Bug 30878

I have the same problem with the tcsh. If the autocomplete fails in the tcsh for one file or
directory, than it also fail in emacs !

Question to all the others: Could it be that it only appears on nfs mounted partitions ???

midnight commander also hide the directories and files that are always
missing from the tcsh autocomplete.

Until now this bug is only noticed on files and dirs on nfs mounted partitions.

midnight commander also hide the directories and files that are always
missing from the tcsh autocomplete.

Until now this bug is only noticed on files and dirs on nfs mounted partitions.

This problem somehow seems to be related to the nfs problem reported as Bug -
30944. After installing the patch from
http://www.fys.uio.no/~trondmy/src/2.4.2/linux-2.4.2-dir.dif as suggested there,
the problem disappeared. At least over here ;-).

More Detail:

tcsh appears to fail when expanding the * wildcard on nfs mounted directory 
from our SGI fileservers. It does not fail when expanding * on an nfs 
mounted directory served from a redhat 7.1 machine or on the local filesystem.

the section of strace which appears to be important:

strace tcsh -c 'echo *' on a SGI nfs directory gives:

rt_sigprocmask(SIG_SETMASK, [], [INT], 8) = 0
open(".", O_RDONLY|O_NONBLOCK|O_LARGEFILE|O_DIRECTORY) = 3
fstat64(3, {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
fcntl64(3, F_SETFD, FD_CLOEXEC)         = 0
getdents64(3, /* 20 entries */, 32768)  = 800
_llseek(3, 18446744073199903873, 0xbfff0200, SEEK_SET) = -1 EINVAL (Invalid 
argument)
lstat64("libglx.so.1.0.1251", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("mga_drv.o", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("mgapdesk-1_00-5beta.i386.rpm", {st_mode=S_IFREG|0644, st_size=0, ...}) 
= 0
lstat64("NVdriver", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("config-2.4.4", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0


Which has an odd _llseek call.

The same on a local file system produces:

rt_sigprocmask(SIG_SETMASK, [], [INT], 8) = 0
open(".", O_RDONLY|O_NONBLOCK|O_LARGEFILE|O_DIRECTORY) = 3
fstat64(3, {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
fcntl64(3, F_SETFD, FD_CLOEXEC)         = 0
getdents64(3, /* 20 entries */, 4096)   = 800
lstat64("config-2.4.4", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("initrd-2.4.4.img", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("kdmrc", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("libGLcore.so.1.0.1251", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("libGL.so.1.0.1251", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
lstat64("libglx.so.1.0.1251", {st_mode=S_IFREG|0644, st_size=0, ...}) = 0

no _llseek call. You also don't get one from a nfs directory served from a 
redhat 7.1 box.

If you do an unsorted directory listing (ls -U) it is always the last listed 
file that is missing.

If you run sh instead (sh -c 'echo *') then this behaves correctly and returns 
all files, so it is definately a tcsh and nfs interaction problem.

Bug also present if you use tcsh 6.09.00 as opposed to 6.10.00 when reading nfs 
mounted partions.

The issue doesn't appear to be entirely tcsh specific as we have reproduced the
wildcard expansion problem with bash 2.02.0(1) on dirs NFS mounted from IRIX
6.5.11 .  Bash 2.04.21(1) which ships with RH7.1 seems to work correctly.

Having looked at a similar bug about NFS (#30944) there is a patch called linux-
2.4.2-dir.dif which can be found here: http://www.fys.uio.no/~trondmy/src/2.4.2/
This is described as: "An experimental patch for fixing a problem that is due 
to NFSv(2|3) readdir returning (32|64) bit unsigned offsets"

I sucessfully applied this to my 2.4.4 kernel using patch -p1 from the kernel 
source root directory. This appears to have solved all the problems that I have 
been having, including allowing StarOffice 5.2 to work properly again.

So my only outstanding query is why this behaviour is only exhibited in tcsh as 
part of redhat 7.1 and not in sh / bash.

