Rob Latham <robl@xxxxxxxxxxx> writes:
> On Mon, Oct 25, 2010 at 09:16:31AM -0600, Ed Hartnett wrote:
>> This problem was caused by an inadvisable call to H5close() inside the
>> netCDF library. Of course it was very silly of me to do this - how can
>> the netCDF library know if you are still using the HDF5 library?
>>
>> The call to H5close() was removed, and thus the problem will not occur
>> for you.
>
> I wonder if you added this close some time after this discussion?
I added the H5close well after that discussion. I don't think it is
related.
>
> http://mailman.unidata.ucar.edu/mailing_lists/archives/netcdf-hdf/2007/msg00107.html
>
> and if so, does removing this call have implications for parallel I/O ?
The only time this can matter is if a program is using netCDF-4 and HDF5
at the same time. (That is, a program is opening some HDF5 files with
the netCDF-4 API, and some with the HDF5 API.)
In this case, when the last netCDF file is closed, the netCDF library
will close the HDF5 library. This is, of course, a bad thing to do.
This will hurt parallel programs the same way that non-parallel
programs will be hurting: subseqent calls to open HDF5 objects will
fail, because closing the library deleted them from memory.
Thanks,
Ed
--
Ed Hartnett -- ed@xxxxxxxxxxxxxxxx
