feat(rust): markdown shortcuts in both desktop playgrounds by tmm · Pull Request #161 · wevm/lexical

tmm · 2026-11-14T18:06:34Z

Description

Typing Markdown in the Rust playgrounds left it as text: the playground
editor never registered the markdown shortcuts, and Return called
insertParagraph directly, so the Enter-triggered transformers (code
fences) could not run.

lexical-playground-core registers register_markdown_shortcuts with
the default transformers. Its script flags become
["richText","list","link","code","markdown"], so an exported session
replays in JS with the same shortcuts. Return now dispatches
KEY_ENTER_COMMAND, as a browser keydown does; rich text's handler
then dispatches INSERT_PARAGRAPH_COMMAND unless the shortcuts or the
code block took it.
GPUI draws code blocks with a monospace font, padding and a tinted
background, so a typed code fence is visible. Its e2e run types a
heading, bold text, a list item and a code fence.
Both READMEs list the shortcuts.
This branch merges rust-playground-blocks (feat(rust): headings, quotes, lists, links and indent in both playgrounds #141), which the
playgrounds' block rendering needs.

Test plan

Before

$ cd rust/apps/lexical-playground-core && cargo test --locked markdown
thread 'tests::markdown_shortcuts_turn_typed_markdown_into_rich_text' panicked at apps/lexical-playground-core/src/tests.rs:497:5:
assertion `left == right` failed
  left: "paragraph"
 right: "h1"
test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 19 filtered out

After

$ cd rust/apps/lexical-playground-core && cargo test --locked
test result: ok. 20 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
$ cd rust/apps/lexical-playground-rune && cargo test --locked   # macOS
test result: ok. 10 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
$ cd rust/apps/lexical-playground-gpui && cargo build --locked --features e2e && \
    LEXICAL_PLAYGROUND_E2E=/tmp/s.json ./target/debug/lexical-playground-gpui   # macOS
e2e: passed
$ cd packages/lexical-conformance && LEXICAL_REPLAY=/tmp/s.json pnpm exec vitest run src/replay.test.ts
 Test Files  1 passed (1)

On macOS, real keystrokes typed into GPUI through System Events (a
heading, lists, a quote, inline formats, a link, a code fence, then ⌘Z)
produced a 163-step session that replays in JS, and the in-app Replay
vectors shows 655/655. Real keystrokes could not be delivered to the Rune
app on the Mac satellite (it does not redraw for synthetic events, also
without this change), so Rune is covered by its runtime tests only.

Stacked on #157 (markdown shortcuts) and merges #141 (playground blocks), so it carries both until they land.

…unds ## Description The Rune and GPUI playgrounds drew every block as a plain paragraph, read only paragraphs at the root, and had no way to make a heading, quote, list or link, though lexical-editor now has rich-text, list and link. `lexical-playground-core` now: - reads the tree recursively: list items with their list's type, depth, number and `checked`, headings, quotes, indents, and links' text runs with their URL; selection points map through undrawn elements (the root, lists), and the tree view shows lists above their items; - registers rich-text, list, check list and link in the session, and sends Enter, Shift-Enter and formatting as the commands a browser dispatches; - adds `set_block_type` (lists through their commands, the same list again makes paragraphs; other types through `$setBlocksType`), `indent`, `tab` (`$indentOverTab`), `toggle_link`, `toggle_checked`, `block_type_at`, `link_at` and `format_at`; - gives the selection the format a browser's `selectionchange` gives it (`onSelectionChange`: a caret takes its text node's format and style, a range the format all its text shares) in the select step before an edit, and the toolbar reads it from the selection the app shows, so a click into bold text shows bold and typing there stays bold, as in a browser; - gets an example with a heading, a bullet list with a link, a check list and a quote. Rune maps list items to its field's bullets and numbers at their depth, a quote to its quote block, links to `Mark::Link`, headings to larger, bolder decorations and a checked item to strikethrough. GPUI shapes each line on its own, at its block's size and indent, and draws list markers (•, 1., ☐/☑), a quote's bar, links in the JS playground's blue, and checked items struck through; a click on a box toggles it. Both get a second toolbar row (Normal, H1-H3, Quote, the three lists, Link, Outdent, Indent, and a check toggle in a check item), and Tab and Shift-Tab indent and outdent. The GPUI end-to-end run now also makes a heading and a list, nests the item with Tab, unnests it with Shift-Tab and ends the list; CI replays that session in JS. ## Test plan ### Before The block-menu test in `lexical-playground-core`, before `$setBlocksType`: ``` test tests::the_block_menu_converts_blocks_and_toggles_lists ... FAILED ``` Without the selectionchange format in the select step: ``` thread 'tests::a_click_gives_the_selection_the_format_there_as_selectionchange_does' panicked left: ("Ru", true) right: ("Ruyst", true) ``` ### After ``` $ cargo test --workspace --all-features --locked (rust/) test result: ok. 19 passed; 0 failed (lexical-playground-core) $ cargo test --locked (apps/lexical-playground-rune) test result: ok. 9 passed; 0 failed $ cargo build --locked --features e2e && LEXICAL_PLAYGROUND_E2E=/tmp/gpui-e2e.json ./target/debug/lexical-playground-gpui (macOS 27, a real window) e2e: passed $ LEXICAL_REPLAY=/tmp/gpui-e2e.json pnpm exec vitest run src/replay.test.ts Tests 1 passed (1) ``` The GPUI session has 26 steps, including INSERT_UNORDERED_LIST_COMMAND, INDENT_CONTENT_COMMAND and OUTDENT_CONTENT_COMMAND. A 26-step session from `lexical-playground-core` (block menu, lists, Tab, links, a check box, undo and redo) also replays in JS, and changing one character of one recorded result makes the replay fail. fmt, clippy `-D warnings` and rustdoc `-D warnings` pass for the workspace; clippy passes for Rune and, on macOS, for GPUI with `--features e2e`. Screenshots of both apps were inspected: Rune headless (`--snap`), GPUI on macOS 27. Amp-Thread-ID: https://ampcode.com/threads/T-16d45229-bfa8-417a-a091-d3e3888d830d

