id(web): world keyboard encoder drops or silently degrades modified special keys — Alt+Backspace sends plain Backspace, Alt+Arrow is dropped to the browser

Summary

terminalBytesForKey in pkgs/id/web/src/world-keyboard.ts handles modified special keys inconsistently: some are silently dropped, and the rest are silently degraded to the unmodified key — so a world application receives a plain Backspace / Enter / Delete that the user never pressed, while other modified keys vanish entirely.

Evidence (verified on main `5db60876`)

The special-key branch (world-keyboard.ts:36-40):

if (event.key in specialKeys) {
  if ((event.ctrlKey || event.altKey || event.shiftKey) && event.key.startsWith("Arrow")) return null;
  if (event.shiftKey && event.key === "Tab") return null;
  return new TextEncoder().encode(specialKeys[event.key]);
}

Only modified arrows and Shift-Tab are rejected; every other modifier combination falls through to the plain sequence. A temporary vitest probe driving the real function (removed after the run) produced:

Input	Result	Plain-key result
Alt+Backspace	`\x7f`	`\x7f` (identical)
Ctrl+Backspace	`\x7f`	`\x7f` (identical)
Alt+Enter	`\r`	`\r` (identical)
Alt+Tab	`\t`	`\t` (identical)
Alt+Escape	`\x1b`	`\x1b` (identical)
Ctrl+Delete	`\x1b[3~`	`\x1b[3~` (identical)
Alt+Delete	`\x1b[3~`	`\x1b[3~` (identical)
Ctrl+Home	`\x1b[H`	`\x1b[H` (identical)
Shift+End	`\x1b[F`	`\x1b[F` (identical)
Alt+PageUp	`\x1b[5~`	`\x1b[5~` (identical)
Alt+ArrowUp	`null` (dropped)	`\x1b[A`
Alt+ArrowLeft	`null` (dropped)	`\x1b[D`
Shift+Tab	`null` (dropped)	`\t`

Contrast with printable keys in the same function: Alt+x correctly returns \x1bx — the Alt modifier is preserved (as an ESC prefix) for printable input, just not for special keys.

Why it matters

Wrong input delivered: Alt+Backspace is word-delete in most editing contexts; the world app instead receives a single-character Backspace. Ctrl+Delete, Alt+Enter, Shift+End etc. all lose their modifier the same way. The caller in pkgs/id/web/src/world.ts (keydown handler, ~line 240) calls preventDefault() whenever bytes are returned, so the browser's own handling is suppressed and the degraded byte sequence is the only thing that happens.
Dropped input leaks to the browser: when the function returns null, the caller does not preventDefault(). Alt+ArrowLeft / Alt+ArrowRight are browser history navigation in several browsers, so a user trying to send a modified arrow to the world can instead navigate away from the world page mid-session.
The decoder's design says nothing should be dropped or misreported: the receiving side, Key.parse in pkgs/id/examples/roc-world/Key.roc, has no modified-key variants, but it deliberately surfaces unrecognised escape sequences as Unknown(bytes) ("nothing is dropped") — e.g. its own test pins Key.parse("\u(1b)[1;5C") == [Unknown("\u(1b)[1;5C")] for Ctrl+Right. The web encoder is the side that breaks that contract, in two opposite ways at once.

Suggested directions (one consistent policy)

Pick one and apply it to all special keys, not just arrows/Tab:

Encode the modifier using the standard xterm modifier parameter (;2=Shift, ;3=Alt, ;5=Ctrl, …) for CSI sequences (\x1b[3;5~ Ctrl+Delete, \x1b[1;3A Alt+Up, \x1b[Z for Shift-Tab) and the ESC-prefix convention the function already uses for Alt+printable where a CSI form doesn't exist (ESC+\x7f for Alt+Backspace). Key.parse will surface these as Unknown today — preserved, not misreported — and a later Key extension can name them.
Or return null consistently for every modified special key the Roc Key module cannot represent, so no modified press is ever silently rewritten as the plain key — and document that modified specials are unsupported (accepting that the browser keeps e.g. Alt+Left).

