[This patch seems harmless enough I guess.

On the other hand, the NFS implementation in AIX 3.x is known to be
quite bad in just about every way.  More recent NFS implementations
are much more robust.  Given that I haven't heard much about this sort
of problem, it would seem like most unices don't have this problem.

-kingdon]

From: Andreas Luik <luik@isa.de>
Date: Mon, 20 Jan 1997 14:42:28 +0100
To: bug-cvs@prep.ai.mit.edu
Subject: Lock error due to errno == ESTALE

>Submitter-Id:   net
>Originator: 	 Andreas Luik
>Organization:
net
>Confidential:   no
>Synopsis:       Lock error due to errno == ESTALE
>Severity:       serious
>Priority:       medium
>Category: 	 cvs
>Class:          sw-bug
>Release:	cvs-1.7
>Environment:
    	
System: IBM RS/6000, AIX 3.2.5


>Description:
	
cvs update may abort with following message:

cvs checkout: Updating odstb/lib/idm
cvs checkout: [04:07:01] waiting for atc's lock in /project/atc/CVSMASTER/odstb/lib/idm
cvs checkout: couldn't stat lock directory `/project/atc/CVSMASTER/odstb/lib/idm/#cvs.lock': Missing file or filesystem
cvs checkout: failed to obtain dir lock in repository `/project/atc/CVSMASTER/odstb/lib/idm'
cvs [checkout aborted]: read lock failed - giving up

The reason seems to be that stat(2) call in lock.c:541 can return
errno == ESTALE (our repository is mounted via NFS) if the lock
directory has just been removed. Other systems correctly seem to
return ENOENT. 

>How-To-Repeat:
	
>Fix:
Perhaps errno == ESTALE should be checked on AIX, too:

*** cvs-1.7/src/lock.c.ORIG	Mon Dec 18 01:55:30 1995
--- cvs-1.7/src/lock.c	Mon Jan 20 14:31:29 1997
***************
*** 543,548 ****
--- 543,553 ----
  	    if (existence_error (errno))
  		continue;
  
+ #if defined(_AIX) && defined(ESTALE)
+ 	    if (errno == ESTALE)
+ 		continue;
+ #endif
+ 
  	    error (0, errno, "couldn't stat lock directory `%s'", masterlock);
  	    return (L_ERROR);
  	}