Allow Int32 indices, switch to SuiteSparse 7

[rayegun]

No description provided.

[codecov-commenter]

Codecov Report

Merging #380 (e953cf0) into main (6ccc974) will decrease coverage by 0.40%.
The diff coverage is 80.41%.

❗ Current head e953cf0 differs from pull request most recent head cc518fa. Consider uploading reports for the commit cc518fa to get more accurate results

@@            Coverage Diff             @@
##             main     #380      +/-   ##
==========================================
- Coverage   93.97%   93.57%   -0.40%     
==========================================
  Files          12       12              
  Lines        7480     7520      +40     
==========================================
+ Hits         7029     7037       +8     
- Misses        451      483      +32     

Impacted Files	Coverage Δ
src/linalg.jl	96.09% <ø> (ø)
src/solvers/cholmod.jl	87.50% <80.14%> (-3.16%)	⬇️
src/solvers/spqr.jl	86.41% <88.88%> (ø)

... and 3 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[rayegun]

I'm going to have to test this on a 32-bit VM to get anywhere on this

[ViralBShah]

In case you don't already have one, the easiest thing is to get the ubuntu 32 bit docker image on a linux box.

[ViralBShah]

doctests need updating: https://buildkite.com/julialang/julia-master/builds/22783#01877398-57ae-43e1-8707-50d39367ca4b

Also most of the QR tests are failing on 32-bit, but cholesky seems ok: https://buildkite.com/julialang/julia-master/builds/22783#01877398-50ea-4bc7-aba6-87ce01031f4a

[rayegun]

Okay that's good info, I'm not surprised QR is a bit worse, I made much more cursory changes there. I'll update.

Will do doctests too.

[rayegun]

The newest version of SPQR will introduce a small regression for us, we have to convert HPinv from Int32->Int64->Int32 when Ti === Int64. There are no _l_ vs _ methods for int32_t in SPQR as far as I can tell.

[ViralBShah]

Is this breaking for the Julia user, or is it just that we need to adjust SPQR bindings to do some extra conversions so that tests pass?

[rayegun]

Just extra conversions.

[rayegun]

@ViralBShah to unblock progress I've temporarily taken out the SPQR 32-bit plat tests. These should "just work" after the upgrade to SS7.1 when Tim fixes this upstream.

I don't see another option rn

[ViralBShah]

I suppose we should wait for SuiteSparse 7.1.