Correctness issues and errors after sorting

[tecosaur]

Hello,

I was recently surprised to find that when sorting a column I read from a CSV ...the numbers changed?

Using Julia 1.11 and the latest releases of SentinelArrays, CSV, and DataFrames, I executed:

CSV.read("my.csv", DataFrame).column |> unique! |> sort

This produced some really surprising values, and turned up some incredibly buggy behavior:

julia> cv = ChainedVector([rand(1:100, 20) for _ in 1:10])
[...]

julia> unique!(cv)
[...]

julia> sum(cv)
5073

julia> sum(sort(cv))
2236548265497175

If I keep on running sum(sort(cv)) the result bounces around a bit, it seems like there's something stochastic happening?

julia> sum(sort(cv))
4399

julia> sum(sort(cv))
4378

julia> sum(sort(cv))
4400

This can also produce an error instead of an incorrect result sometimes.

julia> sum(cv)
4414

julia> sum(sort(cv))
ERROR: ArgumentError: out of range arguments to copyto! on ChainedVector
Stacktrace:
 [1] copyto!(dest::ChainedVector{…}, doffs::Int64, src::ChainedVector{…}, soffs::Int64, n::Int64)
   @ SentinelArrays ~/.julia/packages/SentinelArrays/ob2QK/src/chainedvector.jl:475
 [2] copyto!
   @ ~/.julia/packages/SentinelArrays/ob2QK/src/chainedvector.jl:465 [inlined]
 [3] copymutable
   @ ./abstractarray.jl:1192 [inlined]
 [4] sort(v::ChainedVector{Int64, Vector{Int64}})
   @ Base.Sort ./sort.jl:1720
 [5] top-level scope
   @ REPL[39]:1
Some type information was truncated. Use `show(err)` to see complete types.

[laborg]

resize! is not working for a ChainedVector.

julia> resize!(ChainedVector([[1],[2],[3]]),2)
2-element ChainedVector{Int64, Vector{Int64}}:
   1
 #undef

[laborg]

This line isn't correct. The last chunk of a chained vector shouldn't be resized to zero (probably a one-off error).
https://github.com/JuliaData/SentinelArrays.jl/blob/a2976cb2c63ecb8b2f1c2dff50975835abfd4b0c/src/chainedvector.jl#L549C5-L549C54