Well, going back and looking at the change, the hack wasn't as extensive as
I remembered. This patch is generated off the current CVS tree:

--- mprintf_orig.c 2006-07-06 06:33:56.000000000 -0700
+++ mprintf.c 2006-07-28 16:36:07.453125000 -0700
@@ -606,14 +606,24 @@
   long param; /* current parameter to read */
   long param_num=0; /* parameter counter */
 
- va_stack_t vto[MAX_PARAMETERS];
- char *endpos[MAX_PARAMETERS];
+ va_stack_t *vto /*[MAX_PARAMETERS]*/;
+ char **endpos /*[MAX_PARAMETERS]*/;
   char **end;
 
   char work[BUFFSIZE];
 
   va_stack_t *p;
 
+ vto = (va_stack_t *)calloc( MAX_PARAMETERS, sizeof(va_stack_t) );
+ endpos = (char **)calloc( MAX_PARAMETERS, sizeof(char *) );
+
+ if ( !vto || !endpos )
+ {
+ free( vto );
+ free( endpos );
+ return 0;
+ }
+
   /* Do the actual %-code parsing */
   dprintf_Pass1((char *)format, vto, endpos, ap_save);
 
@@ -982,6 +992,9 @@
     f = *end++; /* goto end of %-code */
 
   }
+ free( vto );
+ free( endpos );
+
   return done;
 }

** End of patch **

This assumes that the implementation of free() accepts a null argument,
which is normally the convention.

Enjoy, for what it's worth!

Gary Maxwell
Casabi, Inc.

-----Original Message-----
From: Gary Maxwell [mailto:gmaxwell_at_casabi.com]
Sent: Friday, July 28, 2006 4:04 PM
To: 'libcurl development'
Subject: RE: Stack smashed when linked with pthreads

We recently ported libcurl to an embedded ARM system with microscopic thread
stacks (3-4 kbytes typical) and ran into this problem.

I traced it down to the mprintf module, where local stack frames are large
and the functions can be called recursively. I hacked a change so that the
large stack frame objects were allocated off the heap (heap space is not an
issue), and that solved the problem.

I haven't considered submitting a patch, unless there is general hue and cry
for one.

Gary Maxwell
Casabi

-----Original Message-----
From: Dan Fandrich [mailto:dan_at_coneharvesters.com]
Sent: Friday, July 28, 2006 3:49 PM
To: curl-library_at_cool.haxx.se
Subject: Re: Stack smashed when linked with pthreads

On Fri, Jul 28, 2006 at 03:58:34PM -0500, Marshall Crocker wrote:
> Hmm...that may be it. I'm not exactly sure of my stack size but I am
> using the linux-tiny patches for the 2.6 kernel which may very well
> reduce the size below the default 8K. I assume 8K would be big enough
> to handle curl and pthreads right?

8K is very small. There is at least one place in the curl app where a 64K
buffer is put onto the stack, and I wouldn't be surprised if there are
buffers in the KB range in libcurl. Some of the call stacks can get pretty
deep, so I'd be surprised if you *didn't* see stack overflows at 8K.

>>> Dan

-- 
http://www.MoveAnnouncer.com              The web change of address service
          Let webmasters know that your web site has moved