feat(python): BytecodeParser can now handle mixed/nested and/or control flow

[alexander-beedie]

Ref: #9968.

This was a fun one... enables two additional ops (POP_JUMP_FORWARD_IF_FALSE and POP_JUMP_FORWARD_IF_TRUE if Python >= 3.11, and POP_JUMP_IF_FALSE and POP_JUMP_IF_TRUE for earlier versions) in order to support mixed and/or chains (previously only supported all and or all or) with, crucially, nesting resolution being determined from jump offsets.

Example

from polars.utils.udfs import BytecodeParser

fn = lambda x: x > 1 and x != 2 or x % 2 == 0 and x < 3
BytecodeParser( fn,"expr" ).to_expression( col="x" )
# (pl.col("x") > 1) & (pl.col("x") != 2) | ((pl.col("x") % 2) == 0) & (pl.col("x") < 3)

Notice that we maintain the parenthesised inner or block:

fn = lambda x: x > 1 and (x != 2 or x % 2 == 0) and x < 3
BytecodeParser( fn,"expr" ).to_expression( col="x" )
# (pl.col("x") > 1) & ((pl.col("x") != 2) | ((pl.col("x") % 2) == 0)) & (pl.col("x") < 3)

Also:

• Tidied-up the loose opname-related sets/dicts/etc; all now bundled within a single OpNames class.

[alexander-beedie]

Working nicely in 3.11 now; will take care of earlier versions after I come back from watching Oppenheimer :)

[ritchie46]

I just happened to stumble on this SO question. 😄

https://stackoverflow.com/questions/76767485/how-to-achieve-pl-colcode-applylambda-x-x88-if-lenx-2-else-x-with

[MarcoGorelli]

Just got a minor comment, but this looks really good!