ofthemachine/libreoffice repository overview

⁠libreoffice — headless LibreOffice + Python office stack

Purpose-built RCTX for office document workflows: docx/pptx/xlsx ↔ PDF conversion, Word-accurate page-count via rendered PDF, OOXML manipulation, and PDF post-processing (split/merge/extract/linearize).

⁠Why this exists

ofthemachine/python3 already carries the Python office libraries (python-docx, pypdf, etc.) but has no native renderer. For accurate page counts, format conversion, or PDF rendering you need a real office engine. LibreOffice is the de-facto open-source choice and runs headlessly under a non-privileged user.

This RCTX was proposed (and validated) against a real use case: building an operon TOOL for cross-machine docx page-counting. A pure-Python structural heuristic gave estimates within ±1-2 pages of Word on typical multi-page docx inputs; a soffice-backed TOOL would deliver Word-exact counts portably.

⁠Key tools

⁠Fonts

Curated minimal set for consistent pagination across rebuilds:

• fonts-liberation + fonts-liberation2 — Arial/Times/Courier equivalents (metric-compatible)
• fonts-noto-core — broad Latin / common-script coverage
• fonts-dejavu-core — fallback

fc-cache -f is run at build time so soffice sees them immediately. Skipped: fonts-noto-cjk (~200 MB; add in a derived image if needed).

⁠Image size

~700 MB. Most of the weight is LibreOffice itself (~400 MB) and the curated fonts (~100 MB). Python libs are slim by comparison.

⁠Quick examples

Build locally:

make libreoffice
make libreoffice R=1    # build + run smoke test
make libreoffice I=1    # interactive shell

Word-accurate page count in a fraglet:

import subprocess, re, os
r = subprocess.run(
    ["soffice", "--headless", "--convert-to", "pdf",
     "--outdir", "/tmp", "/input.docx"],
    capture_output=True, check=True, timeout=120,
)
pdf = "/tmp/" + os.path.splitext(os.path.basename("/input.docx"))[0] + ".pdf"
info = subprocess.run(["pdfinfo", pdf], capture_output=True, text=True, check=True)
pages = int(re.search(r"Pages:\s+(\d+)", info.stdout).group(1))
print({"pages": pages, "method": "soffice+pdfinfo"})

⁠Pairing with operon TOOLs

Register a TOOL anchored to CODE that runs in this RCTX:

# 1. Record the CODE under a ledger ref
operon run ofthemachine/libreoffice page_count_soffice.py \
  --record-ref tooling/docx/page_count_soffice

# 2. Register a TOOL anchored to the CODE
operon tool register tools/docx/page_count_soffice \
  --from-code tooling/docx/page_count_soffice

# 3. Invoke with a docx FILE mounted at /input.docx
operon invoke tools/docx/page_count_soffice \
  --file <docx-ref>:/input.docx

The TOOL ref tools/docx/page_count stays as the stable invocation surface; swapping the backing CODE/RCTX is transparent to consumers.