[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

19990908: McIDAS-X 7.60 GRDCOPY, GRDDISP built with SC5.0

>From: address@hidden
>Organization: Unidata
>Keywords: McIDAS-X 7.60 GRDDISP GRDCOPY Sun SC5.0 f77

Hsie,

I poked around on my system looking for the cause of the segmentation
violations that we both were seeing when running GRDCOPY.  What I found
was disturbing: there is apparently a bug in the Sun SC5.0 Fortran
compiler, f77.  We both have this compiler, so we are both seeing the
same problem.

I verified that this was the problem by:

o examining the source code of the routines that were bombing; there
  are no discernable problems

o putting in debugging information to see where routines were bombing;
  they were screwing up the simple statement in mcgget.for:

  DO 8 JA=1,NSORT

  The values for JA in the DO loop were:

  picking up sort conditions, ja = -268443984
  picking up sort conditions, ja = -268443983
  picking up sort conditions, ja = -268443982
  picking up sort conditions, ja = -268443981
  picking up sort conditions, ja = -268443980
  picking up sort conditions, ja = -268443979
  picking up sort conditions, ja = -268443978

  instead of the expected:

  picking up sort conditions, ja = 1
  picking up sort conditions, ja = 2
  picking up sort conditions, ja = 3
  picking up sort conditions, ja = 4
  picking up sort conditions, ja = 5
  picking up sort conditions, ja = 6

I found that I could get JA to assume proper values by modifying the
do loop as follows:

  DO 8 i=1,NSORT
  JA = i

This change would make GRDDISP work correctly, but not GRDCOPY (they both
call mcgget).

I then compiled the GRD* commands (along with mcgget.for) using the
SC4.2 Fortran compiler, and all of the routines ran correctly.  This
points to a problem with SC5.0.  We will have to research it more and
then report it to Sun.

In order to get you going, I FTPed versions of GRDDISP and GRDCOPY
created with SC4.2 from our Solaris SPARC system to yours.  I verified
that the new GRDCOPY command runs correctly by running uwgrid.sh:

cd workdata
./uwgrid.sh NG 1999251 12

What bothers me is the following: if SC5.0 f77 is bad, then what else
in the McIDAS (and other) package will run incorrectly?

I will try to keep on top of this situation and let you know what the
outcome is.

By the way, I wasn't really sure about how you have things setup on
stratus.  I noted that the copies of executables in
/home/mcidas/mcidas7.6/src are not the same files as those in
/home/mcidas/bin (i.e. the files were not hard links).  Also, it
appears that the /home/mcidas/bin files are the same as those in
/home/mcidas/m7.6/bin.  This agrees with how you have your runtime link
setup.  So, I FTPed the new executables to /home/mcidas/bin so you
could use them.

Tom