Column insertion in DataFrameRow

[bkamins]

Fixes #3390

@nalimilan - I just did the initial implementation. Let us first decide if we like the functionality and if yes, I will document and test it.

[nalimilan]

Makes sense. Have you thought about other functions that we could want to implement (or not) for DataFrameRow? For example, we could allow deletecols!, but it could also be a bit more dangerous...

[nalimilan]

Use Tables.allocate_column? similar is for arrays.

Or maybe just call insertcols! to avoid duplication? Is that slower?

[bkamins]

This is a very good point. I used similar, because we already had it in
https://github.com/JuliaData/DataFrames.jl/blob/main/src/subdataframe/subdataframe.jl#L207
which also requires fixing then.

Other places with the same issue are e.g.:

• DataFrames.jl/src/subdataframe/subdataframe.jl

  Line 239 in 8c32d53

  newcol = similar(old_col, promote_type(T, S), length(old_col))

• DataFrames.jl/src/other/broadcasting.jl

  Line 204 in 8c32d53

  col = similar(Vector{typeof(v)}, nrow(df))

• DataFrames.jl/src/dataframe/insertion.jl

  Line 313 in 8c32d53

  newcol = similar(df1_c, promote_type(S, T), targetrows)

Do you think we should do a systematic review of them and fix this everywhere?

[nalimilan]

I thought we followed the policy that similar is called when we have a vector to pass to it, and otherwise allocatecolumn is used. Apparently that's not really the case, given the example in broadcasting.jl. The two other examples look correct to me. We could try changing the broadcasting.jl one and check whether other places have the same issue.

[bkamins]

we could allow deletecols!

Now I just added functions that match what we support for SubDataFrame. We could support e.g. deletecols! both for SubDataFrame and DataFrameRow, but I would wait till someone asks for it. Column adding is useful as it fills omitted rows with missing and this is what this PR does.

[nalimilan]

Now I just added functions that match what we support for SubDataFrame. We could support e.g. deletecols! both for SubDataFrame and DataFrameRow, but I would wait till someone asks for it. Column adding is useful as it fills omitted rows with missing and this is what this PR does.

Right, I just wanted to think about the broader implications of this PR. Mirroring what we do for SubDataFrame sounds like a good rule.