Skip to content

Avoid repeated dynamic-option classification arrays - #283

Open
OskarEichler wants to merge 1 commit into
trailblazer:masterfrom
OskarEichler:codex/option-classification
Open

Avoid repeated dynamic-option classification arrays#283
OskarEichler wants to merge 1 commit into
trailblazer:masterfrom
OskarEichler:codex/option-classification

Conversation

@OskarEichler

Copy link
Copy Markdown

Summary

Compute dynamic option names once per runtime-options rebuild and use direct short-circuit Proc/Symbol/Uber::Callable checks. Classification order is preserved.

Reproduction

require 'representable/json/hash'
1000.times { Representable::Definition.new(:name, getter: ->(*) { 1 }, default: 7, as: :alias) }
# Measure elapsed time and GC.stat(:total_allocated_objects) around this bounded loop.

Verification

  • Existing master suite on this isolated branch: 545 tests / 701 assertions, zero failures/errors, three existing skips on Ruby 3.2.11. Ruby 4.0.6 retains the same eight legacy inspection-format failures, zero errors and three skips. Existing tests/expectations are unchanged.
  • 308 focused checks on Ruby 3.2.11/4.0.6 for this change. XML first-render uses a separate fresh-process check.
  • Combined release-based verification with other focused fixes and attributed Use Setter in OverwriteOnNil if a :setter is defined #177/Hash value of null breaks parsing #264/Update MultiJSON usage to address deprecations #273 source backports: 544 tests / 700 assertions pass on Ruby 3.2.11, plus 4,410 focused checks and 36 boundary checks. Package retains 104 paths and unchanged compared metadata. All 36 runtime files compile on both Rubies; no newly introduced default Lint offenses.

Isolated YJIT measurement: three alternating baseline/fixed process pairs, each with five batches of 3,000 definition builds. Median time 0.012017→0.009716s; allocations 234,000→171,000. Bounded local evidence, not production throughput.

Breaking changes / limitations

No documented API change intended. A custom dynamic_options override supplies the same list, now evaluated once per rebuild instead of once per field.

No new/modified checked-in tests. Ruby 2.4–3.1, JRuby, Windows and Linux were not executed locally. Ruby >=2.4 and runtime dependency ranges remain unchanged. External test tooling supplies a legacy MiniTest alias and ruby-prof 1.7.x; no assertions were changed. No production verification is claimed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant