Hi all,

I was able to isolate what seems a darcs 2.0 regression in handling
manually moved files.

Darcs 1.0 handled the following "in the right way":

    $ mv a.txt subdir/
    $ darcs mv a.txt subdir/a.txt
    $ darcs record

resulting in a changeset like the following:

    Sun Apr 20 17:08:15 CEST 2008  lele@metapensiero.it
      * Third patch

         ./a.txt -> ./subdir/a.txt

Doing the same with darcs 2.0 I obtain the following instead:

    Sun Apr 20 17:00:53 CEST 2008  lele@nautilus.homeip.net
      * Third patch

         ./a.txt -> ./subdir/a.txt
        R ./a.txt

Darcs 2.0 behaves correctly if I rename the file using darcs itself,
that is if I simply do

    $ darcs mv a.txt subdir/a.txt
    $ darcs record

*without* manually moving it first, resulting in the following
changeset:

    Sun Apr 20 17:12:09 CEST 2008  lele@nautilus.homeip.net
      * three

         ./a.txt -> ./subdir/a.txt

To sum up:

    $ mkdir test
    $ cd test
    $ darcs init
    $ touch a.txt
    $ darcs add a.txt
    $ darcs record -a -m "First"
    $ mkdir subdir
    $ darcs add subdir
    $ darcs record -a -m "Second"
    $ mv a.txt subdir/
    $ darcs mv a.txt subdir/a.txt
    $ darcs record -a -m "Third"

works ok with darcs 1.0, but under darcs 2.0 produces a changeset
where "a.txt" is both moved to "subdir/a.txt" AND removed.

Instead the following

    $ mkdir test
    $ cd test
    $ darcs init
    $ touch a.txt
    $ darcs add a.txt
    $ darcs record -a -m "First"
    $ mkdir subdir
    $ darcs add subdir
    $ darcs record -a -m "Second"
    $ darcs mv a.txt subdir/a.txt
    $ darcs record -a -m "Third"

works ok producing the "same" changeset under both 1.0 and 2.0.

The "redundant" patch does *not* seem to cause any problem, that is,
both darcs 1.0 and 2.0 handle it without a noise. It was noticed
indirectly, because tools like the darcs plugin for Trac was unable to
understand it. I bet that also tools like tailor could be confused
when migrating this kind of patch to another VCs...

Thank you in advance,
bye, lele.
-- 
nickname: Lele Gaifax    | Quando vivrò di quello che ho pensato ieri
real: Emanuele Gaifas    | comincerò ad aver paura di chi mi copia.
lele@nautilus.homeip.net |                 -- Fortunato Depero, 1929.

I'm marking this as a bug, because of this:

         ./a.txt -> ./subdir/a.txt
        R ./a.txt

Whether or not darcs is supposed to recognise (a rm; darcs mv), it does not seem
very sensible for both of these patches to appear...

Lele: could I ask you to submit a formal test case as a shell script in the
bugs/ directory (i.e. a darcs patch).  You can inspire yourself from the other
tests in that directory

On Wed, 07 May 2008 16:25:07 -0000
Eric Kow <bugs@darcs.net> wrote:

> Lele: could I ask you to submit a formal test case as a shell script
> in the bugs/ directory (i.e. a darcs patch).  You can inspire
> yourself from the other tests in that directory

My time-machine worked right this time, you should probably already
have a bugs/issue803.sh... if not, it's a "darcs pull" away ;-)

ciao, lele.
-- 
nickname: Lele Gaifax    | Quando vivrò di quello che ho pensato ieri
real: Emanuele Gaifas    | comincerò ad aver paura di chi mi copia.
lele@nautilus.homeip.net |                 -- Fortunato Depero, 1929.

Perfect!  Now we just need somebody to fix it.

Assigning to David, who already is aware of this from the mailing list.