Option 1 matches the decoder's nothing-dropped contract; option 2 is the smaller change. Either is better than the current split.

Acceptance criteria

One documented policy covers every (special key × modifier) combination in terminalBytesForKey; no modified special returns the plain key's bytes, and the arrow/Shift-Tab special cases are no longer exceptions to a different rule.
Regression tests in world-keyboard.test.ts pin the policy for at least: Alt+Backspace, Ctrl+Delete, Alt+Enter, Shift+Home/End, Alt+PageUp, Alt+Arrow, Shift+Tab — and the existing plain and Alt+printable expectations still pass.
If option 1 is chosen, the emitted sequences are the standard xterm forms, and the Key.roc behaviour for them (currently Unknown) is noted in the PR.
Unmodified special keys, control letters, and AltGr (Ctrl+Alt printable) behaviour are unchanged.

id train triage (2026-10-24): needs-decision — Muse: assign to Session A (phases 0-10 review); web keyboard encoder drops/degrades modified special keys (world-keyboard.ts:36-40).

Intake review — fleet monitor, 2026-10-26T08:26:24Z.

Still live — verified on current main this run: the special-key branch of terminalBytesForKey in pkgs/id/web/src/world-keyboard.ts is byte-for-byte the code the issue quotes — modified arrows and Shift+Tab return null, every other modified special key falls through to the unmodified sequence. No merged PR or commit has touched this behaviour since filing.

Related: #724 / #761 are the sibling world-input correctness findings in the same web client; this issue already carries user/review, and this intake pass is that review's first step — the label stays until the encoding decision below is taken.

Suggested approach: decide the modified-special-key encoding once, for the whole matrix: Alt+key as ESC-prefixed sequence (consistent with the printable-key path in the same function), Ctrl+special as the xterm modifier-parameter form (CSI 3;5~ etc.) where one exists, and an explicit null + no-preventDefault leak only where the browser genuinely owns the chord. Add the issue's probe table as unit tests in world-keyboard tests so the matrix cannot silently regress.

Risks: silently degraded input (Alt+Backspace → plain Backspace) is worse than dropped input — it performs an edit the user did not ask for; that is why this stays priority/normal rather than low despite the small surface.

Re-verified on current main 95cd41f0 (2026-10-27, new-issue-creator-code-b): verbatim copy of pkgs/id/web/src/world-keyboard.ts (sha256 cd62a6a8…, identical both sides) driven under Bun on the laptop — Alt+Enter \x0d, Alt+Backspace \x7f, Alt+Tab \x09, Alt+Escape \x1b, Alt+Delete \x1b[3~, Alt+Home \x1b[H are all byte-identical to the unmodified keys, Alt+ArrowUp still returns null, and Alt+x still correctly returns \x1bx. The inconsistency is unchanged; no separate issue filed.

🤖 Agent note · parked (resume check)

Agent: queue worker (repo-worker-code), Lane F shift
Session / ID: queue task 4622
Task: PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184 (head 64e63c0, only world-keyboard.ts + world-keyboard.test.ts) revalidated fresh: Codex review completed, no findings. CI run 41231798279 still running — test-web-unit, test-unit, test-shard-0 and the full tier pending; test-web-typecheck, web-lint, test-shards 1–3, cargo-check already green.
Failures dispositioned: shfmt-check proven pre-existing on main (ci(id): shfmt-check fails on main — pre-existing, unrelated to PR diffs touching it #3208); web-fmt-check annotations show GitHub cache restore 429 rate-limit (infra); rustfmt/clippy cannot be caused by a TS-only diff. None block this PR.
Expectation: resume ~21:20Z — remaining jobs green → merge PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184, close this issue, complete task 4622 as fixed_merged.
Links: PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184 · run https://github.com/developing-today/code/actions/runs/41231798279

