Promotion in cov

[catawbasam]

See discussion at: http://stackoverflow.com/questions/25052037/cova-mean-0-gives-inexacterror

I did not find any reference to it when searching the issues. Apologies in advance if the issue has been filed already.

I just got a similar error from generic_scale!, reached via in cor at statistics.jl:439.
The versioninfo for my machine is:

Julia Version 0.3.0
Commit 7681878 (2014-08-20 20:43 UTC)
Platform Info:
  System: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i7-2720QM CPU @ 2.20GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

[andreasnoack]

@catawbasam The problem here is not generic_scale!, since it is not possible to scale integers in place. I have therefore changed the subject line.

I think that the problem is that cov doesn't promote to floating points. Hence the error similar to that from stackoverflow

julia> cov(rand(0:9,4,3),mean = 0)
ERROR: InexactError()
 in generic_scale! at linalg/generic.jl:18
 in covzm at statistics.jl:253
 in cov at statistics.jl:286

At the same time it might be possible to improve on type inference here

julia> Base.return_types(cov, (Matrix{Float64},))
1-element Array{Any,1}:
 Any

julia> Base.return_types(cov, (Matrix{Int},))
1-element Array{Any,1}:
 Any

cc: @lindahua

[kshyatt]

Still a problem?

[andreasnoack]

Unfortunately yes

julia> Base.covm(A, 0)
ERROR: InexactError()
 in macro expansion at ./linalg/generic.jl:9 [inlined]
 in macro expansion at ./simdloop.jl:73 [inlined]
 in generic_scale!(::Array{Int64,2}, ::Float64) at ./linalg/generic.jl:8
 in covm(::Array{Int64,2}, ::Int64) at ./statistics.jl:308