make rowiterator an abstractvector

[piever]

This I think simplifies the code and adds feature as we can use the AbstractVector machinery to iterate and get an element by index / slice the table etc...

[nalimilan]

Makes sense, but you need to define Base.IndexStyle to return IndexLinear too I think.

[quinnj]

Can we do some performance tests to just make sure this doesn't affect anything there?

[piever]

Sure, I'll try to find some time today: I should probably specify that this abstractvector prefers linear indexing (as suggested above) and use @boundscheck and @inbounds to make sure we only do inbounds check when necessary (i.e, not when iterating the vector).

EDIT: actually the @inbounds change is maybe best left to a later PR as it may be not 100% trivial

Do you have some standard benchmarks for iteration?

[quinnj]

I don't have anything standard, I usually use these lines from CSV to sanity check things (https://github.com/JuliaData/CSV.jl/blob/master/test/files.jl#L56); the pandas csv file is big enough to hit the row-iteration and the buildcolumns call hits the no-schema path.