Incorrect filtering of subarrays

[magistere]

filter produces incorrect output for subarrays:

julia> a = [1:10]
10-element Array{Int32,1}:
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10

julia> b = sub(a, 5:8)
4-element SubArray{Int32,1,Array{Int32,1},(Range1{Int32},)}:
 5
 6
 7
 8

julia> filter(x -> x < 7, b)
4-element Array{Int32,1}:
 5
 5
 4
 4

julia> filter(x -> x < 7, b[:])
2-element Array{Int32,1}:
 5
 6

[timholy]

The issue seems to be that we're missing an implementation of logical indexing for SubArrays. It's being interpreted as b([1,1,0,0]), and because of the lack of bounds-checking on SubArray indexing no error is thrown.

[ivarne]

@timholy beat me to it.

The problem is that filter creates a boolean array that is used for indexing. Subarray does not support a boolean array for indexing.

The method that gets invoked for boolean array indexing is

function getindex{T,S<:Integer}(s::SubArray{T,1}, I::AbstractVector{S})

because Bool is a subtype of integer. I think this will be fixed if we define

function getindex{T,S<:Integer}(s::SubArray{T,1}, I::AbstractVector{Bool})

[timholy]

Agreed. I'll let you beat me this time, if you're game :-).

[ivarne]

I'll give it a try, but probably it will not be good enough for Stefan.

[ivarne]

And I will not have time to read subarray.jl and understand how arrays and dimensions work until Monday. An implementation just raising an error might be good before we release 0.2.

[StefanKarpinski]

I'll give it a try, but probably it will not be good enough for Stefan.

I'm not sure where I got this stern reputation from. Is it the viking horns?

[magistere]

Yeah, horns are awesome :)

[magistere]

Tim and Ivar, thank you for pointing out the root of the problem.

[ivarne]

@StefanKarpinski I mentioned you because I often see you saying NO to slow implementations and new language features with too specific use cases. Too me it looks like you're doing a terrific job to make the language pleasant, consistent and easy to learn.

The subarray code is also performance critical, so it is highly likely that there is a better way to implement it than what I would end up with.

[StefanKarpinski]

Ah, yes. Saying "no" is definitely a huge part of my job :-)

The current SubArray code is not exactly immaculate, so I'm a little less protective of it. The whole thing is badly in need of an overhaul.