SPB Git

Repository Public License

documents.md 66.3 KB

# Documents, ebooks and presentations — engine notes

Package Packages/EngineDocuments (module EngineDocuments, factory makeEngines()), three engines: documents, ebooks, presentations. Everything below was measured on 2026-11-21 on the development Mac (Apple Silicon, 18 cores, macOS 27.0) with the bundled tools only.

Test result (swift test in Packages/EngineDocuments): ✔ Test run with 35 tests in 7 suites passed after 74.302 seconds. (35 test functions, run on 2026-11-21 after the fixes listed under “Fixed on 2026-11-21”, with ORIX_CONVERT_TOOLS pointing at EmbeddedEngines/tools; the parametrised ones run 99 declared document / ebook paths, 24 declared presentation paths and 43 alternate-backend combinations.)

# Bundled components

Component Version How it runs Size on disk License
Pandoc 3.14, official arm64 build, unmodified separate process 196 964 448 bytes (188 MiB), one executable GPL-2.0-or-later (separate executable, see EmbeddedEngines/pandoc/NOTICE.md)
LibreOffice 26.8.3, official aarch64 build, slimmed separate process, one per conversion, throw-away profile 471 472 KiB (≈ 460 MiB), 2 978 files, 121 Mach-O files — official bundle: 827 784 KiB, 218 Mach-O files MPL-2.0 + third-party notices (see EmbeddedEngines/libreoffice/NOTICE.md)
Apple text system macOS (public AppKit NSAttributedString readers / writers) in process — —
WebKit macOS (WKWebView, NSPrintOperation) system web-content process — —

Together the two helpers add about 650 MiB to the application. That is large; it is the honest cost of converting DOC / PPT / PPTX / ODP with real layout without depending on software installed on the Mac.

# Which backend converts what, and why

Each pair was converted with every backend able to do it, on fixtures generated from one Markdown master (headings on two levels, bold / italic, bullet and numbered lists, a table, a PNG image, a link, a footnote, a code block, inline and display math, title / author / subject / keywords / language). The output was read back by an independent reader written for this package (DOCX / ODT / EPUB / PPTX / ODP package XML, HTML scanner, PDFKit, the Apple text system for RTF and DOC) and scored on: 10 probe words, table cells, image count, link target, bold / italic, heading levels, footnote, title and author. "Automatic" is the backend that kept the most.

Findings per backend:

  • LibreOffice — the only backend that keeps page layout. Best for office ↔ office (DOCX, DOC, ODT, RTF) and for → PDF: everything on the fixture survived (10/10 words, table, image, link, emphasis, heading styles, footnote, metadata). Cost: a process start per conversion. Measured quirk: with a substituted font that has ligatures, the PDF text layer of one word came out as "zebrafissh" for "zebrafish" (the page renders correctly; searching / copying that word is affected). Its plain-text export dropped the footnote text (9/10 words), so it is not the automatic choice for → TXT. Its HTML import is not offered at all: on a hostile test page it tried to load remote resources (stopped only by the dead-end proxy) and aborted in 2 of 3 conversions.
  • Pandoc — best for anything structural: office / EPUB → Markdown, HTML, LaTeX, reStructuredText, EPUB, TXT and markup → office. Kept all ten probe words, table, link, emphasis, heading levels, footnote and math (as TeX or MathML). It does not keep page layout or fonts by design. Its RTF reader is weak: on an RTF written by a word processor it lost heading levels and the footnote and reordered lists, so RTF and DOC go through LibreOffice first (→ DOCX → Pandoc), which kept all of it. It does not read ODT metadata (the engine reads meta.xml itself and hands it over) and writes no document properties into RTF.
  • Apple text system — instant and in process, but measurably the weakest: reading DOCX / ODT it drops images and footnotes (9/10 words), it does not read DOCX hyperlinks, and its DOCX writer flattens tables into paragraphs and writes no links. It is never the automatic choice; it is offered as an alternative where it produces a valid file, and every loss is reported as a warning when it happens. It is never used to read HTML (the AppKit HTML importer loads remote resources).
  • WebKit — HTML → PDF with real pagination (NSPrintOperation: paper size, orientation, margins) or one continuous page (createPDF). JavaScript is disabled, a content-rule list blocks every URL except a private scheme that serves the document and files inside its own folder (symbolic links leading out of it are refused). Markdown, reStructuredText and EPUB → PDF go Pandoc → self-contained HTML (built-in print stylesheet, MathML) → WebKit: a reflowed print rendering, not the source's page design.

