v2.0: remove almost all solver, preconditioner infrastructure in favor of LinearSolve

[j-fu]

(EDIT after #63e1238)
Besides removing old preconditioner structure, this PR overauls the general structure of the package.
It removes the old ExtendableSparseMatrixCSC struct in favor of GenericExtendableSparseMatrixCSC{SparseMatrixLNK}, thus using the previously introduced modular infrastructure as default. This is more flexible, and, hopefully, also easier to understand.

While formally breaking, I took care that the change is non-breaking for ExtendablFEMBase and ExtendableFEM. VoronoiFVM will need an upgrade to v3.0 in order to take this in.

Also, we have now Aqua, ExplicitImports and UndocumentedNames tests.

[pjaap]

Impressive! I found only a minor leftover in the code

[Copilot]

Pull request overview

This PR performs a major v2.0 refactoring of ExtendableSparse.jl, removing the old solver and preconditioner infrastructure in favor of relying primarily on LinearSolve.jl. The key architectural change is replacing the old ExtendableSparseMatrixCSC struct with a modular GenericExtendableSparseMatrixCSC{SparseMatrixLNK} approach, making the package more flexible and easier to understand.

• Removes deprecated factorization and preconditioner infrastructure
• Restructures to use GenericExtendableSparseMatrixCSC with pluggable sparse matrix extensions
• Adds comprehensive code quality checks (Aqua, ExplicitImports, UndocumentedNames)

Reviewed changes

Copilot reviewed 69 out of 70 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/ExtendableSparse.jl	Simplified main module, removed factorization exports, added @public macros
src/typealiases.jl	New file defining backward-compatible type aliases
src/abstractextendablesparsematrixcsc.jl	Enhanced documentation and API consistency with TYPEDSIGNATURES
src/sparsematrixlnk.jl	Fixed copy() implementation, updated documentation style
src/sparsematrixdilnkc.jl	Improved documentation, fixed field name consistency
src/sparsematrixdict.jl	Added comprehensive documentation, fixed I/J swap bug
src/preconbuilders.jl	Added BlockPreconditioner and JacobiPreconditioner implementations
src/sprand.jl	Updated to work with new abstract types
test/*.jl	Updated all tests to use new GenericExtendableSparseMatrixCSC API
docs/*.md	Restructured documentation to reflect new architecture
ext/*.jl	Removed deprecated Pardiso and AMG extensions

Comments suppressed due to low confidence (4)

src/sparsematrixdilnkc.jl:489

• There's an inconsistency in variable naming: the field is rowval (line 489) but was previously referenced as rowvals in the old code. While this change is correct (matching the actual field name), ensure all references throughout the codebase use rowval consistently.
  src/sparsematrixdict.jl:161
• The swap of I and J assignment appears to be a bug fix. Previously, I[i] = icsc and J[i] = rowval[j] meant columns were stored in I and rows in J, which is backwards for sparse matrix COO format where I typically holds row indices and J holds column indices. The fix correctly assigns I[i] = rowval[j] (rows) and J[i] = icsc (columns). However, verify that this doesn't break existing code that might have been compensating for the previous incorrect behavior.
  src/sparsematrixcsc.jl:2
• The documentation string uses $(TYPEDSIGNATURES) which generates detailed type signatures, but this appears on line 2 as a replacement for the previous simpler format. While this is more informative, ensure that the generated documentation doesn't become overly verbose or cluttered with internal implementation details that users don't need to see.
  src/sparsematrixlnk.jl:411
• In the copy function, direct field assignment is used without calling copy() on the arrays. This creates shallow copies where the new SparseMatrixLNK shares the same underlying arrays (colptr, rowval, nzval) with the original. This could lead to unexpected mutations if either the original or the copy is modified. Consider using copy(S.colptr), copy(S.rowval), and copy(S.nzval) to ensure deep copies.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.