sort.c minor cleanup in volatile setjmp code

[Paul Eggert]

I installed this patch.  It doesn't fix any bugs, but it does remove a
confusing and unnecessary "volatile" and I suppose it improves
performance by a few nanoseconds.

2004-12-01  Paul Eggert  <address@hidden>

        * src/ls.c (sort_files): Minor cleanup.  Remove an unnecessary
        'volatile' on a local variable.  Rewrite to avoid unnecessary
        double-assignment to 'func' in the usual case where strcoll does
        not fail.

Index: src/ls.c
===================================================================
RCS file: /fetish/cu/src/ls.c,v
retrieving revision 1.368
retrieving revision 1.369
diff -p -u -b -w -r1.368 -r1.369
--- src/ls.c    19 Nov 2004 21:41:00 -0000      1.368
+++ src/ls.c    2 Dec 2004 00:31:43 -0000       1.369
@@ -2871,10 +2871,15 @@ static int rev_str_extension (V a, V b) 
 static void
 sort_files (void)
 {
-  /* `func' must be `volatile', so it can't be
-     clobbered by a `longjmp' into this function.  */
-  int (* volatile func) (V, V);
+  int (*func) (V, V);
 
+  /* Try strcoll.  If it fails, fall back on strcmp.  We can't safely
+     ignore strcoll failures, as a failing strcoll might be a
+     comparison function that is not a total order, and if we ignored
+     the failure this might cause qsort to dump core.  */
+
+  if (! setjmp (failed_strcoll))
+    {
   switch (sort_type)
     {
     case sort_none:
@@ -2910,13 +2915,8 @@ sort_files (void)
     default:
       abort ();
     }
-
-  /* Try strcoll.  If it fails, fall back on strcmp.  We can't safely
-     ignore strcoll failures, as a failing strcoll might be a
-     comparison function that is not a total order, and if we ignored
-     the failure this might cause qsort to dump core.  */
-
-  if (setjmp (failed_strcoll))
+    }
+  else
     {
       switch (sort_type)
        {