trust a computer you can't lift.


More information about the Mercurial mailing list