Add row-wise macros

[pdeffebach]

No description provided.

[pdeffebach]

Just finishing adding the macros, will add tests tomorrow (should be easy, just the same as the @byrow tests).

[pdeffebach]

Okay I've updated tests and docstrings and a little bit of index.md.

One question for @bkamins and @nalimilan: Should we deprecate

@transform df @byrow 
    ...
end

Given that this can more easily be written as @rtransform? If so, I will add a deprecation message and update the docs, after that this will be ready for merging.

[bkamins]

Should we deprecate

@transform df @byrow 
    ...
end

The question is if it hurts to have it. I understand that we keep @byrow for cases when you call @transform with some operations that are intended to be done row-wise and some that are intended to be done columnwise. Is this a correct understanding? (as if we keep @byrow anyway then I think we can leave it)

[pdeffebach]

Sorry, I should have been clearer. I mean

@transform df @byrow begin 
    :y = f(:x)
    :z = g(:x)
end

That is, when @byrow is an "outer" flag, all operations are row-wise. This is exactly equivalent to @rtransform begin ... end

[bkamins]

Yes - I have understood it. And I meant that as long as we accept @byrow also as an inner flag (per transformation) I am OK to keep it. So - as you prefer. I would make a decision based on the fact if it would simplify the code if you removed it.

[nalimilan]

Would there be a way to avoid duplicating these helper functions, for example by passing subset or subset! to a more generic one?

[pdeffebach]

Yes, that't what DataFrameMacros does. I will do this in a future PR.

[nalimilan]

Why not do it now? :-p

[pdeffebach]

Okay I've simplified the @byrow check but it's hard to actually make everything in a coherent function

1. The ugly t = (fun_to_vec(ex; no_dest=true, outer_flags=outer_flags) for ex in exprs) can't be gotten rid of yet because we still need to support some deprecated functionality in @combine.
2. There are lots of exceptions, for example @subset requires a keyword argument, @orderby doesn't even call a DataFrames function. Writing a function for these cases would be just as easy as writing out the expression.

[pdeffebach]

Okay I think this is also ready for merging!

@nalimilan @bkamins as Milan noted there is some technical debt here but it can be re-worked in a future PR.

[pdeffebach]

Yes, that't what DataFrameMacros does. I will do this in a future PR.

[bkamins]

As commented in #270 maybe we need @filter as @rsubset is not going to allow for filtering of groups?

[pdeffebach]

Maybe I should restrict the row-wise macros to only work on AbstractDataFrames in order to avoid confusion.

[bkamins]

I think it is ok for row-wise macros to work also on GroupedDataFrame. However, given the questions today during the workshop - it is crucial to explain the row-wise vs whole-column behavior. It seems it is a super confusing distinction for newcomers.

[bkamins]

If I do not do a final review of this PR in 2 days from now please ping me (I cannot do it now).

[pdeffebach]

Yes I'm sorry I wasn't there to help field questions. I should have offered.

[bkamins]

Thank you for posting this. Actually I have not expected so many questions (but probably it is a good sign that there were so many questions). I was swamped during the first hour, but then I learned how to sort them.

[pdeffebach]

Thank you for the review!

I have responded to comments.

[pdeffebach]

Sorry about the typos. This is ready to be merged, then.

[pdeffebach]

Okay the merge was actually very easy and there were no conflicts.

I will merge this!