perf: replace string concatenation with arithmetic in DecimalType.serialize (hundreds of ns, x1.3-2.6 speedup)

[mykaul]

Summary

• Replace int(''.join([str(d) for d in digits])) with a multiply-and-add loop in DecimalType.serialize()
• Eliminates per-digit str() allocations and the intermediate joined string

Details

Decimal.as_tuple() returns digits as a tuple of ints (0-9). The previous code converted each digit to a string, joined them, and parsed the result back to int. The new code uses simple arithmetic: n = n * 10 + digit.

Before

unscaled = int(''.join([str(digit) for digit in digits]))

After

unscaled = 0
for digit in digits:
    unscaled = unscaled * 10 + digit

Benchmark results (isolated digit-to-integer conversion)

Digit count	String join (ns)	Arithmetic (ns)	Speedup
3	228	84	2.70x
20	1079	681	1.58x
31	1502	1125	1.34x

Benchmark script included at benchmarks/decimal_serialize.py.

Testing

• All 651 existing unit tests pass (16 pre-existing skips)
• Correctness verified via round-trip serialize/deserialize for small, medium, and large Decimal values

[mykaul]

Equivalence verification

Rebased onto current master (fast-forwarded cleanly, no conflicts) and re-verified that the arithmetic rewrite produces byte-identical serialize() output vs. the original int(''.join(str(d) for d in digits)) implementation.

Checked directly against the live DecimalType.serialize/deserialize round trip and against a side-by-side old-vs-new comparison harness covering:

• zero in all its as_tuple() forms (0, -0, 0.0, 0E+5, 0E-5, ...)
• negative values
• negative scale / positive exponent (e.g. 1E+2, 9E+50)
• very large magnitude (100-300 digit unscaled values)
• very small magnitude / very negative exponent (e.g. 1E-200)
• all-9s / boundary digit patterns
• 20,000 randomized fuzz cases (random digit strings, sign, exponent in [-80, 80])

Result: all of the above match byte-for-byte between old and new implementations. Added an explicit Decimal('0') fixture to tests/unit/test_marshalling.py's marshalled_value_pairs (the existing table had no exact-zero case) so this is now covered by a permanent regression test rather than just ad-hoc verification.

One genuine (non-blocking) behavioral difference found: for Decimal('NaN') / Decimal('sNaN') (as_tuple() returns empty digits), the old code raised ValueError (from int('')), while the new code raises TypeError (from int32_pack(-exponent), since exponent is the string 'n' for NaN). This isn't a new correctness bug: CQL's decimal type has no representation for NaN, so both versions were already incapable of serializing it, and the old code was already inconsistent here too (it raised TypeError for Decimal('Infinity'), same as the new code does). No existing code or test depends on the specific exception type. Flagging for visibility; not treating it as blocking.

CI / review status

All CI checks pass (unit tests across all supported Python versions/event loops, wheel builds, build, snyk). No reviews or review threads on the PR to resolve.

Test results

• tests/unit/test_types.py: pass
• tests/unit/test_marshalling.py (incl. new zero-decimal case): pass
• Full tests/unit/: 720 passed, 88 skipped, 0 failed

Amended the fix into the existing commit and force-pushed (kept as draft).

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Optimizes decimal serialization by replacing string conversion with arithmetic accumulation.

Changes:

• Adds multiply-and-add digit conversion.
• Tests zero serialization.
• Adds performance and round-trip benchmarks.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
cassandra/cqltypes.py	Optimizes decimal digit conversion.
tests/unit/test_marshalling.py	Adds zero-value coverage.
benchmarks/decimal_serialize.py	Adds serialization benchmarks.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.