perf(text): resolve each font slot once per run

[eliahilse]

TL;DR:

Summary:

• measurement resolved and revalidated a run's font fallback chain for every character; w:rFonts fixes the family per slot and bold/italic only vary by whether the slot is complex-script, so (family, bold, italic) is a function of the slot within a run
• resolve the four slots at most once each instead, with no cache, cap, key or normalisation involved; a chain too long to be worth keeping is rebuilt per character exactly as before, so peak memory is never above the base either
• adds the slot-routing coverage the suite had none of: every character class must measure through its own slot's family, w:hint="eastAsia" must move only the ambiguous ones, and both hold when the chain is rebuilt per character rather than kept

Test plan:

• cargo test --workspace green (85 suites, 0 failures)
• cargo clippy --workspace --all-targets --all-features -- -D warnings and cargo fmt --all -- --check clean
• measurement output byte-identical to the base across 45,496 differential cases (132 MB of serialized output) covering per-slot families that differ from each other, w:hint="eastAsia", complex-script style overrides, combining marks, list markers, small caps, bidi and every error path
• benchmarked against the base, interleaved and in release, minimum of ten rounds: no case slower, from 1.00x on a one-character-per-slot run to 0.32x on a long run
• every new assertion verified to fail against a deliberately broken slot mapping and against both ways the rebuilt-chain buffer could leak between slots

[openooxml-bot]

All contributors have signed the CLA — thank you! ✍️ ✅

_{Posted by the CLA bot.}

[greptile-apps]

Greptile Summary

This PR introduces a per-measurement font-chain memo keyed by normalized family and style, then routes paragraph, marker, tab, field, and text-run lookup through it.

• Caches resolved chains and validation verdicts for one measure_paragraph invocation.
• Bounds the memo at 128 entries and recomputes beyond that limit.
• Adds tests covering key separation, cached errors and verdicts, store isolation, and MeasureInput thread-safety.
• Updates the text crate changeset and affected measurement call signatures.

Confidence Score: 4/5

The PR appears safe to merge, with only non-blocking cleanup needed for overly verbose comments and docstrings.

The memo uses the same key normalization and lookup semantics as the existing implementation, remains scoped to one immutable store/input pair, and preserves validation and error behavior across the changed measurement paths.

Files Needing Attention: crates/ooxml-text/src/measure/chain_memo.rs, crates/ooxml-text/src/measure/mod.rs

Important Files Changed

Filename	Overview
crates/ooxml-text/src/measure/chain_memo.rs	Adds bounded, per-call chain and validation memoization with focused behavior tests; explanatory documentation exceeds the repository’s concise-comment convention.
crates/ooxml-text/src/measure/mod.rs	Creates one memo per paragraph measurement and passes it consistently through empty-paragraph, marker, and run preparation paths.
crates/ooxml-text/src/measure/prepare.rs	Replaces direct font-chain resolution and validation with behavior-equivalent memo calls across text, tab, field, and fallback-metrics paths.
crates/ooxml-text/src/measure/list_marker.rs	Reuses the per-call memo for list-marker font-chain validation without changing marker measurement semantics.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    M[measure_paragraph] --> C[Create per-call ChainMemo]
    C --> K[Normalize family and style key]
    K --> H{Memo hit?}
    H -->|Yes| R[Reuse resolved chain]
    H -->|No| I[Resolve from MeasureInput fontChains]
    I --> B{Below 128 entries?}
    B -->|Yes| S[Store chain result]
    B -->|No| R
    S --> R
    R --> V{Validation verdict cached?}
    V -->|Yes| U[Reuse verdict]
    V -->|No| F[Validate IDs against FontStore]
    F --> U
    U --> P[Prepare marker and runs]

Loading

_{Reviews (1): Last reviewed commit: "perf(text): memoize font chain resolutio..." | Re-trigger Greptile}

[greptile-apps]

Trim redundant memo documentation

The module documentation and related inline comments repeatedly explain lifetime and bounded-cache behavior already expressed by the implementation, increasing maintenance and documentation-drift costs. Keep the module summary concise and remove similarly redundant explanations at the memo declaration and integration point.

Context Used: CLAUDE.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!