Problem with SentinelArrays.ChainedVector when limit/skipto is set

[bkamins]

@quinnj:
Start Julia with multiple threads.

Run:

using DataFrames
using CSV
df = DataFrame(x = repeat(["a", "b"], 10^6), y = rand(2*10^6));
f1 = tempname();
CSV.write(f1, df);
df2 = CSV.read(f1, DataFrame, header = false, skipto = 1001, limit = 10000);
sort(df2, 1)

The problem is that in both columns SentinelArrays.ChainedVector the length of arrays does not match inds. Here is an example for Column2:

julia> df2.Column2.inds
4-element Vector{Int64}:
  2853
  5706
  8559
 10000

julia> cumsum(length.(df2.Column2.arrays))
4-element Vector{Int64}:
 2853
 5706
 8559
 9970

and you see that there are 30 elements missing.

If you reimplemented the getindex like:

function Base.getindex(A::ChainedVector, i::Integer)
    chunk, ix = index(A, i)
    x = A.arrays[chunk][ix]
    return x
end

to enable bounds checking you get bounds error when trying to work with df2.

[quinnj]

Thanks for the report; fix is up: #964.