File Release Notes and Changelog
Release Name: 1.5.1
Notes: Release Notes for numarray-1.5.1 I. Comments 1. NumArray Operator Optimization Bug numarray-1.4.0 moved all NumArray operator "hooks" from Python down into C to make operators (e.g. +) as fast as ufuncs (e.g. add()). A number of users doing numarray applications in C or using pickling reported problems. This bug has been fixed and is the predominant motivator for this release. 2. Better NumPy dtype Support. I refined numarray's support for numpy-style numerical dtypes and added the new dtype attribute: >>> a = arange(10, dtype='i') >>> a = arange(10, dtype='i4') >>> a = arange(10, dtype='int32') >>> a = arange(10, dtype='<i4') >>> a = arange(10, dtype=numarray.dtype.int32) >>> a.dtype.char # i.e. int32 'i' >>> a.dtype.kind # i.e. integer 'i' >>> a.dtype.itemsize 4 >>> a.dtype.byteorder # little endian '<' >>> a.dtype.str '<i4' >>> a.dtype.type Int32 >>> a.dtype.isbuiltin True >>> a.dtype.isnative True This is still pretty superficial support for dtype and omits specification of records, strings, and objects. Although numarray's dtype instances record byteorder, only native byteorder arrays can be constructed at present. II. Related Source Forge Tracker Items 1396372 UInt32 AMD64 crash 1412737 NumarrayType enumeration problem 1395938 AMD64 Clipping, casting, and maximum values 1396742 Slow performance in array protocol with string arrays 1397667 Add scipy newcore numpy .dtype attribute 1386807 sum(a boolean array) returns a negative value 1387438 UFunc disease spreads (C operator hook bug) See http://sourceforge.net/tracker/?atid=450446&group_id=1369&func=browse for more details. III. CAUTIONS This release should be binary compatible with numarray-1.3.x. New extensions compiled against 1.5.1 may not work with 1.4.x or earlier versions of numarray.
Changes:
