padNA() broken about dataarrays.jl HOT 11 CLOSED

Also, sometimes where we'd expect an NA we get 0.

julia> padNA(da, 4, 2)
9-element DataArray{Int64,1}:
 140200970311424
 140200928071456
 140200970385312
 140200928071744
               1
               2
               3
               0
               0

EDIT:

Or some other integer, including 1 or 8 ...

julia> padNA(da, 4, 4)
11-element DataArray{Int64,1}:
 140200948841536
 140200948841520
               8
      4294967296
               1
               2
               3
     12884901888
     12884901888
               0
               0

from dataarrays.jl.

This is almost certainly dirty memory.

from dataarrays.jl.

I'm figuring it's part of Chaos month, but can't track it down. Updating to 0.3 now to see if that changes anything

from dataarrays.jl.

Well 0.3 didn't change anything as expected. I couldn't find the source using padNA in repl until I manually went into .julia/DataArrays and did a git checkout master for some reason Pkg.update() wasn't working. In any case, I'll debug the method and see if I can fix it. It's in data vector.jl line 123 of commit 53581c3

from dataarrays.jl.

Well the only change to the method from this resolution JuliaData/DataFrames.jl#154 is the method signature changed Int to Integer for the front and back parameters, but changing this doesn't solve the problem.

The similar method is where the trouble is generated.

from dataarrays.jl.

Somewhere along the line, function similar became function Base.similar and the line (actually the entire code block)

 DataArray(Array(T, dims), trues(dims))

was replaced by

DataArray(Array(T, dims), BitArray(dims))

Using the original code block fixes this issue.

1. Why did BitArray replace trues?
2. Can we put the trues(dims) back in?
3. What happened to using (dims...) ? (both (dims) and (dims...) was previously defined)

from dataarrays.jl.

Re similar, see #10. I think the consensus there was that we should change this back.

from dataarrays.jl.

So the idea is to not call similar at all in padNA but to write the method in some other fashion?

from dataarrays.jl.

I suppose you can pad the back by using [push!(dv, NA) for i = 1:back] (which works). Not sure how to do the front though ...

from dataarrays.jl.

[unshift!(dv, NA) for i = 1:front]

from dataarrays.jl.

Fixed by 96a6241

from dataarrays.jl.