Combined: cache color prefix, short-circuit newline split, and cache humanize

[alanzabihi]

Hypothesis

Combining three independent optimizations in formatArgs can achieve a compound improvement that individually each falls below the 3ms tolerance but together exceeds it:

1. Pre-compute ANSI color prefix at init (thesis Pre-compute ANSI color prefix per instance at init time #5 direction)
2. Short-circuit split/join when no newline present (thesis Short-circuit split/join for single-line messages in formatArgs #6 direction)
3. Cache ms.humanize result for repeated identical diffs (thesis Cache ms.humanize result for identical consecutive diffs #12 direction)

Rationale

Prior theses #5 and #6 were each decided as non_improvement individually, likely because each saved only 1-2ms on the current baseline. But formatArgs runs on every enabled call and these three optimizations target independent bottlenecks within it:

• Prefix computation: eliminates per-call string concatenation for ANSI codes
• Newline check: avoids split/join array allocation for 75% of calls
• Humanize cache: avoids ms library call when diff repeats

Combined, these should produce a measurable improvement.

Scope

• src/node.js init(): cache _colorCode and _colorPrefix on the instance
• src/node.js formatArgs(): use cached prefix, add indexOf newline guard, cache humanize result

Expected impact

Moderate to high. Each optimization independently saves ~1-2ms; combined they should exceed the 3ms tolerance.

Risks

Must verify the combination does not interact negatively. Each optimization is independent (different lines of formatArgs) so interference is unlikely.

[alanzabihi]

Polyresearch approval: thesis #13.

[alanzabihi]

Polyresearch claim: thesis #13 by node alanzabihi/hetzner-d789.

[alanzabihi]

Polyresearch attempt: thesis #13, branch thesis/13-combined-cache-color-prefix-short-circuit-newline-split-and-cache-humanize, metric 247.0400, observation improved.

[alanzabihi]

Polyresearch release: thesis #13 by node alanzabihi/hetzner-d789 (reason: no_improvement).