Announcements

⚠️ COMPLEX256_T is no longer supported ⚠️

This is the new version with a new API.

Enjoy!

Nikolaj

I started to deal with sorted col/row indices for LL/CSC/CSR sparse matrices.

1. Unless there are some urgent needs (add the tag priority to the issues you want me to close urgently), I'll spend the next days (weeks) decoupling the code:

   • The template generation will be completely rewritten to ease the automation process (and avoid some recurrent human errors) and will be an autonomous project; ✅
   • The code generation process will be revisited to allow the use of the Cython debugger (it seems it is complicated to use the Python debugger on OSX). After one day trying... forget about cygdb. gbd itself works well and we now can add debug symbols with a switch in cysparse.cfg. ✅
   • The linalg part will be completely removed from CySparse and each interface with a solver will be an autonomous project (in their respective GitHub repositories). A common interface for all the solvers will be created and allow the interchange of solvers on the fly. ✅ (the linalg repositories still have to be created though).
   • CySparse itself will have its API changed to better reflect the common use in the community (See PySparse, NumPy and SciPy.sparse). ✅
   • A better mechanism will be implemented to allow the mix of special matrix cases (Symmetrical/general matrices, C-contiguous/non C-contiguous, etc). The aim is to introduce complex hermitian matrices (much later). Unfortunately, without real inheritance mechanism, I don't know what to do. So by default, this is done. :white_check_mark:

2. A centralized error/exception mechanism will be added.

3. Memory leaks will be taken seriously and hopefully I'll be able to avoid any memory leak (Valgrind anyone?). Some leaks were left on purpose: I wanted to test some Cython combinations before resolving these leaks (some ask for lots of work). These will be now taken care of.

4. There is a need for a better unit test generation: we need a meta meta generation. I have some ideas that I will test in the next weeks.

5. I have added sorted row/col indices for CSC/CSR matrices. This is a BIG change and needs some serious testing. You might experience some bugs because of this change... Let me know and I'll correct it. Thanks. This is still of actuality.

6. Documentation will be updated accordingly.

All this is very ambitious. I'll try to do it cleverly, step by step, knowing that I might get stuck somewhere before I can finish everything. Each step will be done in dedicated branches so that the master and develop branches can be used as usual. Changes will only be committed in major versions. So basically, there are two types of improvements going on: big non compatible changes (major versions) and business as usual (minor versions).