c/5077: Strange Code Generation

[l.e.matthias@larc.nasa.gov]

>Number:         5077
>Category:       c
>Synopsis:       Strange Code Generation
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Tue Dec 11 10:16:00 PST 2001
>Closed-Date:
>Last-Modified:
>Originator:     Larry Matthias
>Release:        3.0.1
>Organization:
SAIC/NASA Langley Research Center
		(757) 864-7556
>Environment:
System: IRIX64 darrin 6.5 01101245 IP27

	
host: mips-sgi-irix6.5
build: mips-sgi-irix6.5
target: mips-sgi-irix6.5
configured with: ../configure --prefix=/usr/freeware --enable-version-specific-runtime-libs --disable-shared --enable-threads --enable-haifa
>Description:
	We have a software system built with gcc on the Sun using 
version 2.8.1 of gcc that works properly. I am trying to port this
code over to the SGI. I initially built the code using version 2.95.2
and it compiles and links properly. But during execution of the 
application the code exits a loop and then quits (but with no
error messages). I attempted to debug the problem by adding print
statements into the specifc module that was crashing. It was then
that I noticed that by adding or removing printf statements the
code would quit at different places in the loop. The loop iterates
for 29 steps but quits before completing. If I leave a printf
statement in the code it will read ten items and then quit. If I 
remove the printf statement it will read in 20 items and then quit.
Adding and removing printf statements is causing the application
to fail at different times.
	I then tried this code with version 3.0.1 and saw identical
results. So I have an application that works under Sunos using 
version 2.8.1 but fails under IRIX using either version 2.95.2 or
3.0.1. I can think of no reason why the code should crash differently
by the changing of a printf statement. Even more important than
the irratic behavior of the application is to understand why it
is crashing to begin with.
 
>How-To-Repeat:
	Our application consists of over 5000 lines of code. This
code performs subsetting of scientific data from the Multi-angle
Imaging Spectoradiometer (MISR) satellite instrument from JPL. 
It uses the libraries HDF and HDF-EOS to access the scientific 
data. The particular loop that fails is reading global attributes 
and copying them to an output file. The code is allocating and
freeing memory inside of this loop so the problem may have
something to do with memory.
	Since this is a rather complex piece of code I am unsure
how we can get it to repeat the problem in your environment. We
would need to get the HDF and HDF-EOS libraries for you as well
as the application. Please contact me about how we could set 
this up.

>Fix:
	I know of no work around for this problem.

>Release-Note:
>Audit-Trail:
>Unformatted: