Type instabilities

[antoine-levitt]

Work towards #60. Should take care of most of the implicit Float64 conversions.

(based on #59)

[antoine-levitt]

It's actually quite pleasing how few of these we need. The bzmesh is in rational and the G are ints, so we have things like recip_lattice * (G+k) do the right thing all by itself

[mfherbst]

I agree ... if what you committed so far is all that's needed ... that would be pretty good!

[antoine-levitt]

So lda_c_vwn.jl would be pretty annoying to convert to generic. I think the only automatic option is to use ChangePrecision inside a generated function.

[antoine-levitt]

I think apart from xc_fallback/ and occupation.jl we're fine.

[antoine-levitt]

Something like this appears to work:

using ChangePrecision
@generated function f(x::T) where T
    @changeprecision T begin
        :(2/3*x)
    end
end