HDF5

Installing

Mac OS X Lion: Use Brew

If you're on Mac OS X Lion (10.7), you may run into issues. If you're less inclined to get elbow-deep in computer problems, you can just use Homebrew (http://mxcl.github.io/homebrew/), a really nice package manager for Mac OS X. MacPorts and Fink can both be nightmares, but Homebrew manages to do everything hassle-free.

You can install HDF5 using Homebrew by running:

$ brew install hdf5

UPDATE: HDF5 has since moved to Homebrew-Science (https://github.com/Homebrew/homebrew-science). You can "tap" into Homebrew Science to get to the HDF5 recipe.

Configuring

I am using the HDF5 libraries for various software tools available through the CRSim Software repository http://software.crsim.utah.edu/ - specifically, C++ programs using the C++ interface to HDF5 - necessitating the --enable-cxx configure argument:

./configure \
 --prefix=/path/to/hdf5 \
 --enable-cxx

You can optionally use the configure arguments --enable-static-exe --enable-static, though in my experience these aren't necessary to build (and link to) HDF5.

Errors

You may run into problems during make check, something looking like this:

Testing hard normalized long double -> signed char conversions        Command terminated by signal 11
0.31user 0.04system 0:01.64elapsed 21%CPU (0avgtext+0avgdata 19936maxresident)k
824inputs+360outputs (6major+14996minor)pagefaults 0swaps
make[4]: *** [dt_arith.chkexe_] Error 1

The solution, as mentioned here http://kartadikaria.wordpress.com/2011/10/06/how-to-install-hdf5-macintosh-lion/, is to modify the configure flags for GNU compilers (the one used on Mac, if you're using gcc). Change the PROD_CFLAG for gcc compilers version 4 and up to use -O0 instead of the default -O3.

Using

Python and HDF5

Link to docs: http://docs.h5py.org/en/latest/

Link to quickstart: http://docs.h5py.org/en/latest/quick.html

Start by creating an HDF5 file object:

>>> import h5py
>>> import numpy as np
>>>
>>> f = h5py.File("mytestfile.hdf5", "w")

Now that you have a file object, create data sets:

>>> dset = f.create_dataset("mydataset", (100,), dtype='i')

(This is not a file, but a dataset object - this is similar to a Numpy array.)

HDF5 files are stored in a hierarchical way, with a "directory"-like structure.

>>> dset.name
u'/mydataset'

The file object is the root group, equivalent to / on a Unix system:

>>> f.name
u'/'