Lib sparse

[ViralBShah]

From #72

[codecov]

Codecov Report

❌ Patch coverage is 82.16561% with 28 lines in your changes missing coverage. Please review.
✅ Project coverage is 77.96%. Comparing base (b1d0917) to head (02df33c).

Files with missing lines	Patch %	Lines
src/libSparse/wrappers.jl	71.01%	20 Missing ⚠️
src/libSparse/AAFactorizations.jl	83.78%	6 Missing ⚠️
src/libSparse/AASparseMatrices.jl	96.07%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #90      +/-   ##
==========================================
+ Coverage   76.22%   77.96%   +1.73%     
==========================================
  Files           4        7       +3     
  Lines         265      422     +157     
==========================================
+ Hits          202      329     +127     
- Misses         63       93      +30     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ViralBShah]

@luke-kiernan Should we merge this?

[ViralBShah]

The error messages are slightly different in different macOS releases.

[luke-kiernan]

@luke-kiernan Should we merge this?

Sure! I would keep it as an experimental/developer feature, though. For my use case, I found that the performance is surprisingly poor, especially when compared to KLU.jl. I did some profiling and posted a question on the apple developer form, but never got an answer. But maybe someone else will find it more useful.

[ViralBShah]

Oh that's good to know. I'll mark this as a draft PR for folks who want to experiment. I agree that without a performance benefit, there isn't much value to adding more code/dependencies.

[luke-kiernan]

That's specifically the solvers. The sparse matrix multiplication might be better. The other place where there's clear gains: LibSparse has a sparse factorization for symmetric non-positive definite matrices, which would be a new addition (SuiteSparse doesn't have one).

[ViralBShah]

Coming to think of it - why not merge this? The tests are passing, and people can benchmark their own use cases and see what works for them.