[baseballyama]

Summary

loadWorkbookStream's iterRows was several times slower than it needed to be on wide sheets where almost every cell is empty but styled (<c r="X1" s="n"/>). The row walk now drives saxes directly instead of consuming a per-event async iterator, so it runs faster than loadWorkbook on the same bytes instead of 1.5–3x slower.

Motivation

#302 reports ~46 µs per cell on a 30,000 × 227 sheet with 0.4% of cells holding a value. A profile showed saxes itself was a small part of the time. Most of it went to the iterParse layer: one event object per SAX event, Clark-notation names and attribute maps, and an async generator step for each event.

Closes #302

Changes

• iterSheetRows (behind ReadOnlyWorksheet.iterRows / iterValues) registers its handlers on a SaxesParser and collects finished rows synchronously. It crosses an async boundary once per 64K feed instead of once per event. Rows completed before a content-limit error or a corrupt cell are still yielded before the error.
• The feed / decode / DTD-prescan driver moved from src/xml/iterparse.ts to the internal src/xml/sax-feed.ts, and iterParse is now built on it. It lives in its own module so no public .d.ts imports saxes types. saxes' own declarations fail under skipLibCheck: false, and check:consumer caught this.
• coordinateToTuple parses the plain "AB12" form from char codes before falling back to the regex. Both readers call it once per cell.

Testing

• Existing streaming / xml / worksheet suites pass unchanged (pnpm test: 3761 tests).
• New perf gate tests/perf/iter-rows-throughput.test.ts: 5,000 × 227 styled cells, comparing iterRows time against loadWorkbook on the same bytes. On main it fails (ratio 1.12). On this branch it passes (0.81–0.84).
• Manual bench, 10,000 × 227 sheet built with the reproduction from the issue, alternating runs on the same machine: main 4.2–19.9 s, this branch 0.9–3.2 s.

PERF_GATE=1 pnpm test:perf tests/perf/iter-rows-throughput.test.ts

Breaking changes

None

Checklist

• I have read CLAUDE.md and followed the project's conventions.
• I have added or updated tests for the change.
• I have added or updated documentation where user-visible behavior changed.
• If this is a breaking change, I have added a changeset / CHANGELOG entry and
  flagged it above.
• I have re-read my own diff and removed dead code, debug prints, and stale comments.
• If I used an LLM to draft this PR, I have verified each change myself, the PR
  represents real work that warrants a maintainer's review, and I am willing to
  defend each line in review.

🤖 Generated with Claude Code

https://claude.ai/code/session_0142ArZjzXh8UP4dVS623qG3

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.