Re: [Quilt-dev] fix patch version check in configure.ac

[Andreas Gruenbacher]

On Tuesday 24 January 2006 02:43, John Vandenberg wrote:
> On 1/24/06, Andreas Gruenbacher <address@hidden> wrote:
> > On Monday 23 January 2006 22:23, Dean Roehrich wrote:
> > > On Mon, Jan 23, 2006 at 09:44:08PM +0100, Andreas Gruenbacher wrote:
> > > > I still need to fix ftw by either using nftw or something else; then
> > > > we should be ready for 0.43.
> > >
> > > The "something else" that I'm using today is sun-backup-files-2.diff
> > > from http://zeroj.hda0.net/quilt-0.43-patches-v5/
> > >
> > > This appears to be happy on Solaris and linux.  I did have one question
> > > about it, though.  Perhaps the addition of the 'rm -rf' in push.in
> > > should be joined with the preceding backup-files with an &&, as was
> > > done in pop.in?
> >
> > It probably should, but I'd really prefer fixing backup-files instead. It
> > does matter how many programs are forked/exec'd at this place, and it's
> > just unclean for backup-files to leave around trash that it created
> > before.
>
> In order to get some idea of how expensive a a bash only
> implementation would be, I have roughly drafted up a patch that
> implements backup-files -r.
>
> http://zeroj.hda0.net/quilt-0.43-patches-v5/sh-restore-2.diff
>
> There are two exec's per patch, and for push, one exec per file; for pop,
> two.

And a lot of slow shell code. Do you have cpio on all systems? Then files can 
maybe removed with "find -size 0 | xargs -r rm -f" and restored with cpio 
pass-through? But that's probably just as bad.

> While this is probably too expensive for linux kernel development, it
> certainly wouldn't affect me with only a few small patches per series.
>  Could we introduce a bash only implementation for platforms without
> nftw?

I would still prefer to fix backup-files a hundred times.

Andreas