This script demonstrates the bug:

----------------------
#!/bin/sh
set -v
rm -rf test
mkdir test
cd test
darcs2 --version
darcs2 init
echo test >File.hs
darcs2 add File.hs
darcs2 record File.hs -a -m "add File"
rm File.hs
darcs2 record -a -m "rm File"
darcs2 cha
darcs2 unrecord -p "rm File" -a
darcs2 cha
darcs2 record -a -m "re-rm File"
----------------------

Here's the output:

rm -rf test
mkdir test
cd test
darcs2 --version
2.0.2 (+ 70 patches)
darcs2 init
echo test >File.hs
darcs2 add File.hs
darcs2 record File.hs -a -m "add File"
Recording changes in "File.hs":

Finished recording patch 'add File'
rm File.hs
darcs2 record -a -m "rm File"
Finished recording patch 'rm File'
darcs2 cha
Mon Aug 18 10:26:59 BST 2008  Simon Marlow <marlowsd@gmail.com>
  * rm File

    R ./File.hs

Mon Aug 18 10:26:59 BST 2008  Simon Marlow <marlowsd@gmail.com>
  * add File

    A ./File.hs
darcs2 unrecord -p "rm File" -a
Finished unrecording.
darcs2 cha
Mon Aug 18 10:26:59 BST 2008  Simon Marlow <marlowsd@gmail.com>
  * add File

    A ./File.hs
darcs2 record -a -m "re-rm File"

darcs failed:  Cannot remove non-empty file ./File.hs
Your repository is now in an inconsistent state.
This must be fixed by running darcs repair.

Failed to record patch 're-rm File'

I have submitted this as bugs/issue1012_unrecord_remove.sh.  Thanks!

By the way, if you are so inclined, submitting bugs as patches is easy.  Just
stick the shell script in bugs/ add, record and send!  There is virtually no
bureaucracy to put in the scripts either.  Bash and set -ev is all you need :-D
[oh and don't worry about names either, we can take of that stuff]

John: does darcs repair allow you to move on?

I ran darcs repair, and it seemed to work, but it took at least a few hours (I 
had to leave it running overnight). But when I tried to record the last patch 
again, it failed with the same error.

I had another copy of the repository from which I moved on.

Is this a dupe of issue693?  msg3418 suggests that that issue also stems from
unrecording.

On Tue, Aug 19, 2008 at 23:18:13 -0000, Nathaniel Filardo wrote:
> Is this a dupe of issue693?  msg3418 suggests that that issue also stems from
> unrecording.

This appears to be a different bug.

I ran the test this morning by creating a file bugs/tests_to_run with
the line
  issue1012_unrecord_remove.sh

I then went into bugs-shell-old-fashioned.dir/test (the test directory
is created by the issue1012 script) and performed a darcs check.  The
repository was consistent; no removal-of-nonempty-file patches there.

Interestingly, the test passes on hashed and format-2 repositories.

The following patch updated the status of issue1012 to be resolved:

* resolve issue1012: it seems to be fixed by the fix for issue709.