Leo Famulari skribis: > On Fri, Jan 22, 2016 at 03:38:54PM -0500, Leo Famulari wrote: >> On Fri, Jan 22, 2016 at 09:34:39PM +0100, Ludovic Courtès wrote: >> > Leo Famulari skribis: >> > >> > > I thought that perhaps it was referring to B02typeset.ztst, but I don't >> > > think there is anything related in that area of the file: >> > > 540 ( >> > > 541 setopt glob >> > > 542 mkdir -p arrayglob >> > > 543 touch arrayglob/{one,two,three,four,five,six,seven} >> > > 544 fn() { >> > > 545 typeset array=(arrayglob/[tf]*) >> > >> > Out of curiosity, what file system is this on? >> >> Both filesystems are btrfs on LVM. The failing system is also using >> LUKS. OK. Btrfs often behaves differently from ext4, exhibiting issues in various places. The above test uses uses readdir(2) and friends to test glob patterns, so this could be an area where Btrfs yields different results. But anyway, you said /tmp is not on Btrfs. > Well, actually, that only describes /home, which is perhaps not > relevant, although I did try and fail to build in a `guix environment > zsh` "by hand" in /home on the failing machine. > > Failing system (separate /tmp): > /tmp /dev/mapper/bad--vg-tmp on /tmp type ext4 (rw,relatime,stripe=4,data=ordered) > > Working system (/tmp on /): > /dev/mapper/good--vg-root on / type ext4 (rw,relatime,errors=remount-ro,data=ordered) I’m not sure what could happen here. Could you try to analyze the test log and see what exactly is failing? Once you’ve done that, you could try to strace the test on both the “good” and the “bad” file system, to spot the difference. Alternately, you could leave that for upstream to solve. ;-) Thanks, Ludo’.