add friendly error message for multiple return syntax

[HomerReid]

  julia> [V, D] = eig([1.0 2.0; 2.0 1.0]) 
  syntax error: invalid assignment lvalue (call (top vcat) V D)
  in run_repl, /home/homer/hacking/julia/j/client.j:23
  in _start, /home/homer/hacking/julia/j/client.j:153

Instead, detect what they're trying to do and show a nice friendly message.

[ViralBShah]

I think this should be two separate issues. Promotion of integer arrays to float is generally required for all linear algebra operations, and needs a general way to handle.

[StefanKarpinski]

Promotion of arguments to LAPACK functions should be pretty easy since they're all generated with macros anyway. See issue #134.

[ViralBShah]

But, we'd have to change each BLAS/LAPACK/ARPACK macro and such. Isn't there a simpler way?

-viral

On Jul 19, 2011, at 8:47 AM, StefanKarpinski wrote:

Promotion of arguments to LAPACK functions should be pretty easy since they're all generated with macros anyway. See issue #134.

Reply to this email directly or view it on GitHub:
#133 (comment)

[StefanKarpinski]

I'm not sure where you're seeing difficulty here. We need to add things like this:

eig{T<:Real}(x::Array{T,2}) = eig(float64(x))

Assuming that float64 is appropriately vectorized (which I just made a commit (abfc98a) that helps with). Yes, we need to add one for each LAPACK function, but that's hardly a big deal. For BLAS, it's not appropriate because we actually want to use slower but exact integer arithmetic for those operations. eig is going to produce floating point regardless, so converting first is the right way to go. ARPACK has two functions in it and we don't really support them very well yet, so we can just punt for now.

[ViralBShah]

Ok, I guess that's easy enough.