Split off list constructor logic from `pl.concat_list` into `pl.list`

[cmdlineluser]

Problem description

Discussing #8503 sparked the thought - would it make sense to "rename" pl.concat_list to pl.list?

pl.list is currently marked for deprecation as part of the .arr to .list rename.

DeprecationWarning: `pl.list` is deprecated, 
please use `pl.implode` instead.

Similar to how there is pl.struct and the .struct namespace:

pl.struct(...).struct

If pl.list were repurposed it would be:

pl.list(...).list

[stinodego]

Fully agree. Will add this to the milestones for the next breaking release.

[cmdlineluser]

Hi @stinodego - I just read your comment regarding horizontal sum which made me think of this issue.

After the concat_list/namespace rename

pl.concat_list(pl.col("nums") * 2).arr.sum()

Would become

pl.list(pl.col("nums") * 2).list.sum()

Your comment made me wonder if it would be possible to provide a syntax sugar version:

pl.list.sum(pl.col("nums") * 2)

sugar:

pl.sum("nums")      # vertical
pl.list.sum("nums") # horizontal

no-sugar:

pl.col("nums").sum()
pl.list("nums").list.sum()

[lucazanna]

Quick thoughts on this:

i like the idea of aligning the methods name between structs and lists.
however a pl.list(…).list does not seem very readable to me / clear that it works on rows instead of columns (at least at first)

What about keeping concat_list so we have
concat_list(…).list

And then rename struct to concat_struct so that we have
concat_struct(…).struct

i believe that concat word makes it clearer that we are concatenating multiple rows together

thoughts ?

[stinodego]

I am also a bit on the fence about this. On the Rust side, struct is called as_struct. Maybe that's better; not sure. I'll brood on this for a bit.

[ritchie46]

Yes, I agree that we miss an (I don't know the english name of this, adjective? verb?), but something that says "what" we are going to do with the noun "list".

We already have this on many places. read_csv, to_uppercase. So I think we should do one of those:

• as_list, as_struct
• to_list, to_struct,
• pack_list, pack_struct

But the noun list, struct is not clear as we do not state what we do with those.

[stinodego]

We have a set of functions that take one or more expressions and combines them to generate a new column of a certain datatype. We should rename these to reflect that they are essentially doing the same thing. I have given it a lot of thought and I think the as_ prefix is the most expressive. It would result in the following names:

Before	After
datetime	as_datetime
date	as_date
time	as_time
duration	as_duration
struct	as_struct
concat_list	as_list
concat_str	as_string
format*	format

* Convenience wrapper around concat_str, remains unchanged

I plan to do the following:

1. Refactor to put all these functions into a single module, as well as gather any dedicated tests (DONE)
2. Change behaviour to have the expression named after the datatype
3. Add new function names, deprecate old.

[lucazanna]

hi @stinodego,

Would it make sense to add an ‘as_array’ method?

Same logic as ´as_list’ but with the result being an array

[stinodego]

Yes, it's already up on my whiteboard 😄