v0.3.1

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

• D2-1: createDialogs().background edits the last message with no update

• D2-5: over-long callback_data throws a named error into onError

• D2-2: concurrent taps on one stack don't lost-update (withLock + reload)

three fixes that previously had no e2e coverage:

test: cover D2-1/D2-2/D2-5 end-to-end via @gramio/testAdds real-bot integration tests through TelegramTestEnvironment for the

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

Strictly more permissive — not a breaking change. Tests: datadict-interface.

• export AnyData / RenderData

  blocked Window<interface> ⊆ Window<DataDict>); isHidden made generic

• widen Dialog window storage to Window<AnyData> (Getter's covariant return

  interface, and untyped windows alike

  DataDictcollapses toDataDict, so d.dialogData` keeps working for typed,

  render callback: the engine injects dialogData/startData, and `AnyData &

• new RenderData<Data> = Data & DataDict types RenderContext.data and every

• DataDict stays the concrete, indexable bag the engine stores/indexes

  spelled out for intent); ~30 user-facing extends DataDict → extends AnyData

• new AnyData = Record<PropertyKey, unknown> | object constraint (≡ object,

failed extends DataDict (interfaces have no implicit index signature).

feat: accept interface-typed window Data (not just type)A typed Window<Data> / getter now accepts a TS interface, which previously

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

messages. No behavior change.

compiler; the bare alias reads better in hovers and constraint-violation

refactor: simplify AnyData to object (same type, cleaner tooltips/errors)Record<PropertyKey, unknown> | object is identical to object for the

chore(release): v0.3.0Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

check` (CI) failed on formatting only. No logic change.

style: biome-format plugin.test.ts (unbreak CI lint)The D2 e2e tests were committed without running bun run lint; `bun run

data-derived, undefined return, and Url combo).

Add test coverage for function-style resolution (static → ✓,

LinkWidget (Url/WebApp), and SwitchInlineWidget render paths.

Add resolveStyle() helper, wire it through ButtonWidget,

WhenCondition and TextSource for render-data-derived values.

in addition to a static value, matching the pattern used by

feat: support dynamic style function on Button (like TextSource)Make ButtonChrome.style accept a function (rc) => ButtonStyle

chore: v0.3.1

fix: allow undefined return in style function, update CLAUDE.md with full check
Full Changelog: v0.2.1...v0.3.1

v0.2.1

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

Version intentionally not bumped. Tests: feedback-d2; routing/robustness updated.

• D2-6: document that Select.itemId must be collision-free as a string.

  build time, instead of warning then letting Telegram reject the send into onError.

• D2-5: over-long callback_data now throws a clear error naming the widget at

  instead of leaving a no-op stub on the shared gramio context.

• D2-4: scope the reply-keyboard ctx.answer shim — restore it after routing

  (was mutated in memory but dropped, only relevant with StartMode.NewStack).

• D2-3: persist the currentId reselection on the stale/denied return paths

  headlessCtx also fleshes from.is_bot.

  from with a locale, or t) so getters/i18n needing more than from.id work;

• D2-1: background(bot, key, overrides?) — enrich the headless ctx (e.g. real

  createDialogs().withLock so background() callers can serialize against taps.

  (_reloadStore), so concurrent updates can't clobber each other. Exposed as

  update path now locks per stack key and re-reads the store under the lock

fix: address v0.2.0 feedback round 2 (background/persist/routing hardening)- D2-2: lost-update race — StackRepository.withLock keyed async mutex; the live

chore(release): v0.2.1Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
Full Changelog: v0.2.0...v0.2.1

v0.2.0

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

Tests: tests/feedback-fixes.test.ts (6 new; full bun run check green)

DataDict-needs-type-not-interface constraint, and the new APIs

• docs: getter↔dialogData flow, headless getter contract, the

  backed by external state

• feat: Select({ selected, selectedMark }) — auto-✓ for stateless selects

  rendering the target window (back-compat with the bare ShowMode arg)

• feat: switchTo(state, { data, mode }) — merge into dialogData before

• feat: DialogManager.setData(patch) — merge dialog data + persist, no render

  ctx.from.id no longer crash when re-rendered from background()

• fix: headlessCtx synthesizes from: { id: chatId } so getters reading

feat: harden background() renders and add data-navigation APIsAddresses production feedback on the background()/headless seam:

chore(release): v0.2.0Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

Tests: answer-binding (stale + handled taps with a this-dependent answer).

and route the success-path spinner close through the same helper.

Now call it bound on ctx inside try/catch (catches sync throw + rejection),

bot's onError (surfaced as a prod crash on stale/denied taps and timeouts).

and .catch() only catches rejections, so the throw escaped past it into the

this.bot.api.answerCallbackQuery, so the detached call threw SYNCHRONOUSLY —

undefined. gramio's answer is a prototype alias delegating to

fix: bind answer in silentAnswer so detached call can't crash the botsilentAnswer detached the method (const a = ctx.answer; a()), so this was
Full Changelog: v0.1.0...v0.2.0

v0.1.0

feat: update readme.md and .gitignore

fix: ci badge url

fix: biome format

fix: typed text param for strict gramio + quiet changelog tag lookup
Full Changelog: https://github.com/gramiojs/dialogs/commits/v0.1.0