feat(polars): add Intersection and Difference ops

[IndexSeek]

Description of changes

Adds support for Intersection and Difference operations for the Polars backend. The approach uses pl.SQLContext to register the dataframes and query using INTERSECT and EXCEPT, respectively.

I saw that a few other functions used a ctx argument which might be using pl.SQLContext in a similar manner, I wasn't sure if that argument could be used here to avoid some of the boilerplate. Here are some of those functions I am referring to:

ibis/ibis/backends/polars/compiler.py

Lines 69 to 72 in 28bafd1

	@translate.register(ops.InMemoryTable)
	def in_memory_table(op, *, ctx, **_):
	sql = sg.select(STAR).from_(sg.to_identifier(op.name, quoted=True)).sql(Polars)
	return ctx.execute(sql, eager=False)

ibis/ibis/backends/polars/compiler.py

Lines 1287 to 1297 in 28bafd1

	@translate.register(ops.SQLStringView)
	def execute_sql_string_view(op, *, ctx: pl.SQLContext, **kw):
	translate(op.child, ctx=ctx, **kw)
	return ctx.execute(op.query)
	@translate.register(ops.View)
	def execute_view(op, *, ctx: pl.SQLContext, **kw):
	child = translate(op.child, ctx=ctx, **kw)
	ctx.register(op.name, child)
	return child

[cpcloud]

@IndexSeek Thanks for the PR as always 😄

I think it's reasonable to reuse the ctx argument in your implementations here rather than constructing a new SQLContext.

[IndexSeek]

@IndexSeek Thanks for the PR as always 😄

I think it's reasonable to reuse the ctx argument in your implementations here rather than constructing a new SQLContext.

You're welcome! That seems like a good plan. I gave this a try using the ctx from kwargs. I was experimenting with a way to reuse the existing tables as part of the context, but for some reason when I was doing op.left.name it would work standalone but when running tests I would get things like:

AttributeError: 'Sort' object has no attribute 'name'

This was working for me when using a memtable in an IPython shell, but not on the tests.

Do you know if there is a more efficient way than needing to manually register the tables as "left" and "right"?

Here's what I was trying where it seems to work using op.left.name, etc.:

@translate.register(ops.Difference)
def execute_difference(op, **kw):
    ctx = kw.get("ctx")
    sql = (
        sg.select(STAR)
        .from_(sg.to_identifier(op.left.name, quoted=True))
        .except_(sg.select(STAR).from_(sg.to_identifier(op.right.name, quoted=True)))
    )
    result = ctx.execute(sql.sql(Polars), eager=False)
    if op.distinct is True:
        return result.unique()
    return result

In [1]: from ibis.interactive import *
   ...: 
   ...: ibis.set_backend("polars")
   ...: 
   ...: t1 = ibis.memtable({"number_col": [1, 2, 3]})
   ...: t2 = ibis.memtable({"number_col": [2, 3, 4]})
   ...: t1.difference(t2)
Out[1]: 
┏━━━━━━━━━━━━┓
┃ number_col ┃
┡━━━━━━━━━━━━┩
│ int64      │
├────────────┤
│          1 │
└────────────┘

[IndexSeek]

Nice, thank you for pushing those changes! The gen_name was a great idea and I see how you added ctx and used it with translate.

[cpcloud]

@IndexSeek Thank you!