Paste pipeline #

Clipboard HTML → Leaflet blocks.

clipboard text/html ─┐
                     ├─► normalizePastedHTML ─► flattenHTMLToTextBlocks ─► buildBlockFromHTML ─► facts
clipboard text/plain ┘        (dialects)            (block elements)          (one per block)
   └─ markdownToHtml

• normalizePastedHTML.ts rewrites word-processor dialects into ordinary HTML: Word's mso-list paragraphs, Word Online's aria-level list runs, Google Docs' sibling-nested <ul>, class-driven marks that live in a <style> block, tables, task-list checkboxes.
• htmlToBlocks.ts flattens the normalized tree into block-level elements and turns each one into the facts a block needs.

Everything downstream of parsePasteHTMLToElements is pure, which is what the tests drive. useHandlePaste only owns the Replicache/undo/cursor side.

Tests #

• htmlToBlocks.test.ts — one block outline per source (real captures in __fixtures__/captured/, plus hand-written fixtures for sources we have no capture of), then only what an outline can't show: marks, links, image URLs, Leaflet's own block attributes. Structural edge cases and blockquotes are tables of html → outline.
• markdownPaste.test.ts — the text/plain path.
• fixtureInvariants.test.ts — source-agnostic invariants over every fixture.

What Google Docs actually emits #

Worth writing down, because the widely-repeated version is out of date. Docs is not currently wrapping its payload in <b style="font-weight:normal" id="docs-internal-guid-…">; both captured selections — whole-document and mid-paragraph — put that id on the first block element and emit no wrapper. The failure that does bite is list nesting: sublists arrive as siblings of the <li> they belong to (<ul><li>a</li><ul><li>b</li></ul></ul>), which nestSiblingLists repairs.

legacy-docs-inline-wrapper.html still exercises the wrapper shape, since an inline element containing block content is real regardless of Docs (Docs itself does it around images), but it isn't evidence about Docs today.

Adding a real capture #

Captures beat hand-written fixtures — the hand-written ones encode what we believe an app emits, and that belief has already been wrong once. To take one:

1. Open capture-clipboard.html in a browser.
2. Copy from the app you care about, then paste into the box.
3. Save the captured markup as __fixtures__/captured/<source>.html.

fixtureInvariants.test.ts picks up anything in __fixtures__ automatically and checks the invariants that hold for every source. Add an outline for the new capture in htmlToBlocks.test.ts too — that is where a regression actually shows up.