🤖 Agent note · picked up

Agent: queue worker (repo-worker-code), Lane D
Session / ID: queue task 4622, Lane D shift 2026-10-31 16:14 CDT (resume)
Found via: queue claim
Task: Resume PR 3184 (xterm modifiers for special keys, Option 1) — re-verify CI run 41231798279, disposition Codex, merge when green
Expectation: effort s; done = PR 3184 merged and id(web): world keyboard encoder drops or silently degrades modified special keys — Alt+Backspace sends plain Backspace, Alt+Arrow is dropped to the browser #732 closed, or parked again on CI with fresh evidence
Links: PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184 · branch queue/4622-732-world-keyboard-modifiers

🤖 Agent note · picked up (resume → merge)

Agent: queue worker (repo-worker-code), Lane F shift
Session / ID: queue task 4622
Task: Resume PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184 @64e63c01d (xterm modifiers for special keys, Option 1) — CI run 41231798279 has now COMPLETED. Every failing job in it also fails on clean main run 41238667850 @36a3347f (test-unit, test-web-unit, biome, prettier, web-fmt, shfmt, clippy, rustfmt, cargo-fmt, treefmt): test-unit's 5 failures are Rust-only (store_traits/world_ws/sandbox fuel), test-web-unit's single failure is search-panel.test.ts:269 (495 passed incl. all world-keyboard tests), biome's 4 errors are in goto-line/search-panel/world-surface files — none in this PR's two files. PR-relevant jobs green: test-web-typecheck, web-lint, test-shards 0–3. Codex review on head sha: no findings. Lane E substantive review stands. Merging under the standing merge directive.
Expectation: PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184 merged → this issue closes → task 4622 completed as fixed_merged.
Links: PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184 · PR run https://github.com/developing-today/code/actions/runs/41231798279 · main run https://github.com/developing-today/code/actions/runs/41238667850

fix(id/web): encode xterm modifiers for special keys in world keyboar…

🤖 Agent note · done

Agent: queue worker (repo-worker-code), Lane F shift
Session / ID: queue task 4622
Task: PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184 MERGED (squash, merge sha c0fb853) — xterm modifier policy for special keys (Option 1) is on main; this issue closed by the merge. CI run 41231798279's failures were all proven pre-existing on clean main (run 41238667850); PR-relevant checks green; Codex clean. The pre-existing shfmt failure already has its own issue (ci(id): shfmt-check fails on main — pre-existing, unrelated to PR diffs touching it #3208); the remaining main-branch failures (test-unit Rust tests, search-panel aria-label, biome/prettier format drift) are fleet backlog material, not part of this fix.
Links: PR fix(id/web): encode xterm modifiers for special keys in world keyboard encoder (#732) #3184

Navigation Menu

Repository navigation

id(web): world keyboard encoder drops or silently degrades modified special keys — Alt+Backspace sends plain Backspace, Alt+Arrow is dropped to the browser #732

Summary

Evidence (verified on main `5db60876`)

Why it matters

Suggested directions (one consistent policy)

Acceptance criteria

Intake review — fleet monitor, 2026-10-26T08:26:24Z.

53 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

id(web): world keyboard encoder drops or silently degrades modified special keys — Alt+Backspace sends plain Backspace, Alt+Arrow is dropped to the browser #732

Description

Summary

Evidence (verified on main 5db60876)

Why it matters

Suggested directions (one consistent policy)

Acceptance criteria

Activity

dezren39 commented on Oct 24, 2026

dezren39 commented on Oct 26, 2026

Intake review — fleet monitor, 2026-10-26T08:26:24Z.

dezren39 commented on Oct 27, 2026

53 remaining items

dezren39 commented on Oct 31, 2026

dezren39 commented on Oct 31, 2026

dezren39 commented on Oct 31, 2026

dezren39 commented on Oct 31, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Evidence (verified on main `5db60876`)