zsh and network based home directory

Hi david,

Are the clocks of your Mac and the file server in sync with each other?

Hi david,

They are 45 seconds off. Is that too much?

No, I think it is OK.

If the zsh hangs only if HISTFILE=xxxxx exists in .zshrc, then I guess it hangs while creating a lock file (.zsh_history.LOCK) for the history file.

Does the file ~/.zsh_history (and .zsh_history.LOCK) already exist?
If they exit, check their owner and permission, and try removing them.
Also check whether .zsh_history.xxxxxx (xxxxxx is a random string) exits or not.

If this doesn't help, then it would be helpfull to debug the hanging zsh by using gdb. Do you know how to do it?

PowerMac G4 Mac OS X (10.4.5)

If I point the history file to /tmp it works (but no lock file is created).

The lock file is deleted if the history file has been successfully read.

There are a lot of .zsh_history.PID-files in ~/

Zsh first create .zsh_history.xxxxxx and makes a hard link of it as the name .zsh_history.LOCK, and deletes .zsh_history.xxxxxx. So I guess zsh hangs when it tries to make a hard link.

The following is a quick memo of how to debug a running (or hanging) zsh:

First, setup Terminal.app so that you can open a Terminal window with a working shell (bash, tcsh, etc.). I guess you know how to do this (or already did this).

Then oepn a Terminal window (in which bash/tcsh is running), and start a zsh in the window; the zsh hangs, right?

Then open another Terminal window, and use ps command to find the PID of the zsh which is hanging. Also try to figure out whether the zsh is running (consuming the CPU cycle) or just waiting (sleeping); the STAT field in the output of ps command may contain 'R' if the zsh is running (You can also use the top command or the Activity Monitor.app).

Now in the new window, run gdb as

gdb /bin/zsh nnn

where nnn is the PID of the zsh. Then in the (gdb) prompt, type

(gdb) where

This prints the stack frames; #1 is the function in which the zsh is now, #2 is the function which called #1, etc.


If you can build zsh binary from the source code, then download the source code from the zsh website (or ftp server), and configure it with --enable-zsh-debug

./configure --enable-zsh-debug
make
make install (install into /usr/local/bin)

Then run the zsh you have just built (/usr/local/bin/zsh). If it hangs, then (in another window) go to the Src directory of the zsh and debug the hanging zsh by gdb zsh nnn. Now you can do source-level debugging.

PowerMac G4 Mac OS X (10.4.5)

Please double check that .zsh_history.LOCK does not exists. If it exists, then manually delete it.

Is the zsh running or sleeping?

Before starting gdb, please use top (or Activity Monitor.app) to find whether the zsh is consuming CPU cycle or just sleeping.

Or in (gdb) prompt, type

(gdb) finish

to run the program until it exits from the current function. If this returns immediately then zsh was running, while if this hangs then it was waiting for stat() to finish.

PowerMac G4 Mac OS X (10.4.5)

(I think saving history into a file is not so critical and you can live without it for a moment, during an upgrade, for example).

If you look into the source code of lockhistfile() (in hist.c), you will find there is a while loop while(link(tmpfile,lockfile)<0) { ... }</tt>. So my guess is the link() system call is failing with a strange error and your zsh is in an infinite loop.
(If you want to know what is happening, the best way is to download the source, build with --enable-zsh-debug, and do source-level debugging.)
What happens if you compile the following test code and run it in your home directory?
(Please modify the path names so that they point to your home directory)
<pre>
#include <stdio.h>
#include <unistd.h>
int main()
{
char *tmpfile = "/Network/Servers/SERVER/Users/name/.zsh_history.2217";
char *lockfile= "/Network/Servers/SERVER/Users/name/.zsh_history.LOCK";
if( link(tmpfile,lockfile) < 0 ) {
perror(NULL);
}
return 0;
}
</pre>
Save this in lock.c, and
<pre>
cc lock.c
./a.out
</pre>
If you have a file .zsh_history.2217 but do not have .zsh_history.LOCK, then a.out should end without any error message and .zsh_history.LOCK will be created.
By the way, if the pathname to your home directory contains symbolic link(s), what happens if you set HISTFILE to the true path name? For example, if /Network/Servers is a symbolic link to /automount/Servers, then
HISTFILE=/automount/Servers/SERVER/Users/name/.zsh_history
PowerMac G4 Mac OS X (10.4.5)

Sorry for not responding promptly.

% ./a.out
Operation not supported

On my PowerBook (10.4.4), I mounted a folder from my PowerMac (10.4.5), went to the mounted folder and ran the a.out (with appropriate changes in the path names, of course). Then I got the same result.

Also the ln command fails in an AFP-mounted directory:

% cd /Volumes/Server/
% touch old
% ln old new
Operation not supported

Either this is a bug, or a "feature", or there is a (hidden) option for afp_mount which makes the link() system call to work in a AFP-mounted directory...

PowerMac G4 Mac OS X (10.4.5)