Python generated checks: name the invariant and keep rejected cases

Operation contract

The parser accepts signed decimal integer text and must round-trip every number in a bounded range. The loop checks that invariant for 95 inputs, then tests three invalid forms separately. This is a deterministic regression fixture, not exhaustive proof for every integer string and not a replacement for a property-testing engine that shrinks failures.

Failure and ownership boundary

Varying inputs without an assertion does not test anything. Random generation needs a recorded seed and a strategy that includes boundaries; a fixed range is easier to reproduce but narrower. Define whether whitespace, plus signs, leading zeroes, Unicode digits, and overflow are valid before broadening the input space.

Working program

def parse_signed_decimal(raw):
    digits = raw[1:] if raw.startswith("-") else raw
    if not digits or not all("0" <= character <= "9" for character in digits):
        raise ValueError("not signed decimal")
    return int(raw)

checked = 0
for quantity in range(-47, 48):
    assert parse_signed_decimal(str(quantity)) == quantity
    checked += 1

rejected = 0
for raw in ("", "47x", " 47"):
    try:
        parse_signed_decimal(raw)
    except ValueError:
        rejected += 1
print("round_trips", checked)
print("rejected", rejected)

Output

round_trips 95
rejected 3

Costs and limits

The shown loop runs O(n × digits) checks with constant state. A broader generated suite needs a runtime budget and retained failing seeds or examples.

Common Mistakes

• Generated values do not prove the invariant outside their domain.
• Keep rejected-input tests separate from success round trips.
• An unrecorded random seed makes an intermittent failure hard to reproduce.

Connected lessons

• Python unittest subTest: label several cases without aborting the method
• Python mutation testing: prove a suite rejects selected wrong implementations
• Python refactoring: preserve rejected inputs as well as accepted results

Test this boundary.