[netCDF #GNF-884914]: Error: netcdf-4.2-snapshot2012021222

Hi Craig,

> I am trying to build a 32-bit version of NetCDF-4 with the Intel compilers on 
> Mac OS X Lion (10.7.3) for a numerical model that does not run on 64-bit 
> platforms yet.
> 
> I have successfully built 32-bit versions of szip and hdf5.

Did you run "make check" on HDF5, and was it version 1.8.8?
I've found that "make check" fails using the llvm-based C compiler that 
comes with the Xcode-4.2.1 developer tools, but that if you build with
the gcc-4.6 that is available from Macports, the HDF5 library passes all
the tests from "make check".  I haven't had a chance to check if the 
resulting library works with the latest netCDF-4.2.snapshot yet.  

> When I run 'make check | & tee check.log' after building the C version of 
> NetCDF-4.2, I get the following error:
> 
> *** Testing NetCDF-4 with user-supplied sample file.
> *** testing creation of sample file...ok.
> *** Tests successful!
> PASS: tst_files4
> /bin/sh: line 1: 76866 Segmentation fault: 11  ${dir}$tst
> FAIL: tst_vars
> 
> . . .
> 
> ================================================
> 1 of 50 tests failed
> Please report to address@hidden
> ================================================
> gnumake[2]: *** [check-TESTS] Error 1
> gnumake[1]: *** [check-am] Error 2
> make: *** [check-recursive] Error 1
> 
> How can I track down this error? Is it catastrophic or can I plow ahead?

I wouldn't advise proceeding without knowing which test in tst_files4 failed,
and that would require duplicating the problem and doing some debugging.  
Sometimes the problems are in the tests, but more often they indicate a real
bug.  Which exact version of netCDF are you using, netCDF-4.2-rc1 or the 
daily snapshot?

> Fortran build fails too.
> 
> What is missing in the link process?

It looks like you're using ifort, and the error is

> Undefined symbols for architecture i386:
>  "_netcdf_mp_nf90_create_", referenced from:
>      _MAIN__ in f90tst_vars.o

Can you see what architecture f90tst_vars.o is compiled for, using

  otool -L f90tst_vars.o

Are you using the procedure described here for building the separate
Fortran library:

  
http://www.unidata.ucar.edu/mailing_lists/archives/netcdfgroup/2012/msg00019.html

--Russ



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



Ticket Details
===================
Ticket ID: GNF-884914
Department: Support netCDF
Priority: Normal
Status: Closed