add allowscalar

[SobhanMP]

closes #189

the only downside is that revise might act a bit weird.

also maybe RCI should call @eval on exp direclty instead of returning exp?

another detail is that should this also block stuff like

A = sprandn(100, 100, 0.1)
A[i, :]
A[1:10, :]

or just the ones i marked?

This does NOT cause any degradation in any way, unlike adding a state variable.

[codecov-commenter]

Codecov Report

Merging #200 (782144e) into main (134f67a) will increase coverage by 0.27%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #200      +/-   ##
==========================================
+ Coverage   91.85%   92.13%   +0.27%     
==========================================
  Files          11       11              
  Lines        7118     7167      +49     
==========================================
+ Hits         6538     6603      +65     
+ Misses        580      564      -16     

Impacted Files	Coverage Δ
src/abstractsparse.jl	44.44% <100.00%> (+13.40%)	⬆️
src/sparsematrix.jl	95.33% <100.00%> (+0.04%)	⬆️
src/sparsevector.jl	95.60% <100.00%> (+1.42%)	⬆️
src/solvers/spqr.jl	88.31% <0.00%> (-1.03%)	⬇️
src/solvers/cholmod.jl	90.66% <0.00%> (-0.02%)	⬇️
src/linalg.jl	84.50% <0.00%> (ø)
src/SparseArrays.jl	100.00% <0.00%> (ø)
src/higherorderfns.jl	96.66% <0.00%> (+0.11%)	⬆️
src/solvers/umfpack.jl	88.30% <0.00%> (+0.14%)	⬆️
... and 1 more

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[rayegun]

You mean it causes a 2% slow down?

[SobhanMP]

no it does not, Chris said that adding a flag for scalar (with a global const scalar = Ref{Bool}(true)) would cause a 2% slowdown. this does not do that. This should cause NO runtime overhead. That's the whole goal of having a macro that restores the function (and delete it), every thing happens at compile time.

[rayegun]

Are we doing both the static type and the macro?

[SobhanMP]

what static type? FixedSCS? yeah.

[SobhanMP]

bump

[ViralBShah]

Was waiting on the update in base to go through.

I think the allowscalar API should be documented and marked experimental (and not exported). Implementation looks good.