Re: [Quilt-dev] quilt on SuSE 7.2 and Windows

[John Vandenberg]

Hi Andreas,

Thanks for taking the time to consider this.

On 7/25/05, Andreas Gruenbacher <address@hidden> wrote:
> Hello,
>
> On Sunday 24 July 2005 17:19, John Vandenberg wrote:
> > Hi,
> >
> > I have been tinkering with quilt in order to use it on a SuSE 7.2
> > development server and on Windows using msys.  Here is a list of the
> > problems, mainly in the test execution rather than the quilt code.
>
> Okay, test suite bugs bother me less wrt. backward compatibility.
>
> > GNU fileutils 4.0.35
> > --------------------
> > The output of ls -l appears to have changed over the years, causing an
> > error in perms.test on SuSE 7.2.
>
> What changed? This may have to do with the setting of LANG, LC_ALL, or
> similar.

hmmm.  I have compared the output of SuSE 7.2 and FC4, and they are
the same.  maybe this problem was actually an gawk issue.  here is the
change I needed:

http://zeroj.hda0.net/quilt-patches/quilt-fileutils-4.0.35.diff

I'll investigate this problem further tonight.

> > gawk 3.0.6
> > ----------
> > the GNU function asort used in patchfns.in is not available.
>
> Does the old gawk have any sorting mechanism? Otherwise we could use sort
> instead.

The asort code doesn't appear to be necessary:

http://zeroj.hda0.net/quilt-patches/quilt-gawk-3.0.6.diff

> > sed 3.02
> > --------
> > The -i option used in annotate.test is not available.
> > Earlier versions of sed do not interpret \t as 0x09, causing many
> > failures in the test suite.
> > example1.test uses the 4a command, which doesn't appear to be supported.
>
> We could use ed etc. instead.

msys doesn't have ed.  Using different sed syntax works:

http://zeroj.hda0.net/quilt-patches/quilt-sed-3.0.2.diff

> > diff 2.7
> > --------
> > When one of the files doesn't exist, diff 2.7 prints an error on
> > stderr, while later versions operate normally as if the file was
> > empty.
> > (annotate.test line 84 fails)
>
> can you run the test with ``./run -l 84'', and then send me the result of
> running ``quilt --trace annotate jkl''?

here it is: http://zeroj.hda0.net/quilt-annotate-diff-2.7.log

to work around this, I found it was easier to fake the missing file:

http://zeroj.hda0.net/quilt-patches/quilt-diff-2.7.diff

> > bash 2.04
> > ---------
> > diff.in uses some funky pipe syntax that was introduced in 2.05.
>
> "< <(...)" you mean?

that's it.

http://zeroj.hda0.net/quilt-patches/quilt-bash-2.04.diff

> > patch 2.5
> > ---------
> > the output of 2.5 and 2.5.4 are very different.  regex support was
> > added to test/run to allow the tests to handle this in order to
> > minimise the number of changes to quilt/*
>
> Hm... what did you change?

quilt/test/run and the test cases.

http://zeroj.hda0.net/quilt-patches/patch-2.5.diff
- Show quoted text -

> > msys 1.0.11
> > -----------
> > msys is missing column, tac, getopt, and ed, which is indirectly used
> > by patch in annotate.in.  EOL issues abound, and the backup-files
> > restoration command line in push.in and pop.in trigger trigger the
> > msys shell to replace the trailing '-' with '-<current path>'.
> >
> > For anyone interested in trying this madness, column, getopt, and a
> > few other binaries from util-linux can be found here:
> > http://zeroj.hda0.net/util-mingw/
> >
> > A semi-complete binary tarball:
> > http://zeroj.hda0.net/quilt-0.41-bin.tgz
> >
> > and my first patch series:
> > http://zeroj.hda0.net/quilt-patches/
> >
> > The tests that fail are:
> > annotate.test: missing ed
> > dotglob.test: not sure
> > perms.test: umask is not well respected on Windows
> >
> > Are the quilt maintainers interested in accepting patches to support
> > older & stranger platforms like these?
>
> As long as the patches don't badly mess up the code or slow things down too
> badly, I think we will accept UNIX patches, yes.
>
> Porting quilt to Windows sounds more like an exercise in masochism: quilt
> relies on the UNIX toolchain  a lot and Windows has much slower process
> startup times, so quilt will never run well as far as I can see. I'm not sure
> we should even bother. I'll surely take obvious things, and things that won't
> hurt otherwise.

It is now usable on my Windows box, so my itch is done.  the patch
quilt-msys.diff is a bit messy; I don't mind braking it down further
to assist reviewing.

I wanted to try quilt out on Windows in order to break down a rather
large patch that I am maintaining to port APR to MinGW.  Previously I
was creating messy patches using CVS or SVN, and manually manipulating
the output to bring order to the patch.  quilt looked like the best
alternative, and now that I have used it a little, I am sold -- its
great.

Maybe, Redmond based kernel hackers will find it useful too. :)

btw, using quilt is snappy to use on my 400Mhz W2K box with 128Mb RAM;
YMWillV.

Cheers,
John