Need better error for Sparse \ for non-square matrices

[timholy]

There's an error in the show method due to tty_cols no longer being exported, but what I'm really worried about is the last line:

julia> A = float([1 0;1 0; 0 1])
3x2 Array{Float64,2}:
 1.0  0.0
 1.0  0.0
 0.0  1.0

julia> b = float([1:3])
3-element Array{Float64,1}:
 1.0
 2.0
 3.0

julia> A\b
2-element Array{Float64,1}:
 1.5
 3.0

julia> S = sparse(A)
3x2 sparse matrix with 3 Float64 nonzeros:
Error showing value of type SparseMatrixCSC{Float64,Int64}:
ERROR: tty_rows not defined
 in show at sparse/sparsematrix.jl:22
 in anonymous at no file:921
 in with_output_limit at show.jl:898
 in showlimited at show.jl:920
 in writemime at repl.jl:2
 in display at multimedia.jl:109
 in display at multimedia.jl:111
 in display at multimedia.jl:143

julia> S\b
ERROR: Error code -13 in umfpack_solve
 in error at error.jl:21
 in solve at linalg/umfpack.jl:166
 in \ at linalg/umfpack.jl:253

Yikes. Hard to imagine this went undetected for so long.

[ViralBShah: Changed title]

[stevengj]

SuiteSparse provides a sparse-QR solver that could be used for this (http://www.cise.ufl.edu/research/sparse/SPQR/), and which is apparently what Matlab uses.

[mlubin]

SPQR was discussed in #3554. It's still up for grabs, though.

[timholy]

Yes, at the time I filed this I thought it was a simple bug, but then I too discovered that it goes deeper.

In the short term, I wonder if we should at least check for squareness and issue a more informative error. There's also svds which could be roped in to do this, although it's not as good an approach as SPQR.