Skip to content

Reduce CSV conversion memory while preserving upstream output - #13

Merged
DevBehindYou merged 1 commit into
mainfrom
codex/csv-parser-memory
Oct 28, 2026
Merged

DevBehindYou merged 1 commit into
mainfrom
codex/csv-parser-memory

Conversation

@DevBehindYou

Copy link
Copy Markdown
Owner

Large CSV conversions retained every parsed row and a separate list of formatted rows before constructing the complete Markdown. Both normal and archive document engines now use a row-at-a-time CSV formatter for verified MarkItDown 0.1.8/0.1.9, preserving the upstream output and retaining the original converter on unverified versions or a missing helper. The parser-capacity audit records existing limits and the next workbook/native/response-memory boundary; queue concurrency is unchanged.

Three fresh processes per variant, with generated multilingual CSV and identical before/after output hashes: two synchronized 50,000-row conversions used median peak process memory of 297.19 → 221.92 MiB without tracing (about 25% lower), with similar elapsed time. The small fixture's timing regressed; no universal speed, host-capacity or billing claim is made. The full input and output remain materialized.

Validation:

  • MarkItDown 0.1.9: 118 normal + 98 archive tests passed locally.
  • 0.1.8: 25 output/fallback tests passed; 0.1.7: 3 fallback/integrity checks passed with 22 optimization-only cases intentionally skipped.
  • Output comparisons include 100 deterministic ragged/quoted tables, encodings/BOMs, pipes/backslashes/newlines, blank runs, wide rows, error/stream ownership, unknown-version behavior and direct/archive wiring.
  • Existing shared-package integrity checks are preserved; the identical N/Z adapters live outside OCR's shared package. No dependency, UI/blog, database or concurrency changes.
  • Pre-release generated multipart CSV/ZIP probes passed on all four live document workers, including the final Unicode marker. These probes create no Storage jobs and are distinct from browser/Storage-backed acceptance.

@vercel

vercel Bot commented Oct 28, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
llmiser Ready Ready Preview Oct 28, 2026 5:20pm UTC
llmiser-api-n1n Ready Ready Preview Oct 28, 2026 5:20pm UTC
llmiser-api-n2n Ready Ready Preview Oct 28, 2026 5:20pm UTC
llmiser-api-z1z Ready Ready Preview Oct 28, 2026 5:20pm UTC
llmiser-api-z2z Ready Ready Preview Oct 28, 2026 5:20pm UTC

@DevBehindYou
DevBehindYou merged commit 165d2b0 into main Oct 28, 2026
14 checks passed
@DevBehindYou
DevBehindYou deleted the codex/csv-parser-memory branch October 28, 2026 17:23

This branch was successfully deployed

1 active deployment
Preview – mdify-app — dc9730aa Deployed Oct 28, 2026 by vercel[bot]
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.

2 participants