Re: 20000420: netCDF on BeOS

>To: address@hidden
>From: JefF Wong <address@hidden>
>Subject: netCDF on BeOS
>Organization: Canada Centre for Remote Sensing; Natural Resources Canada
>Keywords: 200004201225.GAA10696

Hi Jeff,

> I have been attempting to compile version 3.5a of the netcdf libraries and
> programs on a Pentium-2 based system running BeOS v4.5.2 (which has a POSIX
> layer).  This environment uses gcc version 2.9-beos-980929.  Fortran is
> handled through the combination of fort77 (v1.14a) and f2c (the last recorded
> change in the "readme" file is Nov 15, 1999).

The VERSION file of the current netcdf-3.5 beta release at

  ftp://ftp.unidata.ucar.edu/pub/netcdf/netcdf-3.5-beta2.ZIP
or
  ftp://ftp.unidata.ucar.edu/pub/netcdf/netcdf-3.5-beta2.tar.Z

contains "3.5" rather than "3.5a", so maybe what's available is a
little bit more recent than what you have been working with, but we
don't have access to a BeOS platform, so I'm not sure there have been
any changes to make it any more portable to BeOS.

> I had set the following environment variables prior to attempting anything:
> export CPPFLAGS="-DNDEBUG -Df2cFortran"
> export FC="/boot/home/bin/fort77"
> export FFLAGS="-O -Nx400 -w"
>
> I am attaching the log files from:
> ./configure --prefix=/boot/home/src/netcdf >> configure.log 2>> configure.log;
> make >> make0.log 2>> make0.log
>
> I editted ncgen/ncgentab.c to remove the line containing:
>       #include <values.h>
> and did
> make >> make1.log 2>> make1.log
> make test >> test0.log 2>> test0.log

At least that problem seems to have been fixed in the more recent beta
release; there is no "#include <values.h>" statement in the
"ncgen/ncgentab.c" source.

> I editted nc_test/Makefile to remove the line containing
>       ld_math               = -lm
> and did:
> make test >> test1.log 2>> test1.log
> make install >> install.log 2>> install.log

I'm surprised you could get away with deleting the "-lm" flag to link
in the math libraries, because nc_test makes use of the fabs() and
floor() functions from the math libraries.  I can't get it to link if
I omit "-lm".

> As you can see, there are still errors.
> Do you have any suggestions about these problems?

It looks like you've solved almost all the problems.  From the output
you've sent, there are only two things to be concerned about:

 1.  The output from "make test" in the libsrc/ directory
       cmp test.nc test_nc.sav
       test.nc test_nc.sav differ: char 7681, line 7

 2.  The return from nc_open():

       FAILURE at line 94 of test_read.c: nc_open of nonexistent file
       should have returned system error 

The first problem indicates a difference in the binary netCDF file
created by the test program and a saved version of what it should have
produced.  We expect these to be bit-for-bit identical, but if there
are only tiny differences in floating-point values, for example, this
problem may not be serious.  If you have the "ncdump" netCDF utility
installed on another platform, you could compare the output of "ncdump
test.nc" with the output of "ncdump test_nc.sav" to see if you think
the difference is serious.  If you don't have a working version of
ncdump, send us the file libsrc/test.nc and we'll have a look at it to
see if the differences are anything to be concerned about.

The second problem seems less serious, and just indicates an
lack of an expected error return when trying to open a nonexistent
file.  This must ultimately be a problem with the system call

        fd = open(path, oflags, 0);

in libsrc/posixio.c returning with something other than -1 when
invoked for a nonexistent file.  That sounds like a problem with the
BeOS POSIX I/O, but you could probably work around it pretty easily by
finding out the BeOS way to determine if a file doesn't exist and
using that to return what is expected.

--Russ

_____________________________________________________________________

Russ Rew                                         UCAR Unidata Program
address@hidden                     http://www.unidata.ucar.edu