Re: [Quilt-dev] [BUG] 'quilt diff' shows text from wrong file - more info

[Chris Shoemaker]

On Tue, Oct 03, 2006 at 10:06:39PM +0200, Andreas Gruenbacher wrote:
> Hello,
> 
> On Tuesday 03 October 2006 21:37, Chris Shoemaker wrote:
> > On Tue, Oct 03, 2006 at 02:13:14PM -0400, Chris Shoemaker wrote:
> > > First off, I'm a long-time happy quilt user.  Today, I saw something
> > > strange enough that I thought I should report it.  Sorry I don't have
> > > an actual reproduction script.  If it's absolutely required, let me
> > > know.
> > >
> > > Here's what happened:
> > >
> > > [These first few steps may be unimportant, but I include them just in
> > > case.]
> > >
> > > I'm in the middle of my patch stack.  The next patch, call it
> > > 'sed.diff', is the result of a sed on my tree.
> > >
> > > $ quilt push
> > >
> > > [ The push fails because I've made changes to eariler patches.
> > > Instead of forcing it and fixing it up, I'm just going to re-run the
> > > saved sed script. ]
> > >
> > > $ quilt delete sed.diff
> > >
> > > [ Perhaps I should have used -r here, but I never do. ]
> > >
> > > $ quilt new sed.diff
> > >
> > > $ quilt add `cat saved-list-of-files`
> > >
> > > $ sed -f script.sed -i `quilt files`
> > >
> > > Now I want to see what I've done before I `quilt ref`.
> > >
> > > $ quilt diff -z |less
> > >
> > > Here's where it gets weird, and I had to triple-check that I was
> > > actually seeing what I was seeing.
> 
> The above diff is between the old version of the patch (which is still in 
> patches/sed.diff after the delete, and as far as the original patch applies) 
> and the working tree. What you really want to do here is a simple `quilt 
> diff'.

Aha!  I had expected that `quilt new sed.diff` would start a _new_
empty patch.  That's why I followed it with `quilt add ...`, and
that's why I expected `quilt diff` to be the same as `quilt diff -z`.

I guess `quilt delete mypatch && quilt new mypatch` doesn't actually
give me the new empty patch I want.  This might be less disturbing if
`quilt new mypatch` gave some output like "Re-using existing patch."

> 
> > > The diff looked just as I expected, with the right files and hunks and
> > > line numbers, except for one important difference.
> > >
> > > Most of the changed lines from the original version (those starting
> > > with '-') were not the actual text from the original version, but the
> > > text from the changed version.
> > >
> > > IOW, if my sed script had "s/QQQ/XXX/g" in it, I saw lines like:
> > >
> > > @@ -554,7 +554,7 @@
> > >
> > >
> > >
> > > -  foo XXX bar
> > > +  foo XXX bar
> > >
> > >
> > >
> > >
> > >
> > > And this wasn't just one hunk, but almost every hunk was in the
> > > 'correct' place, but showed identical lines.
> 
> Probably the old and the new versions of the files have some whitespace 
> changes --- otherwise, diff wouldn't generate such hunks.

That's quite likely, since the _old_ patch would have stripped
trailing whitespace when it was refreshed, but the new changes
wouldn't reflect this.

> 
> Could you please check very carefully?
> 
> > > A little baffled, I decided to `quilt ref` anyway.
> > >
> > > Afterward, `quilt diff |less` showed the correct diffs.
> > >
> > > Can anyone explain this?  It seems like a bug to me.
> >
> > More info:
> > $ quilt --version
> > 0.45
> >
> >
> > I can easily repeat this.  Also, this is probably a BIG HINT:
> >
> > When I get the weird `quilt diff -z`, there are 2 of the 20 files for
> > which the diff shows the correct difference.  Those two files are the
> > same two files that failed to apply during the first `quilt push`.
> > Incidentally, I use QUILT_PUSH_ARGS="--leave-rejects" in my .quiltrc
> > file.
> 
> In case you still run into problems, please send a test case. (You can use 
> the 
> attached test case as a starting point; run with the test/run script in the 
> quilt tarball.)

In the future, I'll just keep in mind that `quilt new patch` might
give me a patch containing stuff from some other old patch with that
same name.

Thanks.

-chris

> 
> Andreas