Missing DEFINE in configure script

[rparsons@…]

This problem really touches on two areas, the configure script as well as the 
code in the Mapserver C library. The problem is that virtually any output 
through the IMAGEMAP outputformat results in segmentation faults. I 
encountered the problem in perl Mapscript code, however, I learned that the 
command line utilities, such as shp2img exhibit the same behavior. I'm unsure 
of the scope of the problem. 

An investigation of the problem led me to im_iprintf() routine in the 
mapmygis.c source. This routine builds an output string by appending new 
formatted output onto an existing string, allocating memory as necessary to 
accommodate the growth. The SIGSEGV was being thrown by the realloc() within 
that routine. This indicated to me that non-allocated memory had been trampled 
on at some point, causing the realloc() to fail. Due to the nature of this 
routine I suspected an out-of-bounds memory write so I began searching for the 
culprit. Here's what I found:

The im_iprintf() routine relies on a call to vsnprintf() or vsprintf() to 
format and append the new string to the old. 

#if defined(HAVE_VSNPRINTF)
                n = vsnprintf((*(ps->string)) + ps->string_len,
                              remaining, fmt, ap);
#else
                n = vsprintf((*(ps->string)) + ps->string_len, fmt, ap);
#endif

An examination of the code shows that it is quite possible for the vsprintf() 
call to write past the allocated boundary, however, I initially dismissed this 
as being the cause of the problem because I knew that my Debian Linux platform 
supported vsnprintf(). After some more investigation I realized that vsprintf
() was indeed being called. Although it's quite possible that I overlooked it, 
I just couldn't find a mechanism within the distribution that would properly 
set HAVE_VSNPRINTF. To resolve the issue, I added the following line to my 
configure.in file and re-ran autoconf:

AC_CHECK_FUNC(vsnprintf,      STRINGS="-DHAVE_VSNPRINTF $STRINGS")

This properly detected and set the HAVE_VSNPRINTF value.

This addresses the problem on my platform as the SIGSEGV is no longer thrown, 
however, if there are platforms that don't support vsnprintf(), then the 
existing vsprintf() code could lead to problems as it certainly does write 
outside of the allocated segment. On platforms with good memory protection it 
will likely just throw SIGSEGV, however, on other platforms it could result in 
memory corruption. As such, in addition to looking into the 'configure' issue 
it might be a good idea for someone to review the im_iprintf() routine when 
using vsprintf().

Allow me to apologize in advance if any of my assumptions or assertions are 
incorrect (and they very well could be). Thank you to all of the 
Mapserver/Mapscript developers for your hard work on this fantastic product.

- Rob Parsons

[tbonfort]

test for HAVE_VSNPRINTF is in configure.in since r5130 (see also #1613)