The option Engine (Automatic / Pandoc / LibreOffice / Apple text system) is offered only on pairs where more than one backend really works; every alternative is exercised by DocumentPathTests.alternateEngine (43 combinations).

# Declared paths (129)

Generated from the engines' own declarations by PathTableTests.writeTable. Every row is converted from a real fixture, validated, and its preserves claims asserted, by the test named in the last column.

Source Destination Operation Engine Status Fidelity Backend (automatic) Other selectable backends Preserves Limitations Test
docx doc — documents supported structural LibreOffice Apple text system images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. DocumentPathTests.declaredPath
docx rtf — documents supported structural LibreOffice Apple text system, Pandoc images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. DocumentPathTests.declaredPath
docx odt — documents supported structural LibreOffice Apple text system, Pandoc images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. DocumentPathTests.declaredPath
docx txt — documents supported partial Pandoc LibreOffice, Apple text system — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. Headers, footers and text boxes are not included. DocumentPathTests.declaredPath
docx markdown — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DocumentPathTests.declaredPath
docx html — documents supported structural Pandoc LibreOffice, Apple text system images, links, metadata, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. DocumentPathTests.declaredPath
docx latex — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. DocumentPathTests.declaredPath
docx rst — documents supported structural Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DocumentPathTests.declaredPath
docx pdf — documents supported structural LibreOffice — images, links, metadata, textStyles Laid out by LibreOffice, not by the application that wrote the file: line and page breaks can differ from Microsoft Word or Pages. Fonts that are not installed are substituted. Macros are never run; comments are not printed. DocumentPathTests.declaredPath
doc docx — documents supported structural LibreOffice Apple text system images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc rtf — documents supported structural LibreOffice Apple text system images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc odt — documents supported structural LibreOffice Apple text system images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc txt — documents supported partial LibreOffice → DOCX → Pandoc Apple text system — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. Headers, footers and text boxes are not included. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc markdown — documents supported structural LibreOffice → DOCX → Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc html — documents supported structural LibreOffice → DOCX → Pandoc Apple text system images, links, metadata, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc latex — documents supported structural LibreOffice → DOCX → Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc rst — documents supported structural LibreOffice → DOCX → Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
doc pdf — documents supported structural LibreOffice — images, links, metadata, textStyles Laid out by LibreOffice, not by the application that wrote the file: line and page breaks can differ from Microsoft Word or Pages. Fonts that are not installed are substituted. Macros are never run; comments are not printed. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
rtf docx — documents supported structural LibreOffice Apple text system, Pandoc images, links, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. DocumentPathTests.declaredPath
rtf doc — documents supported structural LibreOffice Apple text system images, links, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. DocumentPathTests.declaredPath
rtf odt — documents supported structural LibreOffice Apple text system, Pandoc images, links, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. DocumentPathTests.declaredPath
rtf txt — documents supported partial LibreOffice → DOCX → Pandoc Pandoc, Apple text system — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. Headers, footers and text boxes are not included. RTF input is first imported by LibreOffice. DocumentPathTests.declaredPath
rtf markdown — documents supported structural LibreOffice → DOCX → Pandoc Pandoc links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. RTF input is first imported by LibreOffice. DocumentPathTests.declaredPath
rtf html — documents supported structural LibreOffice → DOCX → Pandoc Pandoc, Apple text system images, links, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. RTF input is first imported by LibreOffice. DocumentPathTests.declaredPath
rtf latex — documents supported structural LibreOffice → DOCX → Pandoc Pandoc links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. RTF input is first imported by LibreOffice. DocumentPathTests.declaredPath
rtf rst — documents supported structural LibreOffice → DOCX → Pandoc Pandoc links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. RTF input is first imported by LibreOffice. DocumentPathTests.declaredPath
rtf pdf — documents supported structural LibreOffice — images, links, textStyles Laid out by LibreOffice, not by the application that wrote the file: line and page breaks can differ from Microsoft Word or Pages. Fonts that are not installed are substituted. Macros are never run; comments are not printed. DocumentPathTests.declaredPath
odt docx — documents supported structural LibreOffice Apple text system, Pandoc images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. DocumentPathTests.declaredPath
odt doc — documents supported structural LibreOffice Apple text system images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. DocumentPathTests.declaredPath
odt rtf — documents supported structural LibreOffice Apple text system, Pandoc images, links, metadata, tables, textStyles Converted by LibreOffice: compatibility with Microsoft Word is good but not complete (SmartArt, some fields, text effects, exact fonts and spacing can differ). Macros are never run and are not carried over. Embedded OLE objects may be reduced to their preview image. DocumentPathTests.declaredPath
odt txt — documents supported partial Pandoc LibreOffice, Apple text system — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. Headers, footers and text boxes are not included. DocumentPathTests.declaredPath
odt markdown — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DocumentPathTests.declaredPath
odt html — documents supported structural Pandoc LibreOffice, Apple text system images, links, metadata, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. DocumentPathTests.declaredPath
odt latex — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. DocumentPathTests.declaredPath
odt rst — documents supported structural Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DocumentPathTests.declaredPath
odt pdf — documents supported structural LibreOffice — images, links, metadata, textStyles Laid out by LibreOffice, not by the application that wrote the file: line and page breaks can differ from Microsoft Word or Pages. Fonts that are not installed are substituted. Macros are never run; comments are not printed. DocumentPathTests.declaredPath
txt docx — documents supported structural LibreOffice Apple text system, Pandoc — Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt doc — documents supported structural LibreOffice Apple text system — Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt rtf — documents supported structural LibreOffice Apple text system, Pandoc — Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt odt — documents supported structural LibreOffice Apple text system, Pandoc — Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt markdown — documents supported structural Pandoc — — Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt html — documents supported structural Pandoc — — Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt latex — documents supported structural Pandoc — — The result is LaTeX source; it is not compiled to PDF. Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt rst — documents supported structural Pandoc — — Plain text has no structure: paragraphs are split at blank lines and no character is interpreted as markup. DocumentPathTests.declaredPath
txt pdf — documents supported structural LibreOffice — — Laid out by LibreOffice, not by the application that wrote the file: line and page breaks can differ from Microsoft Word or Pages. Fonts that are not installed are substituted. Macros are never run; comments are not printed. Plain text is laid out by LibreOffice with a default font and page setup. DocumentPathTests.declaredPath
markdown docx — documents supported structural Pandoc — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. DocumentPathTests.declaredPath
markdown doc — documents supported structural Pandoc → DOCX → LibreOffice — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. DocumentPathTests.declaredPath
markdown rtf — documents supported structural Pandoc — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Footnotes may be flattened into the text flow; document properties are not written. DocumentPathTests.declaredPath
markdown odt — documents supported structural Pandoc — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. DocumentPathTests.declaredPath
markdown txt — documents supported partial Pandoc — — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. DocumentPathTests.declaredPath
markdown html — documents supported structural Pandoc — images, links, metadata, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. DocumentPathTests.declaredPath
markdown latex — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. DocumentPathTests.declaredPath
markdown rst — documents supported structural Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DocumentPathTests.declaredPath
markdown pdf — documents supported structural Pandoc → HTML → WebKit — images, links, metadata, textStyles Rendered by WebKit with JavaScript disabled and without any network access: remote images, fonts and stylesheets are missing. Typeset with ORIX Convert's built-in print stylesheet; this is a reflowed rendering, not the source's own page design. No table structure, bookmarks or tagged-PDF accessibility information. DocumentPathTests.declaredPath
html docx — documents supported structural Pandoc — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html doc — documents supported structural Pandoc → DOCX → LibreOffice — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html rtf — documents supported structural Pandoc — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Footnotes may be flattened into the text flow; document properties are not written. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html odt — documents supported structural Pandoc — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html txt — documents supported partial Pandoc — — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html markdown — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html latex — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html rst — documents supported structural Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
html pdf — documents supported structural WebKit — images, links, metadata, textStyles Rendered by WebKit with JavaScript disabled and without any network access: remote images, fonts and stylesheets are missing. No table structure, bookmarks or tagged-PDF accessibility information. DocumentPathTests.declaredPath
latex docx — documents supported structural Pandoc — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
latex doc — documents supported structural Pandoc → DOCX → LibreOffice — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
latex rtf — documents supported structural Pandoc — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Footnotes may be flattened into the text flow; document properties are not written. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
latex odt — documents supported structural Pandoc — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
latex txt — documents supported partial Pandoc — — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
latex markdown — documents supported structural Pandoc — links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
latex html — documents supported structural Pandoc — images, links, metadata, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
latex rst — documents supported structural Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
rst docx — documents supported structural Pandoc — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. DocumentPathTests.declaredPath
rst doc — documents supported structural Pandoc → DOCX → LibreOffice — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. DocumentPathTests.declaredPath
rst rtf — documents supported structural Pandoc — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Footnotes may be flattened into the text flow; document properties are not written. DocumentPathTests.declaredPath
rst odt — documents supported structural Pandoc — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. DocumentPathTests.declaredPath
rst txt — documents supported partial Pandoc — — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. DocumentPathTests.declaredPath
rst markdown — documents supported structural Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DocumentPathTests.declaredPath
rst html — documents supported structural Pandoc — images, links, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. DocumentPathTests.declaredPath
rst latex — documents supported structural Pandoc — links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. DocumentPathTests.declaredPath
