pointerset causes compilation failures (even in non-reachable code branches)

[carlobaldassi]

Example:

function h{T}(x::Array{T})
  if isbits(T)
    unsafe_store!(pointer(x), 1)
  else         
    x[1] = 1
  end
  return x
end

the fact that unsafe_store! is not defined for non-bits types should not prevent compilation, but:

julia> h([3,3,3])
3-element Int64 Array:
 1
 3
 3

julia> h({3,3,3})
ERROR: error compiling h: pointerset: invalid pointer type

On a related note, I've seen that (at least in some cases) even though the information required to choose a branch is available at compile time (e.g. the isbits(T) should be trivially known from T), the code is not optimized away. I wonder if there's anything which could be done about this.

[JeffBezanson]

Yes, more can definitely be done with constant branches.