Darcs should probably use ${TMPDIR:-/tmp} for temporary files.  It doesn't:

$ darcs send -o /dev/null -a --dont-sign
Creating patch to "http://darcs.net/unstable"...
Waiting for Emacs...
darcs: //dev/null-0: takeFile //dev/null-0 in /home/twb/VCS/darcs-doc.~1~: openFd: permission denied (Permission denied)

> Darcs should probably use ${TMPDIR:-/tmp} for temporary files.  It doesn't:

No, it shouldn't.  Darcs will sometimes rename a temporary file over a real
file, for that to work efficiently the temp file needs to be in the same
directory as the file it's going to be removed over.

                                        Juliusz

On Sun, Oct 26, 2008 at 09:00:53AM -0000, Juliusz Chroboczek wrote:
>> Darcs should probably use ${TMPDIR:-/tmp} for temporary files.
>
> No, it shouldn't.  Darcs will sometimes rename a temporary file over
> a real file, for that to work efficiently the temp file needs to be
> in the same directory as the file it's going to be removed over.

I'd rather it worked inefficiently than not at all.  I believe the
transcript demonstrates that the "efficient" method at least needs to
be taught about edge cases, because its assumptions are not always
valid.

In this case, the assumption darcs makes is that it has the right to create its
output file.  I think this is a reasonable assumption.

David

On Sun, Nov 02, 2008 at 11:53:17PM -0000, David Roundy wrote:
>> $ darcs send -o /dev/null -a --dont-sign
>> Creating patch to "http://darcs.net/unstable"...
>> Waiting for Emacs...
>> darcs: //dev/null-0: takeFile //dev/null-0 in /home/twb/VCS/darcs-doc.~1~: openFd: permission denied (Permission denied)
>
> In this case, the assumption darcs makes is that it has the right to
> create its output file.  I think this is a reasonable assumption.

Rather, it's making the assumption that it has the right to create
ANOTHER file in the same directory as the output file.  This is not
the case for /dev/.

It could be nice to fix this trying first to create it in the same place as the
real file and backing off if we can't.  I don't know if this will be difficult
to do or not though.  Have a look at Darcs.Lock

On Sun, Aug 23, 2009 at 07:46:08PM +0000, Eric Kow wrote:
> It could be nice to fix this trying first to create it in the same
> place as the real file and backing off if we can't.  I don't know if
> this will be difficult to do or not though.  Have a look at
> Darcs.Lock

While the implementor is at it, I'd also like

    darcs24735-new_87c87c6d663850ba77e6e5e50b716c8e
    darcs-temp-mail-0.~1~
    darcs-record-0.~1~

to end up in either ${TMPDIR:-/tmp} or (if you're concerned about
"performance") in _darcs/.  Currently they're in my working directory,
the top of the working tree, and I periodically have to go through and
delete them.

I noticed just now that git creates its equivalent of darcs-record-N within .git.
This is analogous to my suggestion of putting temp files in _darcs/.