CHANGES

Fixed to locate numpy header files correctly

Automate installation of numpy (no flags to change in setup.py)

Removed some Numeric imports

Moved version number from __init__.py to __version__.py

1.0
--------------------------------------------------------------------------------
Moved modules inside "pysparse" name space.

Updated to use the new numpy, but still compatible with the old Numeric.


0.34.032
---------------------------------------------------------------------------------
Fixed a memory leak in ll_mat.keys(). (Reported by c chui)

Improved the performance of ll_mat.scale(), ll_mat.values() and ll_mat.items().

0.34.031
---------------------------------------------------------------------------------
Modules using the pysparse C API can now define NO_IMPORT_SPMATRIX
together with SPMATRIX_UNIQUE_SYMBOL to avoid multiple definition of
the SpMatrix_API pointer. This fix was suggested by Erica Madeleine Ligner.

0.34.030
---------------------------------------------------------------------------------
Applied patches 1089844, 1089837, 1089831 (supplied by Daniel Wheeler
& Jonathan Guyer): Added three new ll_mat methods:
update_add_at(b,id1,id2), put(b,id1,id2), take(b,id1,id2), fixed OS X
compatibility issues, Numpy compatibility fixes.

Applied some additional fixes for 64-bit machines.

PySparse now works with Numeric 23.7 (also on 64-bit machines).

[1111727] Replaced #include <malloc.h> by #include <stdlib.h> to fix
compatibility issue on OS X (reported by Jonathan Guyer).

Removed reference to libm.a which broke compilation on Win32 (reported
by Mark Dausch).

0.33.029
---------------------------------------------------------------------------------
added new umfpack module. (by Nathaniel C. Domingo)

added Umfpack library source files to distribution (by Nathaniel C. Domingo)

added new module sparray, implementing a multidimensional sparse array
(by Pascal Barbedor)

fixed bug in SSOR preconditioner (precon module) for omega != 1.0 and
steps > 1

0.32.028
---------------------------------------------------------------------------------
added Xavier Vasseur's gmres solver to the itsolvers module. Thanks Xavier!

0.31.027
---------------------------------------------------------------------------------
fixed: call to SpMatrix_MATVEC in Src/pcg.c

0.31.026
---------------------------------------------------------------------------------
PySparse is placed under a BSD style license

0.30.025
---------------------------------------------------------------------------------
fixed bug in "generalize" member function of ll_mat type. Reported by
Xavier Vasseur.

the result tuple returned by jdsym.jdsym now includes an additional
value representing the avg. number of inner iterations. This change
may be incompatble with existing code calling this module function.

0.29.024
---------------------------------------------------------------------------------

PySparse can be build using MinGW on Win32 platforms. Check the
instructions in INSTALL.

0.29.023
---------------------------------------------------------------------------------

fixed memory leaks in SpMatrix_LLMatSetItem and
SpMatrix_LLMatUpdateItemAdd that occur if PyMem_Resize would fail.

More detailed installation instructions for Win32 platforms in INSTALL

0.29.022
---------------------------------------------------------------------------------
new method for ll_mat objects: compress

nicer output for ll_mat objects representing small matrices

optimised matrix-matrix-multiplication routine included (unsym-unsym case)

new version of bicgstab solver (uses less memory)

0.28.021
---------------------------------------------------------------------------------
fixed error handling in ll_mat.generalize

fixed name mismatched in solver kernels in itsolvers

0.28.020
---------------------------------------------------------------------------------
new method itsolvers.bicgstab (implemented by Oliver Broeker)

all solvers in the itsolvers module now check for exceptions after
each precon or matvec call.

solver kernels in itsolvers now have names that conflict less likely
with other modules

0.28.019
---------------------------------------------------------------------------------
uncommented call to PrintStat in superlu/dgssv.c and superlu/dgssvx.c
to prevent unwanted output in ML package

fixed memory leak in sss_mat.c: diag array was not freed

new method for ll_mat objects: generalize

0.27.018
---------------------------------------------------------------------------------
updated test modules to reflect earlier bug fix in poisson.poisson2d()

new default precision in ll_mat.export_mtx(): 16 instead of 6

fixed bug in Src/jdsymmodule.c: solution now copied correctly if not
all requested eigenvectors converged.

0.27.017
---------------------------------------------------------------------------------
fixed bug in SRC/cgs.c: now returns correct iteration count, if initial guess
satisfies stopping criterion

fixed bug in Src/superlumodule.c solve transposed system, since matrix was 
passed row-wise

new method for ll_mat matrices: norm(p)

0.26.016
---------------------------------------------------------------------------------
changed interface of SpMatrix_Matvec(), now supports non-square matrices

0.26.015
---------------------------------------------------------------------------------
reimplemented SpMatrix_LLMatBuildColIndex and SpMatrix_LLMatDestroyColIndex

0.26.014
---------------------------------------------------------------------------------
new module itsolvers_util: encapsulation objects for all iterative solvers

0.25.013
---------------------------------------------------------------------------------
fixed bug in poisson.poisson2d_sym_blk(): diagonal entries are now 4, not 2.

changed superlu_superlu_timer.c to support Win32

0.25.012
---------------------------------------------------------------------------------
test module for superlu

poisson module for generating 1D and 2D Poisson matrices

0.25.011
---------------------------------------------------------------------------------
incorporated a new version of SuperLU (date 2002/09/23)

changed postorder tree traversal in sp_coletree.c to be more stack friendly
(max stack size on Linux seems to be 6 MBytes)

removed stdout in get_perm_c.c

0.25.008
---------------------------------------------------------------------------------
fixed bug in matvec method of csr_mat, occuring if stride of vector y is not one

ll_mat.shift now supports shift of a general matrix by a symmetric matrix

new matvec and matvec:transp object methods for ll_mat type

fixed bug when assigning a symmetric matrix to a block of a general matrix

delete_cols method added for ll_mat objects

0.24.007
---------------------------------------------------------------------------------
fixed bug occurring when the spmatrix C API is used in an extension 
consisting of more than one file. Now SPMATRIX_UNIQUE_SYMBOL has
to be defined. This broke the itsolvers.pcg function.

0.24.006
---------------------------------------------------------------------------------
Bug fixed in spmatrix.h: LLMatType now exported properly

Genereation of boiler plate code for C API is automated now

More functions exported in C API

spmatrix module no longer uses extern symbols

delete_rows and delete_rowcols methods added for ll_mat objects

matvec and matvec_transp methods of csr_mat and sss_mat objects now 
accept strided arrays as input

0.23.005
---------------------------------------------------------------------------------
Header files are now installed to python directory tree. External
packages may now import pysparse.

0.23.004
---------------------------------------------------------------------------------
new module Lib/pysparse_version.py defining pysparse version number

0.23.003
---------------------------------------------------------------------------------
Updated README

Removed some illegal characters from superlu/superlu_timer.c

Removed SuperLU output from superlu/get_perm_c.c stating the matrix
reordering used.


0.23.002
---------------------------------------------------------------------------------
SuperLU is now incorporated into pysparse. No need to build the
library separately from now on.

Now compiles under Win32. Tested under WinXP with MS Visual Studio 6.0
and the Inter Math Kernel Library for LAPACK and BLAS

Removed some unused local variables

Changed interface of LLMat_parse_index. Can now be safely used in sss_mat.c.


0.22.001
---------------------------------------------------------------------
initial release