pl.<ex-horizontal-operation>() gives unhelpful duplicate column error

[mcrumiller]

Issue Description

import polars as pl

pl.DataFrame({'a': [1], 'b': [1]}).select(
    pl.sum('a', 'b').alias('c')
)

exceptions.DuplicateError: column with name 'c' has more than one occurrences

pl.sum() in this context has been replaced with pl.sum_horizontal, but this is the wrong error.

Expected behavior

pl.sum() now takes one argument, an expression. If it gets more than one, it should give an error indicating that the API has changed. The current error is incorrect.

[reswqa]

Since pl.sum() is worked as pl.col("a", "b").sum(), if we pass more than one cols, it will expend into col("a").sum() and col("b").sum(). If we add the following alias to it, there will be a problem. That's to say, this may not a bug.

[mcrumiller]

I think the reason is that pl.sum() performs a sum on each column, so it's basically the equivalent of attempting to perform an alias on multiple columns, e.g.:

pl.DataFrame({'a': [1], 'b': [1]}).select(pl.all().alias('c'))

exceptions.DuplicateError: column with name 'c' has more than one occurrences

[mcrumiller]

I encountered this error in my codebase multiple times, all with pl.sum, pl.any, pl.all, etc. These have all since been replaced by pl.*_horizontal and I think we should give a more informative error when it's applied to multiple columns.

[reswqa]

These have all since been replaced by pl.*_horizontal

I'm a bit confuse about this, shouldn't the behavior of pl.sum ("a", "b") and pl.sum_horizontal ("a", "b") be different(one map to shape(1,2) but another map to shape(1, 1))? What does' replace 'mean here 🤔

[reswqa]

One possible optimization is that when we perform alias on multiple column expressions(e.g. pl.all(), cs.numeric()), we should give a better error information.

[cmdlineluser]

Another one I noticed is this could not convert value 'Unknown' as a Literal message which seems a bit cryptic:

df = pl.DataFrame({"A": [1, 2, 3], "B": [4, 5, 6]})

df.select(pl.all(pl.all() > 5))

# ValueError: could not convert value 'Unknown' as a Literal

[stinodego]

One possible optimization is that when we perform alias on multiple column expressions(e.g. pl.all(), cs.numeric()), we should give a better error information.

This would be the desired solution here. Not sure if it's easy to implement though.