On Wed, Feb 29, 2012 at 21:29:37 +0100, Julien Cristau wrote: > On Tue, Feb 28, 2012 at 20:21:39 +0100, Julien Cristau wrote: > > > On Tue, Feb 28, 2012 at 19:05:23 +0100, vladz wrote: > > > > > On Tue, Feb 28, 2012 at 06:42:59PM +0100, Julien Cristau wrote: > > > > > As a solution, I would suggest to take care of the "mkdir" return codes > > > > > (line 36 and 50). To do not change permissions on failures. > > > > > > > > > This script is set -e AFAICT, which means it already does care about the > > > > mkdir return code. > > > > > > Yes but with the "-p" option, mkdir always return 0 (success): > > > > > > $ mkdir /tmp/dir > > > $ mkdir /tmp/dir > > > mkdir: cannot create directory `/tmp/dir': File exists > > > $ echo $? > > > 1 > > > $ mkdir -p /tmp/dir > > > $ echo $? > > > 0 > > > > > Right, makes sense. I can drop the -p, I guess. Not sure what impact > > that would have on things assuming they can use /tmp/.X11-unix (I > > wouldn't really like to fix this just to have the same issue elsewhere). > > Looking at trans_mkdir > > (http://cgit.freedesktop.org/xorg/lib/libxtrans/tree/Xtransutil.c#n480) > > it *looks* like it should be safe, though. > > > Actually it's not going to work. If /tmp/.X11-unix exists and is a > directory (not a symlink), that's good enough for us, we don't want to > fail in that case. > And while I'm at it I'd also like to fix the $SOCKET_DIR.$$ thing to use a random name instead (probably with mktemp). Cheers, Julien
Attachment:
signature.asc
Description: Digital signature