rst pdf — documents supported structural Pandoc → HTML → WebKit — images, links, textStyles Rendered by WebKit with JavaScript disabled and without any network access: remote images, fonts and stylesheets are missing. Typeset with ORIX Convert's built-in print stylesheet; this is a reflowed rendering, not the source's own page design. No table structure, bookmarks or tagged-PDF accessibility information. DocumentPathTests.declaredPath
docx folder media.extract documents supported lossless ZIP reader (own code) — images Only the embedded image and media files are extracted, byte for byte, under their names in the package; linked (external) images are not fetched. EbookTests.coverAndMedia
odt folder media.extract documents supported lossless ZIP reader (own code) — images Only the embedded image and media files are extracted, byte for byte, under their names in the package; linked (external) images are not fetched. EbookTests.coverAndMedia
epub docx — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub doc — ebooks supported structural Pandoc → DOCX → LibreOffice — images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Word 97–2004 is a legacy format: newer features (modern text effects, content controls) have no equivalent. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub rtf — ebooks supported structural Pandoc — images, links, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. Footnotes may be flattened into the text flow; document properties are not written. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub odt — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles The document is rebuilt from its structure (headings, lists, tables, images, links, notes); the source's visual styling is replaced by default styles. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub txt — ebooks supported partial Pandoc — — Text only: fonts, emphasis, images, table grids, links and page layout are dropped. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub markdown — ebooks supported structural Pandoc — chapters, links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub html — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles Page layout, headers and footers are dropped. Images are embedded in the HTML file as data URIs. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub latex — ebooks supported structural Pandoc — chapters, links, metadata, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. The result is LaTeX source; it is not compiled to PDF. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub rst — ebooks supported structural Pandoc — chapters, links, tables, textStyles Images are kept as references only; the image files are not written (use “Extract media” on the source). Page layout, fonts, colours, headers and footers are dropped; complex or merged-cell tables are simplified. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
epub pdf — ebooks supported structural Pandoc → HTML → WebKit — images, links, metadata, textStyles Rendered by WebKit with JavaScript disabled and without any network access: remote images, fonts and stylesheets are missing. Typeset with ORIX Convert's built-in print stylesheet; this is a reflowed rendering, not the source's own page design. The book's own CSS, embedded fonts and page breaks are not used. No table structure, bookmarks or tagged-PDF accessibility information. DRM-protected books are refused. Obfuscated embedded fonts are not carried over. DocumentPathTests.declaredPath
docx epub — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. DocumentPathTests.declaredPath
doc epub — ebooks supported structural LibreOffice → DOCX → Pandoc — chapters, images, links, metadata, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. Word 97–2004 input is first imported by LibreOffice. DocumentPathTests.declaredPath
rtf epub — ebooks supported structural LibreOffice → DOCX → Pandoc Pandoc chapters, images, links, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. RTF input is first imported by LibreOffice. DocumentPathTests.declaredPath
odt epub — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. DocumentPathTests.declaredPath
txt epub — ebooks supported structural Pandoc — — Plain text has no headings: the book is a single chapter. DocumentPathTests.declaredPath
markdown epub — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. DocumentPathTests.declaredPath
html epub — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. CSS styling and scripts are ignored; remote images are never downloaded. DocumentPathTests.declaredPath
latex epub — ebooks supported structural Pandoc — chapters, images, links, metadata, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. LaTeX is read for structure only: custom macros, most packages, figures drawn in TeX and bibliographies are not interpreted. Math is kept as TeX or MathML where the target allows. DocumentPathTests.declaredPath
rst epub — ebooks supported structural Pandoc — chapters, images, links, tables, textStyles Reflowable book: page layout, fonts, headers, footers and text boxes are dropped. Chapters are split at top-level headings. DocumentPathTests.declaredPath
epub folder epub.chapters ebooks supported structural OPF reader (own code) + Pandoc per chapter — chapters, images, links, tables, textStyles One file per spine item, in reading order (a spine item is not always one chapter). HTML chapters embed their images; Markdown chapters reference images written to a media folder. The book's stylesheets and fonts are not carried. DRM-protected books are refused. EbookTests.chapters
epub folder media.extract ebooks supported lossless ZIP reader (own code) — images Images, audio and video listed in the book's manifest are copied byte for byte. DRM-protected books are refused. EbookTests.coverAndMedia
epub jpeg epub.cover ebooks supported lossy OPF reader (own code), ImageIO when re-encoding — dimensions Works only when the book declares a cover image. The file is copied unchanged when it already is a JPEG; otherwise it is re-encoded. EbookTests.coverAndMedia
epub png epub.cover ebooks supported lossless OPF reader (own code), ImageIO when re-encoding — dimensions Works only when the book declares a cover image. The file is copied unchanged when it already is a PNG; otherwise it is re-encoded. EbookTests.coverAndMedia
pptx pdf — presentations supported structural LibreOffice — dimensions, images, links, metadata, textStyles Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. Animations, transitions, video and audio do not exist in a PDF: each slide is exported in its final state. Speaker notes and hidden slides are not included unless enabled. PresentationTests.declaredPath
pptx png — presentations supported structural LibreOffice → PDF → Core Graphics — dimensions Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. One image file per slide (rendered through an intermediate PDF). Animations, transitions, video, audio, speaker notes and links are not part of an image. PresentationTests.declaredPath
pptx jpeg — presentations supported lossy LibreOffice → PDF → Core Graphics — dimensions Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. One image file per slide (rendered through an intermediate PDF). Animations, transitions, video, audio, speaker notes and links are not part of an image. PresentationTests.declaredPath
