User Software :: CR-Tools :: Installation on Mac OSX 10.6 64 bit (Snow Leopard)

1. Introduction
2. Prerequisites
   • C/C++ compiler and tools
   • Fortran compiler
   • CMake
   • Python
   • Numpy
   • Scipy
   • Matplotlib
3. Installing CR-Tools

The following are instructions for installing CR-Tools from source on Mac OSX 10.6 commonly known as Snow Leopard. While the system kernel itself does not run in 64 bit mode by default, all libraries are compiled 64 bit and the compiler is set to 64 bit architecture by default. Therefore these instructions also compile all the USG software in 64 bit, this may or may not give a speed increase.

We assume a clean system (factory default install of Snow Leopard and nothing else), if this is not the case and you run into architecture mismatch problems please make sure the compiler can find the right libraries.

Before installing the USG software you need to install some other tools and libraries.

Warning Do not use fink or macports to install the prerequisites, this is known to give architecture mismatch problems during linking!

First install the GNU C/C++ compiler optimized for OSX provided by Apple as part of the X Code Integrated Development Environment.

You can either use the OSX Installation DVD which came with your mac or download it from the Apple website.

Double click the package and follow the installation instructions. Then run Software Update from the Apple menu to ensure you have the latest version of all libraries.

1. Install the 64 bit optimized version of the GNU fortran compiler from the High Performance Computing project. Make sure you select the gfortan only binaries which can be found here.
2. Then

   gunzip gcc-bin.tar.gz

   (if your browser didn't do so already) and

     sudo tar -xvf gcc-bin.tar -C /.

   It installs everything in /usr/local.

3. You can test if this worked by typing

   gfortran

CR-Tools uses the CMake build system. This tool needs to be installed separately. Simply download the binary from here, double click to mount the disk image and follow the installation instructions.

There are three ways to install Python. You may use the Enthought Python Distribution binary installer (EPD 6.3 64 bit) DMG (recommended), fink or you may compile Python and all required packages from source manually.

First download the source code from here.

Unpack the tar.gz archive if this hasn't been done automatically, e.g. via

tar -xvzf fink-0.29.21.tar.gz

if it has already been partially unpacked, in a terminal window. Then, in a terminal window, change to the resulting fink-0.29.21 directory, and use

./bootstrap

to start the boostrapping operation, which will install the Fink base setup. When you get the following question: “Your hardware is a 64bit-compatible intel processor, so you have the option of running Fink in 64bit-only mode. This is not recommended for most users, since many more packages are available for the default mode (which is mostly 32bit but includes some 64bit packages). Which mode would you like to use?”

(1)     Default (mostly 32bit)
(2)     64bit-only

Be sure to select option 2 to build the 64 bit version.

Choose a mode: [1] 2

After the installation is completed, running the command

/sw/bin/pathsetup.sh

will set up your environment for Fink (assuming you have installed Fink under /sw). If you open a new terminal window, the session will use these environment settings. Once you have installed fink and the other base packages, the command sequences:

fink selfupdate-rsync
fink index -f

If you have not done so during the bootstrap, switch your fink installation to use the unstable tree by running

fink configure

and selecting the appropriate option. Then run the following commands to update the index and rebuild all packages.

fink selfupdate
fink selfupdate-rsync
fink index -f
fink scanpackages

Finally install the required packages using the following command:

fink install numpy-py26 scipy-py26 matplotlib-py26 pyfits-py26 ipython-py26

Unfortunately Fink also automatically installs a basic python 2.7 which does not work with the current version of boost python. Therefore remove this installation using:

sudo fink remove -r python27 python27-shlib

Also make sure that python points to the default python by creating the correct symlink.

sudo ln -s /sw/bin/python2.6 /sw/bin/python

And check if your path is set correctly

which python

should give you

/sw/bin/python

If you want to build the Python Pypeline you will need to install the latest 2.6 series version of Python. Unfortunately the binaries for OSX Snow Leopard are compiled as 32 bit. So we need to download the source code and compile/install it ourselves. Download the source code from here. Then extract the tarball with

tar xvf Python-2.6.5.tgz

The compilation steps are a bit non standard. First we need to build Python as a framework and tell the compiler to compile both the 32 and 64 bit versions.

./configure --enable-framework MACOSX_DEPLOYMENT_TARGET=10.6 --with-universal-archs=intel --enable-universalsdk=/Developer/SDKs/MacOSX10.6.sdk

To compile run

make

or to compile with debug symbols

make OPT=-g

and finally install.

sudo make install

Then we need to make sure the 64 bit binary is loaded by default. Open the python bin directory /Library/Frameworks/Python.framework/Versions/2.6/bin/. In it, you'll find 32-bit and 64-bit intel binary symbolic links. Relink the 32-bit with the 64-bit binaries.

sudo ln -sf python2.6-64 python
sudo ln -sf python2.6-64 python2.6
sudo ln -sf pythonw2.6-64 pythonw
sudo ln -sf python2.6-64 pythonw2.6

You can test if this worked by starting python and printing the maximum value of an int using

import sys
print sys.maxint

If you are running 64-bit, it will be 9223372036854775807, for 32 bit it will print 2147483647, if the latter is the case please check what went wrong before continuing.

Download the source code for the latest stable version of numpy here. Then extract and install using

tar xvf numpy-1.4.1.tar.gz
cd numpy-1.4.1
sudo python setup.py install

1. Download the source code for the latest stable version of scipy here.
2. Extract the tar-archive:

   tar xvf scipy-0.7.2.tar.gz

3. Build and install Scipy:

   cd scipy-0.7.2
   sudo python setup.py install

Compiling a 64bit version of matplotlib is a bit more involved. We need the latest SVN checkout of the matplotlib source code.

svn co https://matplotlib.svn.sourceforge.net/svnroot/matplotlib/trunk/matplotlib matplotlib

(it has been known to work with revision 8337). Then we need to compile and install matplotlib.

cd matplotlib
PREFIX=/usr/local sudo make -f make.osx fetch deps mpl_build mpl_install
sudo python setup.py install

You can test if this worked by starting python and plotting an image as follows:

python
>>> import matplotlib.pyplot as plt
>>> import numpy as np
>>> x=np.linspace(0,2*np.pi,100)
>>> y=np.sin(x)
>>> plt.plot(x,y)
[<matplotlib.lines.Line2D object at 0x1021d9bd0>]

at this point you should have an image of a sine on your screen.

Now we are ready to install CR-Tools itself.

1. First get the latest source code from the LOFAR USG Subversion repository:

   svn co http://usg.lofar.org/svn/code/trunk lofarsoft

2. Then add the following line to your .profile or .bash_profile:

   export LOFARSOFT=/path/to/lofarsoft
   . $LOFARSOFT/devel_common/scripts/init.sh

   Don't forget to restart your shell or problems will arise along the way.
   NOTE on bash profile files:

   • .bashrc is not loaded by default on Mac OS X.
   • either .bash_profile or .profile is loaded by default, but only one of these. Therefore, make sure you have only one of the two.
3. Now bootstrap the build system:

   cd $LOFARSOFT/build
   ./bootstrap

4. In the build directory, run

   ccmake .

   and set WCSLIB_PYTHON_WRAPPER = OFF, DAL_PYTHON_BINDINGS = OFF.

5. Compile PyCRTools:

   make pycrtools

Grab a good lunch and if all is well, the build of PyCRTools and its dependencies completes in about an hour.

Afterwards, check if Numpy has been built by the build process. This is unwanted as we (normally) use Numpy from EPD. To prevent version conflicts (especially a segfault on importing pycrtools), remove it:

cd $LOFARSOFT/release/lib64/python
rm -rf numpy
rm numpy*

Test by running

python
import pycrtools