[netCDF #XZZ-151006]: static/shared mode of compiling netcdf with intel and pgi compiler

Thanks, Davide!

Would it be OK with you if I responded to the support ticket with the
following, which would also be sent to the netcdfgroup mailing list
(where the discussion took place before)?  Or would you rather that I
leave off the netcdfgroup on the email?  Or do you have any changes?

Hi Andre,

> here are the configurations I used to install the netcdf-library from source
> using the Intel Compiler Suite (Composer XE 2011 Update 4 for Linux,
> intel64_2011.4.191) on an (K)Ubuntu/Linux 11.04 64bit system.
> I installed zlib, hdf5, and netcdf from source.
> zlib-1.2.5: I was not able to compile the shared library with the Intel
> compiler (I don't know why), but it works using the GNU compiler.
> hdf5-1.8.6: no problems.
> netcdf-4.1.3: I had to compile in several iterations, incrementally adding
> optimization flags; I'm not sure why that is. Maybe there is a difference
> between shared and static library?
> My current build using the attached scripts works well.
> I don't have much experience with Intel compilers and source installs in
> general, so if anyone can shed some light on the above issues, that would be
> highly appreciated.

And you asked the netcdfgroup mailing list:

> ... but I'm using netcdf in Python, so I need the shared libs.
> ... is it possible to build shared libs for netcdf using static libs of zlib 
> and hdf5?

We have some new information on this, thanks to Davide Del Vento from NCAR's 
CISL (Computational Information Systems Laboratory):

  We have been able to compile netcdf-4.1.3 in fully shared mode with
  the Intel compiler (and soon with PGI). The issue is HDF5. In fact
  you need to have all the dependencies compiled shared, of
  course. And HDF5 compiles in shared mode only under certain
  circumstances. Most notably, HDF5 does *not* compile in shared mode
  and parallel/MPI at the same time: serial can be shared, parallel
  can only be static. This is a known issue for [HDF5], but it's unclear
  if they'll have the resources to tackle it. The fact that pathscale
  and gcc "worked" was an accident of the way we compiled HDF5 with
  those compilers.

--Russ



Ticket Details
===================
Ticket ID: XZZ-151006
Department: Support netCDF
Priority: Critical
Status: Open