Re: [Quilt-dev] Bug#333480: quilt: no way to avoid patch -E option

[Joe Green]

Martin Quinson wrote:

> Another bug I received in the debian BTS a while ago. I'm really unsure
> about what to do here.
>  
I proposed moving the "-E" option to QUILT_PATCH_OPTS a few months ago:

   http://lists.gnu.org/archive/html/quilt-dev/2005-08/msg00007.html

In the ensuing discussion, I found that I couldn't get diff to generate 
output that represents adding an empty file, though this should 
theoretically be possible.  More importantly, if I generated a 
reasonable representation directly (bypassing diff), patch wouldn't 
accept it and create a file.

So, in practice quilt has more problems with handling empty files than 
just how the "patch -E" option is handled.

> Thanks for any advice,
> Mt.
>
> ----- Forwarded message from Rob Browning <address@hidden> -----
>
> Subject: Bug#333480: quilt: no way to avoid patch -E option
> X-Debian-PR-Package: quilt
> From: Rob Browning <address@hidden>
> To: address@hidden
> Date: Tue, 11 Oct 2005 23:29:52 -0700
> User-Agent: Gnus/5.110004 (No Gnus v0.4) Emacs/21.4 (gnu/linux)
>
>
> Package: quilt
> Version 0.42-1
>
> Quilt currently appears to unconditionally insert -E as a patch
> argument (see /usr/share/quilt/patch).  This causes it to remove files
> that end up with a size of 0 after patching.
>
> The patch manpage suggests that -E isn't normally necessary, and it
> can cause trouble.
>
> For example, after updating all of the autofiles for emacs21,
> aclocal.m4 will have been reduced to a size of 0, but it still exists,
> and the emacs Makefile depends on it existing.  If the resulting
> autofiles diff is represented as a quilt patch, then when it is
> applied, aclocal.m4 will be removed (rather than left with a size of
> 0), and the emacs build will fail.
>
> There may be some reason that quilt depends on this behavior, but if
> not, it would be nice if there were some option to control the use of
> -E when calling patch.  It would be ideal (though perhaps not
> feasible), if the use of -E could be specified per-patch, as with -p,
> but a global control, or additional argument to quilt push, would be
> sufficient.
>
> Also, I listed this bug at normal severity because I wasn't sure
> that it wasn't a real bug, but it might also just rate wishlist.
>
> Thanks
>  
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Quilt-dev mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/quilt-dev
>  


--
Joe Green <address@hidden>
MontaVista Software, Inc.