Hello Guillem,
Guillem Jover [2010-06-02 20:43 +0200]:
> Hmm, while testing and fixing few things here and there I got a bit
> absorbed and ended up touching most of the patch. So I've just taken
> it over (mainly given the amount of changes I've made to it). I've
> added a Signed-off-by line on your behalf, please tell me if that's
> fine with you.
Of course, thanks a lot for cleaning it up!
> It might be a bit confusing as “dpkg -L pkg” will list all files even
> if not in the file system, but that's equivalent to the admin removing
> those after the fact, so I think it's fine. It will also allow in the
> future to ask dpkg which files are missing for example.
Hm, did that behaviour change in your patch then? With mine, the .list
file that dpkg writes after unpack does not contain the skipped files,
and the test suite patch that I sent also verifies that (it calls dpkg
-L to check for present/absent files).
> > + /* We need to keep directories if a glob excludes them, but a more
> > specific
> > + * include glob brings back files; this will probably create more
> > directories
> > + * than necessary, but better err on the side of caution than failing
> > with
> > + * "no such file or directory" (which would leave the package in a very
> > bad
> > + * state). */
> > + if (remove && ti->Type == Directory) {
>
> There's also the possible case of symlinks pointing to directories, so
> it's catching those too now.
Ah, nice catch, thanks.
> And yes, ideally this would get fixed in the future to be more
> precise, the problem is that I can only think about one way to do
> that, with a two pass algo over the tarball.
Either that, or installing a file would create its directory instead
of relying on it being created previously. Would you see anything
wrong with that?
Thanks,
Martin
--
Martin Pitt | http://www.piware.de
Ubuntu Developer (www.ubuntu.com) | Debian Developer (www.debian.org)
--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]
