The last couple of times I ran `darcs pull' in my local copy of the
repository at <http://darcs.net/unstable/>, in which I have never done
anything but `darcs pull' and non-destructive commands such as `darcs
changes', Darcs made some backup files.  When I pulled some weeks ago,
it made Setup.hs-darcs-backup0 and tests/EXAMPLE.sh-darcs-backup0;
when I pulled few minutes ago, it made src/Preproc.hs-darcs-backup0.

At <http://paste.lisp.org/display/85602> is a transcript of my recent
update.  I ran `darcs revert' after I saw the original backup files,
and wanted to ensure that any edits I had made were reverted before I
pulled (although I am pretty sure that I had never edited any files in
the repository).

Eric Kow suggested that this was because those files had been renamed,
but with changes only to the case of the characters in their names,
and I am using Darcs on a case-insensitive file system (HFS+).  Thus
Darcs saw that there already was a file by the target name and backed
it up before writing the same file again.  So this seems harmless,
although slightly confusing.  Perhaps Darcs could detect this
situation and either not back up the file on case-insensitive file
systems, or, to be safer, just check whether the two names are
equivalent under case-folding, and if they are, say so when emitting a
message about backing the file up.

> just check whether the two names are
> equivalent under case-folding, and if they are, say so when emitting a
> message about backing the file up.

This is actually very easy to do in Darcs since we already require such a check
in Darcs.IO

Consider this patch:

Fri Jun 13 08:02:13 BST 2008  Eric Kow <E.Y.Kow@brighton.ac.uk>
  * Fix a backup bug on case-insensitive file systems.
  
  Avoids this scenario.  The task: move foo to FOO and modify FOO
  1. FOO (foo) already exists, so I'll back it up
     (moves FOO [foo] to FOO.darcs-backup0)
  2. Now I can move foo to FOO
     foo no longer exists, but that's ok
     (nothing happens)
  3. Now I can FOO
     hey, where did FOO go?
     (crash!)

So pretty much everything is in place, we need only do a little IO or perhaps
extend the backup functions' IO