20000124: McIDAS 7.60/Solaris 2.5.1

>From: James T Brown <address@hidden>
>Organization: Michigan State University
>Keywords: 200001242310.QAA18423 McIDAS-X gcc Solaris 2.5.1

Jim,

>I am writing to ask if you might know of anyone who has
>successfully compiled McIDAS 7.60 on a Solaris 2.5.1
>platform (sun Ultra 30).

I have built McIDAS on Solaris 2.5.1 in the past, but not Version 7.60.
I believe that there are more than one site out there right now that
has built on 2.5.1, but I believe that they are using vendor compilers.

>I do not have the latest Sun
>compilers and have been trying to compile using gcc/f2c.

OK.

>I have been successful with this type of install on an
>Ultra2 running Solaris 7, but I do need to compile for 
>a lab of workstations that are not as current.

I understand.

>Here is some compile info:
>
>   gcc      - version 2.95.1
>   libf2c.a - version 20000118

It seems like this should work correctly.

>   Copy of ~mcidas/.cshrc:
>
>   > setenv McINST_ROOT      /home/mcidas
>   > if ( ! ${?MCPATH} ) then
>   >    setenv MCDATA        /home/mcidas/workdata
>   >    setenv MCPATH       ${MCDATA}:/home/mcidas/data:/home/mcidas/help
>   >    setenv MCGUI         /home/mcidas/bin
>   >    setenv MCTABLE_READ 
> "${MCDATA}/MCTABLE.TXT;/home/mcidas/data/ADDESITE.TXT"
>   >    setenv MCTABLE_WRITE "/home/mcidas/data/ADDESITE.TXT"
>   >    setenv XCD_disp_file $MCDATA/DECOSTAT.DAT
>   > endif
>   > setenv PATH ${MCGUI}:$PATH
>   > setenv CC /usr/local/bin/gcc
>   > setenv CPPFLAGS -DNDEBUG
>   > setenv CFLAGS -O
>   > setenv FC /home/mcidas/mcidas7.6/src/mcfc
>   > setenv FFLAGS -O
>   > setenv LD_LIBRARY_PATH /usr/local/lib:${LD_LIBRARY_PATH}

This all looks good.

>The vendor flag is also blanked out in the "makefile".

Good.

>Using this setup and using the gcc/f2c options, it appears
>that many routines do compile.  Here is a sample of the 
>output that is displayed after running "make all":
>
>   >...
>   > compile mctheta.c:              done
>   > compile mctmpfil.c:             done
>   > compile mcts.c:                 done
>   > compile mcvort.c:               done
>   > compile mcwrite.c:              done
>   > compile metfunc.c:              done
>   > compile metplot.c:              done
>   > compile metplot2.c:             FAILED (See
>'/home/mcidas/mcidas7.6/src/makelog' for details.)

OK.

>The makelog entry reports:
>
>   > ./mccomp -O -I. -I../netcdf/libsrc -c metplot.c
>   > gcc -c -O -I. -I../netcdf/libsrc -I/usr/dt/include -I/usr/openwin/include 
> metplot.c
>   > compile metplot.c:              done
>   > ./mccomp -O -I. -I../netcdf/libsrc -c metplot2.c
>   > gcc -c -O -I. -I../netcdf/libsrc -I/usr/dt/include -I/usr/openwin/include 
> metplot2.c
>   > metplot2.c: In function `pcoord2':
>   > metplot2.c:899: Internal compiler error in `final_scan_insn', at
>final.c:2920
>   > Please submit a full bug report.
>   > See <URL:http://www.gnu.org/software/gcc/faq.html#bugreport> for
>instructions.
>   > compile metplot2.c:             FAILED

This sounds like a gcc problem.  Line 899 in metplot2.c is the end of
the pcoord2 routine:

} /* pcoord2 */

This looks inocuous enough.  Will the routine compile if you change the line
to:

/* pcoord2 */
}

>Again, all of this does compile on my Solaris 2.7 machine (although
>that does have a slightly newer version of gcc - 2.95.2).  Just 
>wondering if this a GNU GCC problem, or if this is something that
>can be fixed with the proper setting of a compile flag.

I would say that it is a gcc or 2.5.1 patch level problem.

>At any rate, without the latest Sun compilers, I was hoping to compile
>using gcc/f2c, but it appears that I have hit a wall.

This should work.

>I have looked for information
>in the McIDAS support archives, but was unable to find anything relating
>to builds on Solaris2.5.1.

That is because people havn't run into any problems with.  Then again,
most use vendor compilers on Solaris SPARC.

>I am wondering if I maybe am overlooking something

I don't think so.

>and would like to know if a successful compile on Solaris 2.5.1 machine
>is possible using gcc/f2c

Yes, it should be.

>before I bang my head against this problem any longer.

I understand and would take exactly the same approach.

>Any tips or suggestions would be greatly appreciated.

Please check to see if changing line 899 in metplot2.c allows you to 
proceed.  I really think, however, that the problem is with the scanning
routine that gcc is using.

Tom Yoksas

>From address@hidden  Tue Jan 25 10:52:14 2000

re: try changing line 899 in routine that bombs
>This change did not work.

I decided to install an earlier version of GNU gcc.  I 
switched to version 2.8.1 (down from 2.95.1).  It seems 
to have solved the problem.  Everything compiled and the 
initial McIDAS tests look good.  I won't have everything 
setup for awhile yet, but at least it compiled.

In summary, building for Solaris 2.5.1:

   GNU gcc 2.95.1:   doesn't work
   GNU gcc 2.8.1:    works fine


At any rate, I think the main hurdle has been crossed.
Thanks for your quick response last night.  Will keep 
you posted should I encounter any other difficulties.


Jim
-- 
==========================================================
James T. Brown   
UNIX Systems Administrator
Geography/Fisheries & Wildlife
Michigan State University
email: address@hidden