Minor things

[jagot]

• Added back cache kwarg for now, as requested in Apply changes to ExponentialUtilities.jl to the ExpRK integrators OrdinaryDiffEq.jl#517.
• Remove a few unused arguments to the *_step! methods
• Fixed indentation of tests

[codecov]

Codecov Report

Merging #7 into master will not change coverage.
The diff coverage is 100%.

@@           Coverage Diff           @@
##           master       #7   +/-   ##
=======================================
  Coverage   96.86%   96.86%           
=======================================
  Files           6        6           
  Lines         223      223           
=======================================
  Hits          216      216           
  Misses          7        7

Impacted Files	Coverage Δ
src/krylov_expv.jl	90% <100%> (ø)	⬆️
src/arnoldi.jl	95.55% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 16ed363...dfdfb75. Read the comment docs.

[ChrisRackauckas]

When this tags, OrdinaryDiffEq should get a minor update which bumps the minimum version to the new tag, then the cache fallback can be dropped, and then OrdinaryDiffEq.jl can safely drop it. That is a zero-hiccup deprecation that's nice and clean. Sorry it has an extra step but keeping tests green and not force merging to METADATA is just nicer overall. (Technically we should have a maximum version bound in there, but we can ignore that because it gives the resolver issues)

[MSeeker1340]

I'll tag this now and after this gets registered, I'll push the corresponding OrdinaryDiffEq.jl version bump.

@jagot Are you planning on more changes to ExponentialUtilities.jl? I see that there're still things on the plate in #1.

[jagot]

Short-term, no. Long-term, probably.

• In the future, I'll want to work with multi-electron problems, which will require partitioning of the state vector, in which case the type agnosticism à la KrylovKit.jl will come in handy.
• I'll also be exploring various absorbing boundaries, that will yield complex-symmetric operators, rather than Hermitian ones, at which point I'll need the Arnoldi to work with the alternative interface. However, whilst looking into that, I did not get hseqr2 (QR/Schur factorization of upper Hessenberg matrices) working properly. On top of that, implementing an optimized exponentiator for triangular matrices, as described by Davies and Higham (2003), is slightly daunting at the moment. Such a routine should go into Base.LinearAlgebra, I guess. One "easy" way out is to butcher Base.LinearAlgebra.exp! and turn it into a cache-friendly version. This would not, however, utilize the upper-Hessenbergness (?) of the subspace matrix.