with_columns_seq should assign by expression

[uditrana]

Description

Polars values itself in having a clean, readable interface. To that end, they should expose a with_columns style option/function that allows you to do sequential operations with assignment between operations. This has come up before #11570 .

Current:

df.with_columns(
    pl.col("a").add(1).alias("a_plus_1"),
).with_columns(
    pl.col("a_plus_1").add("a").alias("2a_plus_1"),
)

Ideally:

df.with_columns_seq_assign(
    pl.col("a").add(1).alias("a_plus_1"),
    pl.col("a_plus_1").add("a").alias("2a_plus_1"),
)

[McNickSisto]

I second this.

[cmdlineluser]

The actual previous discussion for the desired functionality is:

• add a sequential run order in with_columns execution #9062

(The _seq in this function is about parallelism.)

[uditrana]

Yea, seems like @ritchie46 was pretty firm with his process in that Issue. It is a bit odd to me because the Polars API seems to have introduced multiple variants (e.g. with_columns_seq) of methods... so maybe philosophy has changed on this matter.

[avimallu]

For reference, @uditrana, I don't think the philosophy has changed significantly. with_columns_seq was created (#10322) specifically to

serve[s] merely as an optimization on small data, when parallelism overhead is too costly.

[ritchie46]

This has been discussed and we will not change the semantics of a with_columns. They will get their state from their context, not from incremental updates. The seq refers to sequential and is recommended to not be used. It is a micro-optimization that sometimes is worth it on tiny data.

[uditrana]

Yea, I wasnt really suggesting changing with_columns core semantics, but rather an alternative option like the seq version for specific use cases. But not worth pushing further if you all don't see it as a useful stylistic addition.