Truncate Python integer division in integer arithmetic so long quotients stay exact [patch]

[matt-edmondson]

Fixes #192

What changed

PythonGenerator.GenerateBinary lowered integer / to int(a / b) and integer % to a - b * int(a / b). Python's / always returns a float. For a long above 2^53 the quotient was rounded before it was truncated, so the quotient came out wrong and the remainder could come out negative. For example, n % 7 with n = 1700000000123456789 returned -75 instead of 2.

The quotient is now truncated using integer arithmetic only:

(a // b if (a < 0) == (b < 0) else -(-a // b))

When the operands share a sign, flooring and truncating agree. When they don't, negating the dividend turns floor division into truncation. The remainder is still a - b * <quotient>. As before, an operand that is a variable or a literal is written inline, and anything else (such as a call) is bound once through a lambda. The IntegerDivisionTests call-count assertion still checks that.

int operands get the same lowering. It is exact for them too, and one form for both keeps the generator simple.

Tests (IntegerDivisionTests)

Python_LongOperandsAbove2To53_MatchTheCFamily runs the generated Python and compares it with C#'s / and %. It covers:

• long operands passed as parameters, through a call operand (the lambda path), and as the issue's repro (long n = 1700000000123456789; return n % 7)
• all four sign combinations, 9007199254740993 / 1, long.MaxValue, and long.MinValue + 1

The test module starts with long = int, because the generator still writes long as an undefined annotation (#109, open PR #124). That keeps this test about the arithmetic.

With the generator change reverted, the test fails on its first check: div_long_literal() = 242857142874779552, expected 242857142874779541. With the change, the existing Python_MatchesTheCFamily cases still pass, and the full suite passes (1116/1116).

The JavaScript side (long values outside ±(2^53−1)) is a separate issue, #168, and is not touched here.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CkCrYLchcFZxtGF9DyXZHi

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud