benwbrum / fromthepage / 36489529837

75%

push

github

Make the accessible PDF export usable by screen readers and Acrobat's checker (#5861)

* Fix Acrobat accessibility checker failures in the accessible PDF export

The Grover/Chrome "accessible PDF" export (accessible_pdf.pdf) failed three
Adobe Acrobat accessibility checks:

- Bookmarks: large documents must have bookmarks. Chrome only emits a PDF
  outline when asked, so pass `outline: true` (with `tagged: true`) to the
  PDF generator. Bookmarks are built from the h1/h2/h3 heading structure the
  export already produces (work title, per-page titles, notes, etc.).

- Tagged content: every page carried one untagged text run - the CSS
  `@page { @bottom-center { content: counter(page) } }` page number. Chrome
  renders margin-box generated content without a structure-tree tag and
  without an Artifact marker, so Acrobat flags it. Remove the page-number
  footer from the export layout.

- Table regularity: transcribers can build TEI tables whose rows have
  different cell counts. Add Work::Export::TableRegularizer, which pads short
  rows with non-breaking-space cells so every row has the widest row's column
  count before the HTML is handed to Chrome. Rowspan tables are left alone.

Verified end to end against the Grover pipeline: resulting PDF has a document
outline, zero untagged text operators on every page, and regular table
structure elements.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MMBG8j4PEQR84hBdE32dSz

* Flatten links and line breaks in the accessible PDF for screen readers

Windows Narrator could not read transcription body text in the accessible
PDF: it announced only the hyperlink URLs and skipped everything else.

Chrome/Skia tags every inline break in the text flow as its own NonStruct
marked-content run. Transcription HTML is dense with inline elements - one
<span class="line-break"> per source line, one <a> per subject link - so each
paragraph shatters into 10-15+ fragment... (continued)

2899 of 4467 branches covered (64.9%)

Branch coverage included in aggregate %.

83 of 84 new or added lines in 3 files covered. (98.81%)

11462 of 14737 relevant lines covered (77.78%)

210.5 hits per line