On 25 May 2010 15:39, Tony Theodore <to...@logyst.com> wrote: > On 25 May 2010 13:23, Brandt Belson <bbel...@princeton.edu> wrote: >> Hi again, >> Ok, I installed numexpr and the newest Cython, and now the latest trunk >> version gives another error related to hdf5. > > [...snip] > >> .. ERROR:: Could not find a local HDF5 installation. >> You may need to explicitly state where your local HDF5 headers and >> library can be found by setting the ``HDF5_DIR`` environment >> variable or by using the ``--hdf5`` command-line option. >> error: Setup script exited with 1 >> I have set the HDF5_DIR env variable to the right value, along with > > [snip...] > > Most of what you're doing looks like it should work. You seem to be > using python from python.org and distutils is calling gcc4.0. > > There are some things I'd try: > > Check output of gcc -v. If it's 4.2 it will default to buidling a > 64-bit hdf5 that won't be usable. You'll have to specify either -arch > i386 or CC when configuring. > > Use the file command to determine the archs of > ../hdf5-1.8.4/hdf5/lib/libhdf5.dylib (or similar). > > Try installing hdf5 to /usr/local, I've generally had the most success > with this. > > I'm not at a Mac at the moment, I'll have a look later on.
Ok, so I think I've figured out a simple general approach, but haven't tested it. The main problem is the mix of available python interpreters and different arch defaults for compilers. Check python arch: [1] $ file `which python` /Library/Frameworks/Python.framework/Versions/Current/bin/python: Mach-O executable i386 For comparison, the Apple supplied python: [2] $ file /usr/bin/python /usr/bin/python: Mach-O universal binary with 3 architectures /usr/bin/python (for architecture x86_64): Mach-O 64-bit executable x86_64 /usr/bin/python (for architecture i386): Mach-O executable i386 /usr/bin/python (for architecture ppc7400): Mach-O executable ppc In the case of [1], you're lucky, you simply need to compile an i386 library. Configure hdf5 (and any others - lzo, szip etc) as follows: env CC=gcc-4.0 ./configure --prefix=/usr/local Using /usr/local is reasonably fail-safe; you don't need to pass anything else to pytables, specify library paths, and easy_install/distutils can find it. Setting CC is the quick way to get a default i386 binary, you could also specify CFLAGS='-arch i386' and LDFLAGS='-arch i386'. You can also pass --enable-shared --disable-static to configure if you like. I believe the the python.org version is a universal i386/ppc binary, so the same applies. In the case where the python on your path includes x86_64, you have to know a little more about what changes you've made to the system. On a vanilla system with just OSX and XCode, the following should work for hdf5: ./configure --prefix=/usr/local If you've been playing with paths, defaults write *.* Prefer-32-Bit, gcc versions etc., you'll probably want to reverse those changes and force an x86_64 build: env CC=gcc-4.2 ./configure --prefix=/usr/local or env CC=gcc-4.2 CFLAGS='-arch x86_64' LDFLAGS='-arch x86_64' ./configure --prefix=/usr/local I've never been able to get an i386 build working with an x86_64 python, I'm sure it's possible, but I'd need to know a lot more about universal binaries and distutils than I have time for. After installing hdf5, confirm: $ file /usr/local/lib/libhdf5.dylib is the correct arch. Once that is done, the various ways of installing pytables (tarball, svn, easy_install) usually "just work". Hope that helps, I'll try and put some notes on the wiki since this comes up often. Tony ------------------------------------------------------------------------------ _______________________________________________ Pytables-users mailing list Pytables-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/pytables-users
