ContextQMD
Back to Plate

Lexical tab node harvest rows need clipboard and browser proof boundaries

docs/solutions/best-practices/2026-05-09-lexical-tab-node-harvest-rows-need-clipboard-browser-proof-boundaries.md

53.0.63.2 KB
Original Source

Lexical tab node harvest rows need clipboard and browser proof boundaries

Problem

Lexical TabNode tests look like one portable tab feature, but they cover several different owners: plain-text clipboard fallback, Google Docs HTML paste, keyboard indentation, command-level tab insertion, node serialization, and selection around a TabNode class. Slate only owns the literal tab/newline paste behavior in this slice.

Symptoms

  • Plain-text tab/newline paste needed package proof in the DOM clipboard owner.
  • Google Docs Apple-tab-span paste needed browser proof in the paste-html example.
  • A trial switch from tx.nodes.insert to tx.fragment.insert removed an empty paragraph for the tab fixture but flattened existing nested-list paste rows.
  • A browser assertion that expected exactly two paragraphs failed because one empty paragraph could remain around the visible pasted content.

What Didn't Work

  • Treating Lexical's TabNode class as a Slate model target. Slate stores literal tabs as normal text.
  • Broadly changing the paste-html importer insertion API to satisfy one paragraph-count assertion. That traded a narrow tab proof for regressions in the existing list corpus.
  • Asserting all paragraph wrappers after paste. That tests incidental wrapper count instead of the visible tab/newline invariant.

Solution

Split the source rows by owner:

  • add a slate-dom clipboard package row for text/plain hello\tworld\nhello\tworld;
  • add a paste-html browser row for the Google Docs Apple-tab-span HTML;
  • assert the two non-empty paragraph text values exactly as Hello\tworld, not the total paragraph count;
  • reject Lexical TabNode schema/serialization/type-guard rows;
  • defer keyboard indentation and adjacent literal-tab selection replacement to explicit code/list/input or text-boundary owners.

Why This Works

The copied behavior is not "Slate should have a TabNode." The useful invariant is that editor paste paths preserve literal tab characters while honoring line boundaries. Package clipboard proof covers model fallback; Playwright covers the browser HTML parser path.

Keeping the paste-html importer on its existing insertion path also preserves the already-proved nested-list corpus. A cleaner insertion API might be possible later, but it needs the full paste-html browser file as the first proof, not one new tab row.

Prevention

  • For upstream node tests, split clipboard/browser behavior from node serialization and command API before editing Slate.
  • When touching paste-html insertion, run the whole paste-html browser file, not only the new row.
  • In paste browser rows, assert visible imported content and exact text where possible; avoid paragraph-count assertions unless wrapper count is the real contract.
  • Treat tx.fragment.insert as paste-like policy only after checking existing list/table corpus behavior.