An unannotated def returning a COMPARISON typed its result from the operands

PyInferDefRetType reads the first return <expr> and asks PyInferExprType for its type, and that scanner had no case for a comparison or a membership test — it typed the expression by the OPERANDS it recognised:

def is_bool_setting(section, key, value):
    return (section, key) in BOOLEAN_SETTINGS or value in {"0", "1"}   # -> TPyList
def bigger(a, b):
    return a > b                                                       # -> str

So the def's REGISTERED result type disagreed with what the body actually returns. The caller read a Boolean as an object pointer (or a string handle) and crashed on the next use — silently wrong before it is fatal, and it hits the most ordinary Python there is: a predicate.

Also fixed with it: tuple/list VALUE equality. A tuple lowers to a TPyList, so ("Options", "Debug") in BOOLEAN_SETTINGS compared two distinct objects by IDENTITY and answered False where Python compares element by element (PyVarEq now compares two TPyList payloads element-wise, recursively).

Fix

PyInferExprType answers tyBoolean for a depth-0 comparison / membership operator (== != < <= > >= in) or a leading not. and / or are deliberately NOT included: Python hands back an OPERAND, and NilPy implements that.

Gate

test/test_nilpy_membership_bool_return.npy, CPython-diffed; make test-nilpy; self-host fixedpoint byte-identical.

Log

• 2026-07-28 — resolved, commit dcbca98c1.

Resolution

Fixed by dcbca98c1 ("fix(nilpy): return inference must agree between passes"). Re-verified 2026-07-28: the ticket's is_bool_setting (a membership test ORed with another) and bigger (a comparison) both return the Boolean CPython does. The ticket was left in backlog/ by that commit.