multidiff

[DiTo97]

Is your feature request related to a problem? Please describe.
multidiff feature handling SQL statements with more than one expressions.

Describe the solution you'd like
I have implemented such feature as part of a text-to-SQL suite, opting for a Levenshtein-like approach if the two statements have a different number of expressions using empty expressions as filler padding and adjusting the diff accordingly.

from typing import Any, overload

from sqlglot import diff, parse, Expression
from sqlglot.diff import Keep, Insert, Move, Remove, Update
from sqlglot.optimizer import optimize


padding = Expression()
Edit = Keep | Insert | Move | Remove | Update


@overload
def multidiff(
    source: list[Expression],
    target: list[Expression],
    schema: dict[str, Any] | None = None,
) -> list[list[Edit]]:
    ...


@overload
def multidiff(
    source: str,
    target: str,
    dialect: str | None = None,
    schema: dict[str, Any] | None = None,
) -> list[list[Edit]]:
    ...


def multidiff(source, target, dialect=None, schema=None):
    source = parse(source, dialect=dialect) if isinstance(source, str) else source
    target = parse(target, dialect=dialect) if isinstance(target, str) else target

    if schema:
        source = [optimize(S, schema=schema) for S in source]
        target = [optimize(T, schema=schema) for T in target]

    maxlen = max(len(source), len(target))

    source += [padding] * (maxlen - len(source))
    target += [padding] * (maxlen - len(target))

    multidifference = []

    for S, T in zip(source, target):
        difference = diff(S, T)
        
        if S == padding: difference.pop(0)
        if T == padding: difference.pop()
            
        multidifference.append(difference)

    return multidifference


def get_diff_distance(difference: list[Edit]) -> int:
    return sum(0 if isinstance(E, Keep) else 1 for E in difference)


def get_multidiff_distance(multidifference: list[list[Edit]]) -> int:
    return sum(map(get_diff_distance, multidifference))

[tobymao]

this is very cool but we won't be implementing this in core, feel free to make a library for this if you would like