pptx ppt — presentations supported structural LibreOffice — dimensions, images, metadata, tables, textStyles Converted by LibreOffice: compatibility with PowerPoint is good but not complete. Animations and transitions are mapped where an equivalent exists, SmartArt becomes plain shapes, embedded video may not play, macros are dropped. Fonts are referenced, not embedded. PresentationTests.declaredPath
pptx odp — presentations supported structural LibreOffice — dimensions, images, metadata, tables, textStyles Converted by LibreOffice: compatibility with PowerPoint is good but not complete. Animations and transitions are mapped where an equivalent exists, SmartArt becomes plain shapes, embedded video may not play, macros are dropped. Fonts are referenced, not embedded. PresentationTests.declaredPath
pptx txt — presentations supported partial Package XML reader (own code) — — Text only, slide by slide in presentation order, with speaker notes: no layout, images, charts or formatting. Text inside pictures, charts and SmartArt is not read. PresentationTests.declaredPath
pptx markdown — presentations supported partial Package XML reader (own code) — tables Text only, slide by slide in presentation order, with speaker notes: no layout, images, charts or formatting. Text inside pictures, charts and SmartArt is not read. PresentationTests.declaredPath
pptx folder media.extract presentations supported lossless ZIP reader (own code) — images Embedded image, audio and video files are copied byte for byte; linked (external) media are not fetched. PresentationTests.declaredPath
ppt pdf — presentations supported structural LibreOffice — dimensions, images, links, metadata, textStyles Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. Animations, transitions, video and audio do not exist in a PDF: each slide is exported in its final state. Speaker notes and hidden slides are not included unless enabled. PresentationTests.declaredPath
ppt png — presentations supported structural LibreOffice → PDF → Core Graphics — dimensions Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. One image file per slide (rendered through an intermediate PDF). Animations, transitions, video, audio, speaker notes and links are not part of an image. PresentationTests.declaredPath
ppt jpeg — presentations supported lossy LibreOffice → PDF → Core Graphics — dimensions Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. One image file per slide (rendered through an intermediate PDF). Animations, transitions, video, audio, speaker notes and links are not part of an image. PresentationTests.declaredPath
ppt pptx — presentations supported structural LibreOffice — dimensions, images, metadata, tables, textStyles Converted by LibreOffice: compatibility with PowerPoint is good but not complete. Animations and transitions are mapped where an equivalent exists, SmartArt becomes plain shapes, embedded video may not play, macros are dropped. Fonts are referenced, not embedded. PresentationTests.declaredPath
ppt odp — presentations supported structural LibreOffice — dimensions, images, metadata, tables, textStyles Converted by LibreOffice: compatibility with PowerPoint is good but not complete. Animations and transitions are mapped where an equivalent exists, SmartArt becomes plain shapes, embedded video may not play, macros are dropped. Fonts are referenced, not embedded. PresentationTests.declaredPath
ppt txt — presentations supported partial LibreOffice → PPTX → package XML reader — — Text only, slide by slide in presentation order, with speaker notes: no layout, images, charts or formatting. Text inside pictures, charts and SmartArt is not read. PPT is first converted to PPTX by LibreOffice, then read. PresentationTests.declaredPath
ppt markdown — presentations supported partial LibreOffice → PPTX → package XML reader — tables Text only, slide by slide in presentation order, with speaker notes: no layout, images, charts or formatting. Text inside pictures, charts and SmartArt is not read. PPT is first converted to PPTX by LibreOffice, then read. PresentationTests.declaredPath
ppt folder media.extract presentations experimental structural LibreOffice → PPTX → ZIP reader — images PPT has no package to read: the file is converted to PPTX by LibreOffice first, so images are the ones LibreOffice re-saved (they can be re-encoded). PresentationTests.declaredPath
odp pdf — presentations supported structural LibreOffice — dimensions, images, links, metadata, textStyles Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. Animations, transitions, video and audio do not exist in a PDF: each slide is exported in its final state. Speaker notes and hidden slides are not included unless enabled. PresentationTests.declaredPath
odp png — presentations supported structural LibreOffice → PDF → Core Graphics — dimensions Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. One image file per slide (rendered through an intermediate PDF). Animations, transitions, video, audio, speaker notes and links are not part of an image. PresentationTests.declaredPath
odp jpeg — presentations supported lossy LibreOffice → PDF → Core Graphics — dimensions Slides are rendered by LibreOffice, not PowerPoint or Keynote: SmartArt, charts, text fitting and theme effects can look different, and missing fonts are substituted. One image file per slide (rendered through an intermediate PDF). Animations, transitions, video, audio, speaker notes and links are not part of an image. PresentationTests.declaredPath
odp pptx — presentations supported structural LibreOffice — dimensions, images, metadata, tables, textStyles Converted by LibreOffice: compatibility with PowerPoint is good but not complete. Animations and transitions are mapped where an equivalent exists, SmartArt becomes plain shapes, embedded video may not play, macros are dropped. Fonts are referenced, not embedded. PresentationTests.declaredPath
odp ppt — presentations supported structural LibreOffice — dimensions, images, metadata, tables, textStyles Converted by LibreOffice: compatibility with PowerPoint is good but not complete. Animations and transitions are mapped where an equivalent exists, SmartArt becomes plain shapes, embedded video may not play, macros are dropped. Fonts are referenced, not embedded. PresentationTests.declaredPath
odp txt — presentations supported partial Package XML reader (own code) — — Text only, slide by slide in presentation order, with speaker notes: no layout, images, charts or formatting. Text inside pictures, charts and SmartArt is not read. PresentationTests.declaredPath
odp markdown — presentations supported partial Package XML reader (own code) — tables Text only, slide by slide in presentation order, with speaker notes: no layout, images, charts or formatting. Text inside pictures, charts and SmartArt is not read. PresentationTests.declaredPath
odp folder media.extract presentations supported lossless ZIP reader (own code) — images Embedded image, audio and video files are copied byte for byte; linked (external) media are not fetched. PresentationTests.declaredPath

