19990323: segmentation violation with GRDDISP and McIDAS-X 7.5

>From: Robert Mullenax <address@hidden>
>Organization: National Scientific Balloon Facility
>Keywords: 199903230406.VAA25625 McIDAS-X 7.50 SC5.0 mcgget.for

Robert,

>I compiled, installed and configure 7.5 on our new Ultra Sparc powered
>system tonight.  Everything seem to be fine until I 
>ran GRDDISP to view model data.  I get the error "program terminated,
>segmentation violation" in the command line.
>
>There were no errors that I saw during the build.
>
>Here are the specifics:
>Solaris 7
>Solaris Workshop 5.0 compilers, 64-bit

A quick note is called for here.  We have a beta release of SC5.0 here
at Unidata.  The beta Fortran compiler that we are using will not even
compile grddisp.pgm!  I have been waiting for the production release
of SC5.0 to get here (it trickles down through the NCAR/UCAR maze before
we get to get it) so I could further explore whether or not the problem
is one of coding in grddisp.pgm or a compiler bug.

I logged onto your machine and started poking around.  The first thing
that scared me was the '64-bit' note you included above.  I was relieved
to find that the code had been built as 32-bit, however.

>Any ideas?

I am attempting to find out exactly where the routine is core dumping.
Once I have that, I will be able to make the determination as to whether
this is a McIDAS or Sun Fortran compiler bug.  If the problem is with
the McIDAS source, I will fix it; if it is with the Sun SC5.0 compiler,
then the only thing I can offer is to FTP a version of grddisp.k compiled
with SC4.2 to your machine and then report the bug to Sun.

>Thanks,

Sorry for the problems

>From address@hidden  Tue Mar 23 20:09:54 1999

I got back on your system this evening and did some more poking
around.  I found that not only does GRDDISP produce a segmentation
violation, but GRDINFO and GRDCOPY also produce the same results.

All of these routines call MCGGET.  If I compile it with optimization
turned off and relink GRDDISP, GRDINFO, and GRDCOPY, they run without
problems.  As soon as mcgget.for is compiled with optimization, all of
these GRD* routines dump core.

I don't know the reason for the cause of the failure yet, but at least
I know where to look.

I am leaving your system with an unoptimized version of mcgget.o
included in libmcidas.a, so relinks of whatever routines that call it
(there may be more than above; I havn't checked yet) should produce
functional code.  I will upload a fix as soon as I can figure out
exactly what is happening (assuming, of course, that the problem is
in the mcgget.for code and not in the compiler).

Tom