[kshyatt]

I was really torn about what to call this check. check_hermiticity seems accurate but is ugly, but check_hermitian is already taken by the function itself.

The idea here is to allow users to avoid the check_hermitian calls if they are sure the input matrix is hermitian, ie if they just called project_hermitian! on it. This comes up for example in the C4v CTMRG case where we compute eigh_vals on something that was just projected.

[codecov]

Codecov Report

❌ Patch coverage is 84.05797% with 11 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...ixAlgebraKitEnzymeExt/MatrixAlgebraKitEnzymeExt.jl	84.44%	7 Missing ⚠️
src/implementations/eigh.jl	83.33%	4 Missing ⚠️

Files with missing lines	Coverage Δ
src/implementations/eigh.jl	95.04% <83.33%> (ø)
...ixAlgebraKitEnzymeExt/MatrixAlgebraKitEnzymeExt.jl	78.96% <84.44%> (+0.82%)	⬆️

🚀 New features to boost your workflow:

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

[lkdvos]

Overall definitely fine by me, although maybe we can discuss the API of this a little bit:

In general, my idea was that the algorithm struct holds all of the settings, so no additional keywords are needed for the f!(in, out, alg) methods. This mostly is beneficial to reason about signatures etc, for example I think the non-inplace methods wouldn't properly pass on the check_hermiticity keyword and instead would pass that to the select_algorithm function. I think most of our algorithm structs are generic so you can do something like get(alg.kwargs, :check_hermiticity, true) to get it out generically?

For the name, do you think it might make sense to just have a package-wide unsafe keyword that basically disables all checks, or something similar?

[kshyatt]

Extremely open to API changes. I didn't go with generic unsafe since only eigh has a similar check (e.g. I think no other function checks for posdef or something). But putting it in the alg can also be done and we just ignore that for eig and co. I'll try that and we can see if we like how it looks, it will simplify things for Enzyme too.