Date: Mon, 28 Oct 2002 23:52:07 +0200
From: Laurynas Biveinis <lauras AT softhome DOT net>
X-Mailer: The Bat! (v1.61) Personal
X-Priority: 3 (Normal)
Message-ID: <8942416321.20021028235207@softhome.net>
To: Andrew Cottrell <acottrel AT ihug DOT com DOT au>
CC: DJGPP Workers <djgpp-workers AT delorie DOT com>
Subject: Findutils 4.1.7 port bugs & fix
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-OriginalArrivalTime: 28 Oct 2002 21:49:31.0595 (UTC) FILETIME=[E5A371B0:01C27ECB]
Reply-To: djgpp-workers AT delorie DOT com

Hello,

There are two problems with find which can be seen during GCC
bootstrap on 2K:
c:/devel/djgpp/bin/find.exe: error waiting for rm: No child processes (ECHILD)
c:/devel/djgpp/bin/find.exe: c:/devel/djgpp/gnu/build.gcc/gcc/include/.. changed during
execution of c:/devel/djgpp/bin/find.exe

'error waiting for rm' is caused by bitrot in DJGPP port - find 4.1
upstream uses wait(), 4.1.7 uses waitpid(). DJGPP port used to provide
its own version of wait() which does DTRT instead of stock lib
version. I've updated it to provide special waitpid() instead, that
solves the first problem. Patch (and ChangeLog, in the case you'd
like to forward it upstream) are at the end.

The second problem is that find stats directories during their
traversal and caches their inodes and in the end we receive error that
inode of some directory has changed.

Is it possible that our stat() implementation returns different inode
for the same directory? If yes, what could we do about it? If no, I'd
like to know if this bug is 2K specific or not. I'd appreciate if
someone downloads find-4.1.7 from 2.04 test build page, does something
like (this is from GCC):
$ touch Done
$ find . -name '[Dd][Oo][Nn][Ee]' -exec rm -f '{}' ';'

and reports any error messages on Win98/DOS, maybe on XP and other
platforms too.

2002-10-28  Laurynas Biveinis  <laurynas DOT biveinis AT mif DOT vu DOT lt>

            * find/pred.c (wait): removed.
            (wait_pid): new function.
            (DOS_PID): new macro.
            

--- pred.c.old  2001-11-19 12:17:22.000000000 +0000
+++ pred.c      2002-10-28 22:54:44.000000000 +0000
@@ -1267,14 +1267,19 @@ static void dos_signal_catcher (int sig)
     }
 }
 
-/* Replacement for the `wait' library function that doesn't
+#define DOS_PID ((pid_t)42)
+
+/* Replacement for the `waitpid' library function that doesn't
    merely fail.  */
-pid_t wait (int *status)
+pid_t waitpid (pid_t pid, int *status, int options __attribute__((unused)))
 {
-  if (child_running)
+  if ((child_running) || (pid != DOS_PID))
+  {
+    errno = ECHILD;
     return (pid_t)(-1);
+  }
   *status = proc_status;
-  return (pid_t)42;
+  return DOS_PID;
 }
 
 /* The macros we need with `wait'.  */
@@ -1357,7 +1362,7 @@ launch (struct predicate *pred_ptr)
        _exit (1);
       }
 
-    child_pid = (pid_t)42;
+    child_pid = DOS_PID;
   }
 
 #else /* not __MSDOS__ */