Simplify "value type (Val)" usage

[expnn]

In 0.7, Val(N) is the instance of Val{N}, so it does not require Type{Val{N}} any more.

julia> isa(Val(4), Val{4})
true

Even though, this is not true for 0.6 and early versions of julia, all tests are passed till now.

[c42f]

Looks like a good simplification, thanks. Compat defines the Val(N) constructor, so you can rely on this being available.

However, this is a rather large breaking change to the public API - I think it deserves a round of deprecations at least, and possibly just continued support for both versions until 0.7 is out. (In fact, I wonder whether constant propagation will be powerful enough in 0.7 that we can remove some of these.)

There's two general cases:

1. Functions where we needed to break compatibility with Base to get the output type inferred (eg, diagm)
2. Functions where we use exactly the same input types as Base (eg, the pivot argument to lu).

For the first case, it seems we could do a round of deprecations, and just go with Val(N) immediately. For the second case, I'd be uncomfortable about breaking consistency with 0.6 until 0.7 is out, and I think we should continue to support both versions until 0.7 has been released.

So, from my point of view, here's a plan to get this merged: (1) Make a depwarn version of all functions in class (1). For all functions in class (2) create a version which takes the old arguments and forwards them to your new version.

BTW, the CI tests are currently failing in LU, so I think there's some more work to do there.

Thanks for contributing!

[c42f]

Actually on further reflection, perhaps we should postpone merging this until it's clearer what the situation regarding constant propagation will actually be in 0.7? It could be that we can just remove most of our uses of Val entirely which would allow us to be exactly compatible with Base. See for example JuliaLang/julia#24011 and linked issues.

It would be unfortunate to force users of StaticArrays through an extra unnecessary deprecation cycle.

[c42f]

Seems this was sorted out in #473 and elsewhere. Sorry we dropped the ball on this PR :-/