test 1x1 cases

[stevengj]

I notice that several of your tests are currently only testing 2x2 and larger matrices. They should be fixed to ensure that 1x1 works too.

(The trivial 1x1 case does not need to be optimized, but it should still work. I'm not sure if you were confused by this in JuliaLang/julia#84 — I suggested removing the n==1 special case from the code to simplify the code paths and to remove an apparently superfluous optimization … the generic implementation should ideally be written in such a way as to work for n==1 without modifiction.)

[smataigne]

I added the case n==1 to all tests, except for those for tridiag.jl . For this one it is a bit more touchy because the sub-diagonal must be an empty complex array and I am not sure yet that everything works for that case.
For example, dot(x, A, y) doesn't work. However it doesn't work either for A SymTridiagonal because the implementation suppose the access to the 2 first elements of x.

[smataigne]

I notice that several of your tests are currently only testing 2x2 and larger matrices. They should be fixed to ensure that 1x1 works too.

(The trivial 1x1 case does not need to be optimized, but it should still work. I'm not sure if you were confused by this in JuliaLang/julia#84 — I suggested removing the n==1 special case from the code to simplify the code paths and to remove an apparently superfluous optimization … the generic implementation should ideally be written in such a way as to work for n==1 without modifiction.)

I understand very well the philosophy.

[stevengj]

However it doesn't work either for A SymTridiagonal because the implementation suppose the access to the 2 first elements of x.

That's a bug, I've filed an issue JuliaLang/LinearAlgebra.jl#945.

In the meantime you could always disable that particular check for n == 1, or write a special check.

[smataigne]

I fixed dot and mul! for the n==1 case

[stevengj]

Just your friendly periodic reminder to mention the issue (#89) in any PR and/or commit message that addresses it, so that it will be linked here.