Attempting to run sparse `qr` produces StackOverflow when run on a sparse matrix of `ForwardDiff.Dual`. about sparsearrays.jl

[rayegun]

Yes

[SobhanMP]

What would be a good solution? I think it's easy to check if typeof(float(A)) === typeof(A), but what about types where float(A) has a cyclic behavior? For instance, imagine 3 matrices A, B, and C where float(A) == B, float(B) == C, and float(C) == A. Can such matrices exist in Julia? (Assuming a pure function float)

[lukas-weber]

I think cycles can exist per-se.

How about checking isprimitivetype(real(float(A)))? This works only if the type directly decays to one of those, but for types that need multiple iterations of float to decay (if they exist), it could be the responsibility of the user to do the conversion.

[SobhanMP]

@rayegun the QR factorization only works on Float64, right? Can we directly cast to Float64?

[rayegun]

And ComplexF64

[SobhanMP]

But can we just cast it directly to F64/CF64? (i.e. replace float(A) with Float64.(A))?