20001206: McIDAS-X and compilers at stc

>From: alan anderson <address@hidden>
>Organization: St. Cloud State
>Keywords: 200012061433.eB6EXQo22789 McIDAS GEMPAK Sun compilers

Alan,

>I got a copy of mcidasx7.7 this morning and the build seems to be going
>fine.  

And I havn't even announced its availability yet :-)  By the way, you
are the _fifth_ site to download McIDAS-X before my announcement.

>My primary reason for writing concerns the compiler issue.  We are using 
>gcc and f2c and that works fine, but a colleague, Tony Hansen will want
>to build GEMPAK before too long, and we may have other packages we want 
>to use as well.  From the NOTES regarding MCIDASX, it seems that Sun 
>compilers, even the latest ones, may have some problems.  While the 
>solutions may be simple for a full time unix person, I think you know
>that would not apply to us.  

The Sun Fortran compiler issues are enough to drive one mad, all right.

>We are scheduled to get some sort of educational site license from Sun
>in the near future, which will give us access to their compilers at 
>low or zero cost, but I am wondering if using them may bring us more
>problems than we have now.  

One thing for both you and Tony to note is that Sun will _not_ be
providing a Fortran compiler for Solaris x86 in their WS6 compiler
set.  This is the set currently available for both SPARC and x86, so
your educational site license would naturally make the latest compiler
distributions available to you.  In order to build GEMPAK/McIDAS-X on
x86 using the Sun-supplied Fortran compiler, you would have to grab
SC5.0 and stick with it (patching it as patches become available, of
course).

>Does using gcc & f2c limit us in some way that Sun compilers would not?

Tests at SSEC show that code built with gcc/f2c runs slower than code
built with any vendor's native compilers.  Most Unidata sites are not
stressing their software in a way that they would really feel the
impact in this speed difference, however.

>I looked at the Installation Guide for GEMPAK and it seems to imply that 
>I need the compilers appropriate for the OS being used, without any 
>explicit mention of gcc or f2c.

Right.  It turns out that there are a couple of McIDAS routines that
"push the wrong button" in the compiler's optimizer.  The form of the
routines (subroutines with ENTRY points) is not used anywhere in
GEMPAK, so it never runs into the problems that I uncovered.  So, you
can build GEMPAK/NMAP with confidence right now.  The lack of a Fortran
compiler for Solaris x86 in WS6, however, is an issue that must be
remembered/dealt with.

>Any suggestions, especially regarding GEMPAK?  

Build with Sun compilers.  Keep SC5.0 around even if WS6 is available.
Stay tuned to see if Sun relents and adds Fortran to WS6 in the future
(doubtful given their proclimations regarding what they will support
Solaris x86 for).  Trust me to patch McIDAS builds in a way that avoids
the Sun compiler bugs (which, we _have_ been in touch with them about,
but so far have not gotten any satisfaction).

>From address@hidden Wed Dec  6 09:07:13 2000

>Oh Tom, 7.7  works great.  Thanks for all your good work.

Thanks for the nice comment.  The bad news is that you will be facing
mandatory upgrades throught the Addenda process as I tweak some of the
code that serves up the compressed NIDS data that becomes freely
available on January 1.  The code that is in the distribution right now
(as of 4 am this morning) works fine, but I want to add some
performance enhancement tweaks to it.  Also, I will be diving off into
the MCGUI interface for a total rewrite in the next couple of months so
that sites will (should, at least) have a fully functional, ADDEized
McIDAS GUI.

By the way, I will be punching up the McIDAS Release Notes today and
then officially announcing the distribution to the mcidas-x email
list.  When you see the announcement, you _should/must_ read the
Release Notes online docs to see the things you will need to worry
about on your system that runs McIDAS-XCD decoders.

Talk to you later...

Tom