[Matrix-SIG] LLNLPython2 released

Paul F. Dubois Paul F. Dubois" <dubois1@llnl.gov
Fri, 5 Jun 1998 15:50:08 -0700

ftp://ftp-icf.llnl.gov/pub/python/LLNLPython2.tgz is now available. Also
available are windows installers for NumPy and RNG.

It contains the beta release of CXX, Release 1.2 of NumPy, and all the other
LLNL-distributed packages. The file README.html in that same directory
explains the different packages.

Documentation for each package is either included in a Doc subdirectory or
is at:
http://xfiles.llnl.gov. Below are the release notes which are available as
HTML files in the release.

LLNLPython is distributed as a public service by X-Division at Lawrence
Livermore National Laboratory. Please read the legal notice which explains
in more detail that we are doing this only as a public service and are not
responsible for this open-source community-based software, which you use at
your own risk.

Release 2 (June, 1998)
Several "Setup" files had a space after the *shared* which mixed up FreeBSD.
(Thomas Gellekum).
See Numerical/README.htm (below)
Bug fixes in Gist3D/Demo/gistdemohigh.py, Gist3D/Lib/quadmesh.py (Thomas

Numerical Python Version 1.2
Release Notes (June 5, 1998) Paul F. Dubois
Changes from version 1.1:
General Changes
Lib/Numeric.py: Added very inefficient (but working) pickling of arrays
based on the copy_reg
mechanisms. Should be replaced by a __reduce__ method someday. (David
Made comparisons of NumPy arrays raise TypeErrors.
Fixed slice bug reported by Andrew Mullhaupt and fixed by Chris Chase a very
long time ago but never integrated into the souce. See:
http://www.findmail.com/listsaver/matrix-sig/?start=464 (David Ascher).
Unfortunately, the fix itself had an error, so I fixed the fix. (Dubois).
Removed a memory leak in a slicing routine. (Dubois)
Removed a space in Setup that was confusing FreeBSD. (Thomas Gellekum).
Fixed the following operators: <<, >>, &, |, ^.  These had been totally
nooperative due to an error in which the implementing functions were
transferred from one module to another via a dictionary lookup and the name
used was wrong but the lookup status was not checked. The internal names are
now left_shift, right_shift, bitwise_and, bitwise_or, and bitwise_xor.
Henk Jansen pointed out that UserArray has a typecode() method that is
destroyed by assignment to self.typecode. Instead, the data member was
renamed self._typecode and typecode() taught to use it, as he suggested. I
didn't try to get into the other issues about UserArray that have been
raised. (Dubois)
MacIntosh changes (Rob Managan & Jack Jensen)
In f2c.h we added a #ifndef so abs is not redefined.
#ifndef abs
#define abs(x) ((x) >= 0 ? (x) : -(x))
in fast_umath and umath we just had to replace <math.h> with "mymath.h".
mymath.h basically does some stuff before including math.h due to funny
stuff CW does for 68K machines. Mainly hypot crashes since it returns 10
byte floats. This file is included in Jack's stuff, not NumPy.
In zlapack_lite Jack Jensen added a hack to allow one long routine to
compile for 68K machines. Not sure if that is well tested or not.