Output conventions: one file per conversion, written in the work directory. Slide images are several files named <name>-slide-NN.png|jpg, listed individually in ConversionOutput.files. Folder outputs (media.extract, epub.chapters) are one directory named <name> media / <name> chapters.

# Safety behaviour (all covered by SafetyTests)

  • Macros are never executed. A document with a Basic macro bound to its load event was assembled by hand. Control: the official LibreOffice with macro security off runs it on open (the marker file appears). Through the engine (→ PDF, DOCX, DOC, TXT, HTML, Markdown, RTF; LibreOffice, LibreOffice → Pandoc and Apple routes) the marker never appears; the user is told the macros are dropped; the outputs contain no macro storage. Three independent layers: the profile (MacroSecurityLevel 3, DisableMacrosExecution), headless --convert-to (did not run the macro even with security off), and the slimmed bundle (Basic libraries, Python, Java and scripting providers removed). The control step runs only where the official bundle is present (EmbeddedEngines/libreoffice/work, i.e. after fetch.sh); it was present for this run.
  • No remote resource is ever requested. Two local counting listeners (the "remote host" and the proxy that Pandoc / LibreOffice are pointed at) received zero connections across 29 conversions of an HTML page (remote image, stylesheet, script, font, @import, url(), iframe, video poster, object, fetch()), a Markdown file, a DOCX and an ODT with linked pictures and an EPUB with a remote image, then one each from a control connection. HTML and EPUB content is neutralised before a tool reads them, because Pandoc's HTML reader fetches <iframe> targets while parsing (found by this test).
  • Includes stay inside the document's folder. reStructuredText include / literalinclude / :file: / :url: and LaTeX \input / \include / \import / \lstinputlisting / \verbatiminput are resolved by the engine: files inside the folder are inlined, anything remote, absolute or climbing out of the folder is removed with a warning. HTML → PDF serves only files inside the folder.
  • Protected files are refused, not opened: EPUB with Adobe ADEPT, Apple FairPlay or any content encryption (font obfuscation alone is accepted and reported), password-protected OOXML (OLE2 container) and encrypted ODF.
  • Damaged, truncated, empty and garbage inputs fail with malformedInput / unreadableInput; ZIP entries are inflated with a size cap and checksum; XML is parsed without external entities and with node / depth limits.
  • Eight LibreOffice conversions in parallel succeed (a gate allows max(2, min(6, cores / 2)) processes at once, each with its own profile). Cancellation and the time limit terminate the process; no soffice is left running and the work directory is empty afterwards.