Created attachment 44177 [details]
strace tcsh -cf 'ls F99UA1.L.full.segment.final.5.SPHERE*coord'

I'm having similar globbing problems (wildcards not matching when they should)
in tcsh on RedHat 7.1 with IRIX 6.5 nfs file systems.

pulvinar 46% uname -a
Linux pulvinar 2.4.9-12 #1 Tue Oct 30 18:33:49 EST 2001 i686 unknown
pulvinar 47% echo $SHELL
/bin/tcsh
pulvinar 48% ls *coord
ls: No match.
pulvinar 49% sh
sh-2.04$ ls *coord
BAK.F99UA1.L.full.34132.flat.CartStd.coord
BAK.F99UA1.L.full.34132.sphere.SphStd.coord
F99UA1.L.full.34132.flat.at0.5mm.CartStd.coord
F99UA1.L.full.34132.flat.CartStd2.At1mm.coord
F99UA1.L.full.34132.flat.CartStd.At1mm.coord
F99UA1.L.full.34132.flat.CartStd.At1mm.newcut.sm.coord
F99UA1.L.full.34132.flat.CartStd.At1mm.newcut.sm.FLAT_CYCLE1.34132.coord
F99UA1.L.full.34132.flat.CartStd.At1mm.newcut.sm.FLAT_CYCLE2.34132.coord
F99UA1.L.full.34132.flat.CartStd.At1mm.newcut.sm.FLAT_CYCLE3.34132.coord
F99UA1.L.full.34132.flat.CartStd.coord
F99UA1.L.full.34132.sphere.SphStd.At1mm..coord
F99UA1.L.full.34132.sphere.SphStd.coord
F99UA1.L.full.fiducial.SmoothedMW_ACorigin.34132.coord
F99UA1.L.full.fiducial.SmoothedMW_ACorigin.At0.5mm.34132.coord
F99UA1.L.full.fiducial.SmoothedMW_ACorigin.At1mm.34132.coord
F99UA1.L.full.inflated.At1mm.34132.coord
F99UA1.L.full.segment.final.5.CompressedMedialWall.34132.coord
F99UA1.L.full.segment.final.5.ellipsoid.34132.coord
F99UA1.L.full.segment.final.5.fiducial.34132.coord
F99UA1.L.full.segment.final.5.fiducial.clean.34132.coord
F99UA1.L.full.segment.final.5.fiducial.clean_closed.34132.coord
F99UA1.L.full.segment.final.5.fiducial.clean_closed.smMW.34132.coord
F99UA1.L.full.segment.final.5.inflated.34132.coord
F99UA1.L.full.segment.final.5.InitialFlat.34132.coord
F99UA1.L.full.segment.final.5.InitialFlat.sm.34132.coord
F99UA1.L.full.segment.final.5.raw.34132.coord
F99UA1.L.full.segment.final.5.sphere.34132.coord
F99UA1.L.full.segment.final.5.SPHERE.34132.coord
sh-2.04$ 

Unlike the problem described in bug 30944, it is not just one/last file
missing.  Also, bug 30944 says Patch is in released errata kernel 2.4.3-12;
wouldn't this mean it would be included in higher kernel releases (above results
are using 2.4.9-12)?

I also thought the problem was restricted to tcsh, since bash results are good
in the above output; however, the way I found this problem directory was by
running this sh script on both IRIX and Linux and comparing the output:

#!/bin/sh

REV=`uname -a | cut -f1,3 -d' '|sed 's/ /_/g'`
OUTFNAME=files_$REV.txt
echo $OUTFNAME
#rm $OUTFNAME
FILE_SYSTEMS="/surface02 /sums2 /sums3 /stpb"

for FILE_SYSTEM in $FILE_SYSTEMS
do
        find $FILE_SYSTEM -name "*coord" | sort -uf >> $OUTFNAME
done

In this case, files_Linux_2.4.9-12.txt was missing
F99UA1.L.full.segment.final.5.SPHERE.34132.coord -- the last file in the
interactive sh results above.  (I confirmed the file was not missing/moved when
the script was run.)

Using kernel 2.4.2, I also have globbing problems, but with different
directories/files.

The attached file globtrace.txt
(https://bugzilla.redhat.com/bugzilla/showattachment.cgi?attach_id=44177) shows
the output from strace tcsh -cf 'ls F99UA1.L.full.segment.final.5.SPHERE*coord'.

We are seeing the same symptom - will try to set up a test with a
more modern version to see if its been subsequently fixed.

*** This bug has been marked as a duplicate of 73877 ***

Changed to 'CLOSED' state since 'RESOLVED' has been deprecated.