Re: [Quilt-dev] quilt and debian (was something else)

[Jean Delvare]

Le lundi 23 novembre 2009 17:05, Greg KH a écrit :
> On Mon, Nov 23, 2009 at 03:57:53PM +0100, Jean Delvare wrote:
> > I still believe we want this shell-based backup-script, no matter how
> > high the performance penalty is. quilt was written in bash after all,
> > which pretty much implies that we do no care about performance;
> 
> Um, no, I would disagree.  I deal with a tree that is almost always over
> 300 patches, and push and pop all of them many, many, times a day.
> Making 'quilt pop' 3 times slower would drive me crazy, I already feel
> it's a bit too slow in places...

Feel free to rewrite quilt entirely in C then, I'm sure it would be
much, much faster :)

After my cleanups, the slowdown factor for "quilt pop" was "only"
2.4. But it was a long time ago, I don't know if these numbers are
still accurate. With some more work I hope we can get it down to 2
or less.

One thing I had in mind was turning backup-files into a shell
function rather than a standalone shell script... I'm fairly certain
that the cost of starting a separate bash instance is significant so
we could save on this. Not sure how to deal with getopt then though.

And I can't disagree entirely with you, I find "quilt pop" slow as
well. Note that "quilt pop -af" is much faster, and in many cases is
equivalent to "quilt pop -a". But of course it's easy to lose changes
if used inappropriately. Another trick is -q, which makes the output
less verbose, that can make a significant difference with a slow
display and/or remote connection. You might want to add
"QUILT_POP_ARGS=-q" to your ~/.quiltrc.

-- 
Jean Delvare
Suse L3