Use DataAPI.rownumber

[VEZY]

Now using the newly proposed DataAPI.rownumber interface (JuliaData/DataAPI.jl#60). We add an implementation for: Tables.ColumnsRow, Tables.DictRow, Tables.IteratorRow and Tables.MatrixRow.

Note: a similar function (getrow) existed already for ColumnsRow and IteratorRow. These functions were not removed as a precaution, but rather used for implementing DataAPI.rownumber (i.e. rownumber(c::ColumnsRow) = getrow(c))

The implementation for Tables.Row is just rownumber(x::Row) = rownumber(x.x). Maybe we should add a fallback or throw an explicit error if the row type does not implement rownumber ?

[bkamins]

Also DataAPI.jl version requirement needs to be changed below.

[VEZY]

Ho yes I forgot this you're right. Done in aefedf6

[bkamins]

Can you please add tests for all 5 types?

[VEZY]

Hoo, when making the tests I did realise I was completely misinterpreting the row argument, it was not an index, but the actual data of the row... That's why tests are so valuable ^^.

So I don't think we can get the row number without adding it as a field of the rows structures for: Tables.ColumnsRow, Tables.DictRow, Tables.IteratorRow and Tables.MatrixRow.

There's two choices from here, either we add a field for the row index into these structures, or we don't define DataAPI.rownumber for these.

I made an example implementation of a row number in DictRow (5f091b0). If you think that's a good path forward, I'll implement it the same way for the others.

[bkamins]

There's two choices from here, either we add a field for the row index into these structures, or we don't define DataAPI.rownumber for these.

Let us wait for @quinnj to decide on this. My thinking is that:

• it should not add performance overhead
• having this information can be useful for users

so I would be OK to add it.

The only downside could be that in some cases these objects could be generated from sources that do not have a proper row numbering. However, I am not sure if this is possible. @quinnj should know.