Amp-Thread-ID: https://ampcode.com/threads/T-01a1daf1-150e-709b-aaed-996443a64459

## Description Typing Markdown in the Rust playgrounds left it as text: the playground editor never registered the markdown shortcuts, and Return called `insertParagraph` directly, so the Enter-triggered transformers (code fences) could not run. - `lexical-playground-core` registers `register_markdown_shortcuts` with the default transformers. Its script flags become `["richText","list","link","code","markdown"]`, so an exported session replays in JS with the same shortcuts. Return now dispatches `KEY_ENTER_COMMAND`, as a browser keydown does; rich text's handler then dispatches `INSERT_PARAGRAPH_COMMAND` unless the shortcuts or the code block took it. - GPUI draws code blocks with a monospace font, padding and a tinted background, so a typed code fence is visible. Its e2e run types a heading, bold text, a list item and a code fence. - Both READMEs list the shortcuts. - This branch merges `rust-playground-blocks` (#141), which the playgrounds' block rendering needs. ## Test plan ### Before ``` $ cd rust/apps/lexical-playground-core && cargo test --locked markdown thread 'tests::markdown_shortcuts_turn_typed_markdown_into_rich_text' panicked at apps/lexical-playground-core/src/tests.rs:497:5: assertion `left == right` failed left: "paragraph" right: "h1" test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 19 filtered out ``` ### After ``` $ cd rust/apps/lexical-playground-core && cargo test --locked test result: ok. 20 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out $ cd rust/apps/lexical-playground-rune && cargo test --locked # macOS test result: ok. 10 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out $ cd rust/apps/lexical-playground-gpui && cargo build --locked --features e2e && \ LEXICAL_PLAYGROUND_E2E=/tmp/s.json ./target/debug/lexical-playground-gpui # macOS e2e: passed $ cd packages/lexical-conformance && LEXICAL_REPLAY=/tmp/s.json pnpm exec vitest run src/replay.test.ts Test Files 1 passed (1) ``` On macOS, real keystrokes typed into GPUI through System Events (a heading, lists, a quote, inline formats, a link, a code fence, then ⌘Z) produced a 163-step session that replays in JS, and the in-app Replay vectors shows 655/655. Real keystrokes could not be delivered to the Rune app on the Mac satellite (it does not redraw for synthetic events, also without this change), so Rune is covered by its runtime tests only. Amp-Thread-ID: https://ampcode.com/threads/T-01a1daf1-150e-709b-aaed-996443a64459

ampagent added 3 commits November 14, 2026 06:46

chore(rust): merge rust-playground-blocks (#141) into the markdown stack

d9668e3

Amp-Thread-ID: https://ampcode.com/threads/T-01a1daf1-150e-709b-aaed-996443a64459

tmm force-pushed the rust-playground-markdown branch from 1730dea to 7bfbfca Compare November 14, 2026 18:15

Navigation Menu

Repository navigation

feat(rust): markdown shortcuts in both desktop playgrounds - #161

feat(rust): markdown shortcuts in both desktop playgrounds#161
tmm wants to merge 3 commits into
rust-markdown-shortcutsfrom
rust-playground-markdown

tmm commented Nov 14, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

tmm commented Nov 14, 2026

Description

Test plan

Before

After

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants