The following are brief installation instructions (from source code) for Numarray. Since Numarray uses Python's distutils, this should be a fairly automatic process. We are not yet supplying windows binaries so having VC++ available is a requirement. We currently use Tim Peters' doctest for regression testing. ============================================================== 1. UNIX: We have tested on several UNIX platforms including: * Linux-x86 (RedHat 8.0/ gcc-3.2 / Python-2.2.2) * Solaris-UltraSparc (Solaris 8 / Sun C / Python-2.2.2) * Tru64-Alpha (OSF1 / v5.1 / DEC C / Python-2.2.2) Installation ok, works fine. Issues with over- and underflow in test-suite. ========================================================================== 2. UNIX: Unpack the distribution tarball into a directory: % gunzip -c numarray-xxx.tar.gz | tar xf - % cd numarray-xxx ========================================================================== 3. UNIX: Build Numarray using the distutils: Do a default installation if you can: % python setup.py config install --gencode This will try to install numarray in python's site-packages directory. Since site-packages is already on PYTHONPATH you're ready to selftest. ----- OR ------ For more flexibility (if you need to do a "user" level install): % python setup.py install --gencode --install-lib= --install-headers=/numarray which will install everything under /numarray, with the exception of the file 'numarray.pth' which is stored in . Since you picked , you need to: setenv PYTHONPATH "/numarray:$PYTHONPATH" or: export PYTHONPATH="/numarray:$PYTHONPATH" or: cp numarray/Doc/sitecustomize.py setenv PYTHONPATH The latter solution lets you set up as a user-level site-packages. See the comments in site-customize.py for more explanation. In short, you only have to add to PYTHONPATH to support *multiple* packages located within . ----- OR ------ For even more installation pleasure: % python setup.py config install --gencode --prefix= for Windows and Unix (on Unix the installation will go into: /lib/python2.x/site-packages/numarray) ----- OR ------ Starting with version 0.41, numarray also supports two step installation: python setup.py config build --gencode python setup.py install (--gencode should be omitted!) ----- OR ------ For Unix this form will probably be more convenient: % python setup.py config install --gencode --home= (ends up in the lib/python directory) Lastly, if you need a hammer, add --force to force setup to install the files even if it thinks they're already up to date. Another frequently usefuly technique is to completely delete your *old* numarray installation prior to installing the *new* one. To see more distutils options: % python setup.py --help For the latest distutils documentation from the Python crew look at: http://python.sourceforge.net/devel-docs/inst/inst.html ============================================================== 5. UNIX: Self-test Numarray: [jmiller@halloween ~/numarray-0.6]$ python Python 2.3 (#3, Jul 30 2003, 13:30:29) [GCC 3.2 20020903 (Red Hat Linux 8.0 3.2-7)] on linux2 Type "help", "copyright", "credits" or "license" for more information. >>> import numarray.testall as testall >>> testall.test() numarray: ((0, 1123), (0, 1123)) numarray.records: (0, 48) numarray.strings: (0, 166) numarray.memmap: (0, 75) numarray.objects: (0, 72) numarray.examples.convolve: ((0, 20), (0, 20), (0, 20), (0, 20)) numarray.convolve: (0, 42) numarray.fft: (0, 75) numarray.linear_algebra: (0, 46) numarray.image: (0, 2) numarray.image.combine: (0, 15) numarray.random_array: (0, 53) numarray.ma: (0, 671) The numbers (tuples) in the results represent the number of errors and number of tests executed for each numarray module, e.g.. there were zero errors detected in 1123 tests of numarray. The number of test cases is likely to change in the future, so don't be alarmed if your counts don't match exactly. Tuples of tuples indicate multiple passes of the same basic test, for various reasons. ============================================================== Win-32: (Installation from source) ============================================================== 0. Win-32: To install numarray, you need to be in the Administrator account. As a general rule, always remove (or hide) any old version of numarray before installing the next version. 1. Win-32: We have tested Numarrray on several Win-32 platforms including: Windows-XP-Pro-x86 ( MSVC-6.0) Windows-XP-Home-x86 ( MSVC-6.0) Windows-NT-x86 ( MSVC-6.0) Windows-98-x86 ( MSVC-6.0) 2. Win-32: First, unpack the distribution: (NOTE: You may have to download an "unzipping" utility) C:\> unzip numarray.zip C:\> cd numarray 3. Win-32: Build it using the distutils defaults: C:\numarray> python setup.py install --gencode This installs numarray in C:\pythonXX where XX is the version number of your python installation, e.g. 20, 21, etc. 4. Win-32: Once you have installed numarray, test it with: C:\numarray> python Python 2.3 (#3, Jul 30 2003, 13:30:29) [MSC 32 bit (Intel)] on win32 Type "copyright", "credits" or "license" for more information. >>> import numarray.testall as testall >>> testall.test() numarray: ((0, 1123), (0, 1123)) records: (0, 48) strings: (0, 166) memmap: (0, 75) objects: (0, 72) numarray.examples.convolve: ((0, 20), (0, 20), (0, 20), (0, 20)) numarray.convolve: (0, 42) numarray.fft: (0, 75) numarray.linear_algebra: (0, 46) numarray.image: (0, 2) numarray.image.combine: (0, 15) numarray.random_array: (0, 53) numarray.ma: (0, 671) The numbers (tuples) in the results represent the number of errors and number of tests executed for each numarray module, e.g.. there were zero errors detected in 1123 tests of numarray. The number of test cases is likely to change in the future, so don't be alarmed if your counts don't match exactly. Tuples of tuples indicate multiple passes of the same basic test, for various reasons. ============================================================== Win-32: (Installation from self-installing executable) ============================================================== 0. Win-32: To install numarray, you need to be in the Administrator account. As a general rule, always remove (or hide) any old version of numarray before installing the next version. 1. Click on the executable's icon to run the installer. 2. Click "next" several times. I have not experimented with customizing the installation directory and don't recommend changing any of the installation defaults. If you do and have problems, let me (jmiller@stsci.edu) know. 3. Assuming everything else goes smoothly, click "finish". 4. Test numarray as above in Win-32: (Installation from Source) ============================================================== Special Notes: ============================================================== 1. i386 linux -- if compiling against GNU libc on i386 and enabling SSE processor functions (with something like "-march=athlon-xp" or using other libraries that utilize SSE such as atlas or Intel IPP) then libc version 2.3.3 or above will be needed.