# Measured performance

LibreOffice starts a new process with a new profile for every conversion. Measured cost for a one-page DOCX → PDF: 1.2–2.4 s (table below; 1.2–1.3 s in the quietest measurements of the day). Re-using a profile directory gave no gain (1.7–2.6 s reused vs 1.2–1.9 s fresh, measured back to back under the same load), and a persistent listening instance was not adopted: it would share one process between untrusted documents, could not be killed per job, and needs a UNO client. Eight parallel conversions took 3.9–4.6 s in total.

The table is one run of PerformanceTests.measure, executed alone. The Mac was shared with other heavy jobs all day (load averages: 38,14 81,74 79,92 on 18 cores at the time of this run), so these are not best-case figures; when the whole test suite ran in parallel under a load average above 100, the same single conversions took 4–10 s.

Conversion Time
LibreOffice DOCX→PDF, 1 page, run 1 1.40 s
LibreOffice DOCX→PDF, 1 page, run 2 2.36 s
LibreOffice DOCX→PDF, 1 page, run 3 1.61 s
LibreOffice DOCX→PDF, 1 page, run 4 1.92 s
LibreOffice DOCX→PDF, 1 page, run 5 1.28 s
LibreOffice DOCX→ODT 1.49 s (engine call incl. reading source and output)
LibreOffice DOC→DOCX 1.34 s
LibreOffice PPTX→PDF, 3 slides 1.63 s
PPTX→PNG, 3 slides at 150 dpi 1.51 s (1654 × 930 px each)
PPTX→Markdown text (package XML only) 0.01 s
Pandoc Markdown→DOCX 0.22 s
Pandoc Markdown→HTML 0.21 s
Pandoc DOCX→Markdown 0.06 s
Pandoc DOCX→EPUB 0.08 s
Pandoc EPUB→HTML 0.40 s
LibreOffice + Pandoc DOC→Markdown 1.20 s (two steps)
Pandoc + WebKit Markdown→PDF, 1 page 0.41 s
WebKit HTML→PDF, 1 page 0.14 s
Apple text system RTF→DOCX 0.02 s (in process)
Pandoc Markdown→DOCX, long document 0.33 s (29864 words, 13 KiB DOCX)
LibreOffice DOCX→PDF, long document 2.60 s (51 pages)
Pandoc + WebKit Markdown→PDF, long document 1.43 s (51 pages)
Pandoc DOCX→EPUB, long document 0.46 s

# Evaluated and NOT implemented

Item Decision Why
Pages (.pages) unsupported, no path, no inspection No documented local converter. The "embedded preview" idea was checked on real packages: of 1 026 iWork template packages shipped with the Pages and Keynote apps on this Mac, none contains QuickLook/Preview.pdf, and an iWork '08 document found on disk has only QuickLook/Thumbnail.jpg. With no real file to test on, an "extract embedded preview PDF" operation would be an untested claim.
Keynote (.key) unsupported Same finding.
LaTeX → PDF not declared Needs a TeX distribution, which is not bundled. Rendering LaTeX through HTML would not be LaTeX's typesetting. LaTeX is a structural source / target only (↔ Markdown, HTML, DOCX, ODT, RTF, EPUB, RST, TXT); custom macros, packages, TikZ and bibliographies are not interpreted and a warning says so.
PDF as a source not in this engine Owned by the PDF engine.
Presentation → HTML dropped LibreOffice's export is one absolutely-positioned XHTML page without slide structure and without the speaker notes.
LibreOffice as HTML reader dropped Remote-load attempts and crashes on the hostile test page (see above).
RTFD not declared No catalog format; it is a folder package.
Images written next to Markdown / LaTeX / RST output not implemented Would make the output two items. Images stay as references and a warning is emitted; media.extract gives the files.
Warm / persistent LibreOffice instance not adopted No measured gain from a warm profile; a shared instance weakens isolation.
Removing soffice.cfg from LibreOffice (36 MiB) reverted Every conversion fails or hangs without it.
PPT inspection (slide count, text) not implemented Binary format; read only during conversion through LibreOffice. inspect reports nothing it did not read.
Hyphenation / spelling hyphenation patterns kept, spelling data removed Hyphenation changes line breaks; spelling is not used when converting.

