convert(::Type{Ptr{T1}}, x::Array{T2}) should fail if T1 != T2

[stevengj]

Right now, Julia will happily do:

julia> convert(Ptr{Int}, [1.0])
Ptr{Int64} @0x000000010367e470

This is an invitation for bugs; I just ran into this today with a student who was doing ccall(:foo, Void, (Ptr{Cint},), x) and didn't understand why it was failing for x = [1,2,3] or x = [4.0, 5.0].

My preference would be to throw an error in this case. We can still allow convert(::Type{Ptr{T1}}, x::Ptr{T2}) to work, so if you want to do an explicit cast from one pointer type to another you can still do so.

[ivarne]

+1

We have reinterpret when you want to do this. Should we keep convert(Ptr{None}, [1,2]) as a special case?

[JeffBezanson]

Yes I think this is reasonable along with a special case for Ptr{None}.

[stevengj]

Whoops, this breaks a bunch of the tests (sorry, I though I had run make testall before pushing); need to add some more fixes first.

[stevengj]

And if it breaks several places in Base (exposing some bugs, actually), it will probably break packages as well, so we'll need to deprecate the old behavior.

[meggart]

Thanks for this change, it made me discover a bug in the NetCDF package, too. Is there a way to get a more meaningful (error?)-message than the deprecation warning? Or will the warning persist for julia 0.3 and an error be generated for julia 0.4?