On Wed, 23 May 2012 16:28:52 -0300 Raphael Kubo da Costa <rakuco@...>
> Vincent Torri <vincent.torri@...> writes:
> > On Mon, May 21, 2012 at 10:28 AM, Joerg Sonnenberger
> > <joerg@...> wrote:
> >> On Mon, May 21, 2012 at 12:15:10AM -0300, Raphael Kubo da Costa wrote:
> >>> After talking about the previous version of the patch with raster on
> >>> IRC, I'm sending a more conservative version that does not change the
> >>> Linux behavior -- /dev/shm is expected to work on Linux, and the
> >>> non-Linux case works at the expense of leaving some files in /tmp if a
> >>> crash occurs before _cleanup_tmpf() is called.
> >> Can you not hard-code /tmp, but try to honor TMPDIR?
> > TMPDIR does not exist on Windows. TMP and TEMP do
> New attempt attached: we now check TMPDIR, TMP and TEMP (in this order)
> before falling back to "/tmp".
beyond using TMPDIR, TMP and TEMP env vars... i dont see how this helps much :)
in fact they probably should be there in if (fd < 0). on non-linux it STAR TS
with fd being -1 so it goes right it the failure case. for linx this means
failure case still works. we.l that and the ifdef around the unlink.
i've put in an alternate patch in svn that keeps the tmpdir fallback for evas.
------------- Codito, ergo sum - "I code, therefore I am" --------------
The Rasterman (Carsten Haitzler) raster@...