[CIG-SHORT] Follow up on hdf5 issue while compiling

[Brad Aagaard]

On 11/04/2014 07:52 AM, Romain Jolivet wrote:
> Dear Brad et al.,
>
> Following my previous email:
> I compiled pylith using the installer on an iMac (brand new), running Mac os X
> 10.9 (not Yosemite, Mavericks).
> As things were not going that smoothly, I decided to enable every options
> available with the installer, but not gcc (I have gcc4.9 from macports).
>
> When compiling hdf5, it stops, saying:
>
> ../../../hdf5-1.8.11/tools/lib/h5tools_dump.c:635:9: error: expected expression
> before '/' token
>            //HGOTO_ERROR(dimension_break, H5E_tools_min_id_g, "Could not allocate
> buffer for ptdata”);
>
> The compiler used is mpicc (given what the Makefile says) that has been compiled
> using the pylith installer.
>
> I went inside the hdf5 code and modified the way comment were implemented at a
> few places, replacing // by the classic /* */.
> It compiled, but this small issue must hide something bigger because the tests
> fail in pylith due to some hdf5 opening problem in pets (see testmeshio.log file
> attached).


This is a clear HDF5 bug. C files should use C style comments /* */, not 
C++ style comments //. Your fix is correct. I am not sure why gcc 4.8.2 
let this pass when I built HDF5.

> The tests that fail are in unittests/libtests/meshio
> It also fails when running the 2d subduction examples and any other example that
> writes hdf5 outputs.
>
> Any idea? In the mean time, I will try with the 4.7 version of gcc from MacPorts
> (faster to install than the one from the installer),

What are the config arguments to the PyLith installer? I want to know if 
you are using the tarballs or repositories for PETSc and PyLith. If 
using the repositories, what branches are you using?

Brad