Compare structural similarity

Use these library helpers when you already have Python function ASTs and want a cheap comparison of statement shapes. This is separate from the docstring-based echo_code and echo_check tools.

Python

import ast

from axm_echo import jaccard_similarity, statement_set

left = ast.parse("def first(x):\n    return x + 1\n").body[0]
right = ast.parse("def second(y):\n    return y + 99\n").body[0]
assert isinstance(left, ast.FunctionDef)
assert isinstance(right, ast.FunctionDef)

score = jaccard_similarity(statement_set(left), statement_set(right))
assert score == 1.0

The functions deliberately compute different values. Their normalized shapes match because constants and Name identifiers are replaced. A score of 1.0 is therefore not semantic equivalence.

statement_set() flattens supported compound statement bodies and returns a frozenset[str]; order and repetition are lost. The flattening includes if/loop branches, with bodies and ordinary try handlers/finally blocks. It does not uniformly flatten every Python construct: match and try* are not special-cased. Attribute names and function argument declarations are not all erased by the identifier normalization.

For a single statement, normalize_dump(stmt) returns a normalized string or None if dumping fails. flatten_body(body) exposes the flattening step when needed. jaccard_similarity(a, b) returns intersection/union: two empty sets score 1.0, exactly one empty set scores 0.0.

These helpers use the standard-library AST and do not load torch. Importing the package still loads its normal Python dependencies; this is not a separate dependency-free installation.

See Python API for signatures.