[joe-sharp]

Two memoization gaps in VersionGem::Api, fixed as separate commits.

1. _to_a used = instead of ||= (⚡️ Memoize split version string in _to_a). The VERSION string was split again on every call to major, minor, patch, and pre.
2. pre never cached nil (⚡️ Cache nil pre-release in pre). ||= treats nil as unset, so a version with no pre-release re-ran _to_a on every pre call. The fix checks defined?(@pre) instead.

Each commit adds a spec that fails without its fix, a Keep a Changelog entry under Unreleased → Fixed, and a DCO sign-off.

Summary

Cold lookups are about 30% faster, and warm pre on a version with no pre-release is about 9× faster. The split count drops from 4–6 per cold lookup to 1.

Benchmark

Ruby 4.0.9, 20,000 iterations per run, median of 7 runs. splits/op counts String#split calls per operation. Three interleaved runs of main and three of this branch, rerun on the same machine:

Scenario	splits/op main → branch	µs/op main (3 runs)	µs/op branch (3 runs)
Cold lookup, 1.2.3 (major, minor, patch, pre, to_h, to_a)	6 → 1	15.47, 14.34, 14.34	10.35, 10.15, 10.34
Cold lookup, 1.2.3.pre-78	4 → 1	15.07, 13.55, 13.85	10.32, 10.40, 10.73
Cold lookup, Epoch 3012.34.56	6 → 1	18.27, 17.69, 18.30	13.75, 13.41, 13.42
Warm pre, 1.2.3 (no pre-release)	1 → 0	0.80, 0.81, 0.88	0.085, 0.084, 0.083
Warm pre, 1.2.3.pre-78	0 → 0	0.082, 0.083, 0.081	0.083, 0.083, 0.083
Warm major	0 → 0	0.084, 0.081, 0.081	0.080, 0.080, 0.081
Warm to_a (Epoch)	0 → 0	0.084, 0.084, 0.083	0.081, 0.083, 0.081

Takeaways

• Cold lookups are about 30% faster (≈14.7 µs → ≈10.3 µs) and split the version string once instead of 4–6 times.
• Warm pre on a version with no pre-release is about 9× faster (≈0.83 µs → ≈0.08 µs). This is the fix 2 win.
• Warm lookups for cached values are unchanged, as expected.
• Caveat: these were run on a laptop with other load, so absolute numbers are noisy. The split counts are exact.

Reproducing

The benchmark script is not in the repo. It loads lib/ directly, prepends a counter onto String#split, and times ITER calls of each scenario. Run it with:

ruby -I lib bench_api.rb <label>

Run it against a checkout of main (git archive main lib) and against this branch to compare.

Testing

• bundle exec rspec: 32 examples, 0 failures.
• bundle exec rubocop lib/version_gem/api.rb spec/version_gem/api_spec.rb: no offenses.
• bundle exec kettle-test was not run, because mise is not installed in the environment used for this change. The full RSpec suite was run directly instead.

🤖 Generated with Claude Code