# Fixed on 2026-11-21 (found by the cross-engine sweep)

  • Documents whose only images are remote could not be converted. The remote image was blocked and reported, then the engine's own validation failed with “Images kept: source 1, output 0” (HTML → PDF / DOCX / ODT / RTF, Markdown → PDF / DOCX / EPUB, LaTeX → DOCX, RST → HTML, EPUB → PDF). DocumentFacts now counts the images whose source is a remote URL (remoteImageCount; HTML, Markdown, LaTeX, reStructuredText, EPUB) and both the validation and the loss accounting compare the remaining local / embedded images only. One remote-content-blocked warning (data loss) says how many resources and how many images are left out and why (“… 1 image is not included because it is stored on a remote server (ORIX Convert never fetches remote content)”); preflight says it beforehand; the fact images.remote-not-included records the count. A local image next to a remote one is still required in the output. Test: SafetyTests.documentsWhoseOnlyImagesAreRemoteStillConvert (18 routes, zero connections on the two counting listeners, then a control connection on each).
  • Presentations → text failed when slide text looked like a heading. The plain-text export was validated by counting lines starting with “Slide N”, so body text such as “Slide 1 of 3” was counted (“slides 3, sections 6”). A section header is now the line “Slide N” / “Slide N: title” followed by its rule line, looked for once per exported slide, in order, and the count is also compared with the number of sections the writer recorded (slides fact). The Markdown export never had the flaw: the writer escapes every # of body text, so no body line can start with ## . Test: PresentationTests.bodyTextThatLooksLikeASlideHeading (PPTX and ODP; body text and speaker notes containing “Slide 1 of 3”, “Slide 2: the plan”, “Slide 3” and a 40-dash line; UTF-16 + CRLF; slide subsets; and two damaged outputs that the check must still refuse).
  • media.extract on a file without media now stops with “This document / presentation / book contains no embedded image, audio or video file to extract (0 media files found in the DOCX package) …” and preflight says so before converting (no-media) for documents and books as it already did for presentations. Test: EbookTests.coverAndMedia.

Checked and found correct: the WebKit HTML → PDF renderer (WebPDF.swift) already compiled its content-rule list with one rule per scheme and treated a compilation failure as an error (the SVG renderer of the vector engine did not — see vector.md).

Left as it is: AppKit prints attributedStringScaled … and CoreText note: Client requested name ".SFNS-Regular…" on stderr while WebKit prints a PDF (NSPrintOperation). They come from AppKit's print path, not from this code, and are harmless; nothing in the engine's API silences them.

# Known defects and limits

  • PDF text layer from LibreOffice: with some substituted fonts a ligature cluster is mapped to extra letters (observed: "zebrafissh"). Visual output is correct. Not fixable from outside LibreOffice.
  • Font substitution: fonts missing on the Mac are replaced (the fixture's theme font was replaced by Linux Libertine G). preflight lists the missing fonts by name; it cannot predict the exact replacement except for the metric-compatible pairs (Calibri → Carlito, Cambria → Caladea, Arial / Times New Roman / Courier New → Liberation).
  • → DOC and → PPT cannot be re-read structurally by an independent reader here; the tests read them back through a LibreOffice round trip, which is the same program that wrote them.
  • EPUB chapter split works per spine item; a spine item is not always one chapter.
  • Remote images are recognised in HTML, Markdown, LaTeX, reStructuredText and EPUB sources. Linked (external) pictures of DOCX / ODT are reported as remote references but are not counted as images at all.
  • A local image whose file is missing is still counted as an expected image: a document whose only image is a broken local reference fails the “Images kept” check (the conversion is refused rather than delivered without it).
  • Speaker-notes and title detection depend on placeholder types in the file; text boxes that are not placeholders are exported as ordinary slide text (true of the hand-written test deck).
  • Declared page counts of DOCX / ODT are reported as "declared by the application", never as a computed layout.
  • The compatibility figures above come from one synthetic master document and one three-slide deck. They show what each backend keeps on those constructs; they are not a claim of compatibility with arbitrary Word or PowerPoint files.
Files main
Menu

Theme