Support reading tables in Google Docs

[ndisidore]

Previously in Google Docs docTabToMarkdown() previously processed only paragraph structural elements. This updates it to include tables as well

Google documentation

The Google Docs API explicitly exposes table cells through documents.get():

• Document structure overview: explains that documents.get() returns the complete document as structured JSON.
• StructuralElement: body elements can be paragraphs, section breaks, tables, or tables of contents.
• Table: contains tableRows[].
• TableRow: contains tableCells[].
• TableCell: contains recursive content[] structural elements, including the cell’s paragraphs and nested tables.
• documents.get: retrieves that document structure.

i.e.

Document
└─ Tab.documentTab.body.content[]
   └─ StructuralElement.table
      └─ tableRows[]
         └─ tableCells[]
            └─ content[]
               └─ StructuralElement.paragraph
                  └─ elements[].textRun.content

---

[github-actions]

Preview: pr518-fix-google-do-2cddb46f

https://pr518-fix-google-do-2cddb46f-router.cloudflare-os-previews.workers.dev

Dashboard · deleted when this PR closes

[ask-bonk]

Submitted 1 actionable inline finding.

github run

[ask-bonk]

Submitted 1 actionable inline finding.

github run

[ask-bonk]

Submitted 2 actionable inline findings.

github run

[ask-bonk]

Submitted 2 actionable inline findings.

github run

[ask-bonk]

Submitted 2 actionable inline findings.

github run

[ask-bonk]

Submitted 3 actionable inline findings.

github run

[ask-bonk]

Submitted 1 actionable inline finding.

github run

[ask-bonk]

Submitted 1 actionable inline finding.

github run

[ask-bonk]

Submitted 2 actionable inline findings.

github run

[ask-bonk]

Submitted 2 actionable inline findings.

github run

[ask-bonk]

• Medium packages/gatekeeper-google/src/markdown-converter.ts:1475: Replacing an interrupted list item 2. Second with 3. Second previews 3, but bullet recreation discards the requested ordinal and Google commits 1. Second. Normalize unsupported restarts or reject them before approval.

github run

[ask-bonk]

Medium: The documented glyphType: "NONE" renders an empty glyph, but this classifies every present glyph type as a numbered list. A valid cell whose paragraph has bullet: { listId: "L" } and whose lists.L.listProperties.nestingLevels[0] is { glyphType: "NONE", glyphFormat: "%0" } is displayed by Docs with no marker; this renderer emits <ol><li>Text</li></ol>, inventing 1. (and the body path likewise invents 1. ). Handle NONE as a markerless level rather than ordered.

[ask-bonk]

Medium: This boundary check protects only positions inside a delimiter run, not inside the inline style span that the delimiters form. For a provider paragraph containing plain malformed-but-supported text *x, replaceText("*x", "*x*") previews a complete italic span, and parsing the whole replacement would insert x with italic: true. Here the shared *x is retained, however, so materialization passes only the final * to markdownToDocRequests; the generated batch merely inserts that literal character with an empty text style, leaving all three characters plain. The same occurs when completing ~~x to ~~x~~. Expand trim bounds to the complete resulting inline formatting span (as is already done for links), and cover completion of an unmatched delimiter.

[ask-bonk]

• Medium markdown-converter.ts:1579: Changing a bullet item to numbered treats both as the same list boundary. Simulation omits the blank separator that appears after materialization, invalidating dependent edits.
• Medium markdown-converter.ts:548: Valid glyphType: "NONE" lists render as <ol>, inventing numbering for markerless table-cell lists.

github run

[ask-bonk]

Submitted 1 actionable inline finding.

github run

[ask-bonk]

• Medium markdown-converter.ts:957: Token-boundary trimming is quadratic. A 280 KB formatted replacement took ~3.8s; larger valid Docs can exhaust Worker CPU. Index ranges instead of rescanning all ranges per character.
• Medium markdown-converter.ts:1511: Style ranges include characters Google strips from insertText. Inputs containing controls/private-use characters can style neighboring text or invalidate the batch. Sanitize before calculating offsets.
• Medium markdown-converter.ts:462: Table cells emit Google’s U+000B manual line breaks as raw control characters, so HTML consumers lose the visible break. Render them as <br>.

github run

[ask-bonk]

Medium: Canonicalizing only the trimmed fragment misses a newly created plain-paragraph boundary. For an existing A\n\nB, replaceText("A\n\nB", "A\nX\nB") trims the change to just X, so simulation remains A\nX\nB. Materialization recognizes the insertion point as following a paragraph and inserts \nX, creating three provider paragraphs; the next read renders them as A\n\nX\n\nB. Any dependent queued edit against the approved preview is then invalidated. Please canonicalize/rebuild with the adjacent block boundaries and add an action-level replay case.

[ask-bonk]

Submitted 1 actionable inline finding.

github run