Fix SUNDIALS v7.4 Krylov solver segfaults on macOS

[ChrisRackauckas]

Summary

This PR fixes segmentation faults that occur when using Krylov iterative linear solvers with SUNDIALS v7.4 on macOS.

Problem

The Julia wrappers for SUNDIALS Krylov solvers were missing the required SUNContext parameter introduced in SUNDIALS 6.0+. This caused immediate segmentation faults on macOS when attempting to use GMRES, FGMRES, BCG, PCG, or TFQMR linear solvers with KINSOL, CVODE, or IDA.

Solution

• Added the missing SUNContext parameter to all Krylov solver constructor calls
• Updated the C library wrapper definitions to properly pass the context through to SUNDIALS

Changes Made

Files Modified:

1. src/simple.jl - Added ctx parameter to all Krylov solver constructors in KINSOL interface
2. src/common_interface/solve.jl - Added ctx parameter for CVODE, CVODES, and IDA Krylov solver calls
3. lib/libsundials_api.jl - Fixed ccall signatures to include SUNContext parameter

Affected Solvers:

• SUNLinSol_SPGMR (GMRES)
• SUNLinSol_SPFGMR (FGMRES)
• SUNLinSol_SPBCGS (BCG)
• SUNLinSol_PCG (PCG)
• SUNLinSol_SPTFQMR (TFQMR)

Testing

Tested on macOS (Darwin 24.6.0, Apple Silicon) with SUNDIALS v7.4.1:

✅ All 5 Krylov solvers now work with KINSOL
✅ All 5 Krylov solvers now work with CVODE
✅ All 5 Krylov solvers now work with IDA
✅ No more segmentation faults

Test script confirming all solvers work:

# All tests pass - 15 PASSED, 0 FAILED
# GMRES, FGMRES, BCG, PCG, TFQMR all work with CVODE, KINSOL, and IDA

Impact

This fix is critical for macOS users as it resolves crashes when using any Krylov iterative solver. The changes are backward compatible and follow the SUNDIALS 7.x API requirements.

Fixes segmentation faults reported when running test suite on macOS with SUNDIALS v7.4.

🤖 Generated with Claude Code

[ChrisRackauckas]

@ViralBShah Mac should be fine now. Not sure how other tests worked without it 😅 but the memory context is local now on those.

[ViralBShah]

Still failing on mac. I have been using Julia 1.11.6 (but fails on 1.10 and 1.12 as well):

WARNING: Method definition f_oop(Any, Any) in module Main at /Users/viral/.julia/dev/Sundials/test/kinsol_banded.jl:22 overwritten at /Users/viral/.julia/dev/Sundials/test/kinsol_nonlinear_solve.jl:38.
Test Summary: | Pass  Total  Time
Kinsol        |  131    131  2.7s

[11518] signal 11 (2): Segmentation fault: 11
in expression starting at /Users/viral/.julia/dev/Sundials/test/handle_tests.jl:39
SUNSparseMatrix at /Users/viral/.julia/artifacts/a67d3539b46859461a015e756e14f4707979c8dd/lib/libsundials_sunmatrixsparse.5.4.0.dylib (unknown line)
SUNSparseMatrix at /Users/viral/.julia/dev/Sundials/lib/libsundials_api.jl:8858
unknown function (ip: 0x136fdc157)
jl_apply at /Users/julia/.julia/scratchspaces/a66863c6-20e8-4ff4-8a62-49f30b1f605e/agent-cache/default-honeycrisp-R17H3W25T9.0/build/default-honeycrisp-R17H3W25T9-0/julialang/julia-release-1-dot-11/src/./julia.h:2157 [inlined]
do_call at /Users/julia/.julia/scratchspaces/a66863c6-20e8-4ff4-8a62-49f30b1f605e/agent-cache/default-honeycrisp-R17H3W25T9.0/build/default-honeycrisp-R17H3W25T9-0/julialang/julia-release-1-dot-11/src/interpreter.c:126

[ViralBShah]

Minimal reproducer:

julia> using Sundials, Test

julia> Sundials.SUNSparseMatrix(3, 3, 3, Sundials.CSC_MAT)

[15022] signal 11 (2): Segmentation fault: 11
in expression starting at REPL[2]:1
SUNSparseMatrix at /Users/viral/.julia/artifacts/a67d3539b46859461a015e756e14f4707979c8dd/lib/libsundials_sunmatrixsparse.5.4.0.dylib (unknown line)
SUNSparseMatrix at /Users/viral/.julia/dev/Sundials/lib/libsundials_api.jl:8858
unknown function (ip: 0x1265d4157)

[ChrisRackauckas]

Tests on mac pass on my computer, and pass in #489

[ViralBShah]

The SUNSparseMatrix constructors are missing the Suncontext in the sundials_api.

I don't quite understand why it is working on linux and CI, but it is probably getting lucky with some default code paths. I am able to consistently crash with that one liner on my mac.

[ChrisRackauckas]

It is interesting that it doesn't consistently segfault, but it can if the context is missing. This new SunContext is pretty silly.

[ViralBShah]

Since we are calling it with fewer arguments, it is either somehow getting a context luckily or NULL is acceptable or something else that is acceptable is getting picked up.