djust 0.9.2rc2

dj-transition now accepts 1-token short form, matching documented grammar. Closes #1273. The dj-transition-group docs at 43-dj-transition-group.js:22-23 advertise short form like dj-transition-group="fade-in | fade-out" where each half can be a 1-token spec — but _parseSpec at 41-dj-transition.js:45-60 required 3 tokens, so the ENTER side was silently rejected and no animation fired. Fix: extend _parseSpec to accept 1-token form, return {single: <class>}; _runTransition handles single by applying the class on next frame and waiting for transitionend. 2-token form remains rejected as ambiguous (matches dj-remove). 2 new regression cases in TestParseSpecAcceptsShortForm + behavioral test (tests/js/dj_transition.test.js).

AsyncResult now serializes to a dict templates can navigate. Closes #1274. assign_async() returns AsyncResult instances that templates expect to read as {{ users.loading }}, {{ users.ok }}, {{ users.failed }}, {{ users.result }}, {{ users.error }}. Before this fix, neither normalize_django_value nor DjangoJSONEncoder.default had an AsyncResult branch — the value fell through to str(), producing "AsyncResult(loading=True, ...)" which templates couldn't navigate. Result: every assign_async demo rendered blank. Fix: new AsyncResult.to_dict() method + register in both serializer paths; normalize_django_value recurses into the dict so a non-primitive result payload (Django Model, datetime, Decimal nested in result) is normalized too. 11 regression cases in python/djust/tests/test_async_result_serializer.py.

Form submit now flushes pending debounced dj-input handlers before dispatching. Closes #1278. Text/email/password inputs with dj-input defaulted to 300ms debounce; a user who typed and immediately clicked submit raced the submit handler past the pending input events. Views that depended on server-side state populated by dj-input handlers (e.g., WizardMixin's wizard_step_data) saw stale state at submit time. Fix: debounce() now exposes a .flush() method; new _flushPendingDebouncesInForm(form) iterates the form's [dj-input] descendants and flushes any pending wrappers; _handleDjSubmit calls it before reading FormData / dispatching. 4 regression cases in tests/js/dj_submit_debounce_flush.test.js.

dj-dialog client-close (ESC/backdrop/dialog.close()) now syncs back to server. Closes #1267. Previously dj-dialog was one-way (server→client); the user closing a dialog client-side left server state believing it was still open. Re-opening from the server became a no-op because re-asserting dj-dialog="open" wasn't a value change. Fix: new dj-dialog-close-event="..." attribute opts into a native close event listener that dispatches the configured event name to the server. Idempotent across re-syncs (WeakMap guard); reads attribute at fire time so morph updates take effect. 5 new regression cases in tests/js/dj_dialog.test.js.

SSE transport: EventSource and dispatch POST now send Django session cookie. Closes #1277. Authenticated views over SSE failed check_view_auth on every mount because the EventSource GET (and the message POST) didn't carry credentials. Result: infinite mount→navigate loop on authenticated views. Fix: 03b-sse.js:69 opens EventSource with {withCredentials: true}; 03b-sse.js:367sendMessage POST sets credentials: 'include'. 2 new regression cases in tests/js/sse-transport.test.js.

mount() lifecycle: queued async work and push events are now drained after the mount frame. Closes #1280 (assign_async() / start_async() called from mount() never resolved over WebSocket — view stayed at initial loading-state HTML forever) and #1283 (push_event() called from mount() or on_mount hooks queued events that never reached the client). Both root at the same site: LiveViewConsumer.handle_mount() ended with send_json(response) without draining _async_tasks or _pending_push_events. The fix mirrors the established pattern in handle_event() / _flush_deferred_activity_events(): send the response frame, then drain push events, then dispatch async work. 3 regression cases in TestHandleMountSourceShape (python/djust/tests/test_handle_mount_drains_queues.py).

data_table integration restored over WebSocket — emit defaults renamed to match on_table_* mixin handler convention. Closes #1275 (tag emitted 23 event names that didn't match any handler), #1291 (pagination handlers entirely missing from the mixin), #1279 (handlers mutate state but never refresh rows). Single root cause: the WS dispatcher does exact-match getattr(view, event_name, None) (websocket_utils.py:173), but the tag-emit defaults previously used bare table_* strings while DataTableMixin uses on_* Phoenix-style handler names — so every default WS interaction returned "no handler found". Fix: rename tag-emit defaults across 4 files (92 lines: templatetags/djust_components.py, mixins/data_table.py class-level attrs + _PRE_MOUNT_TABLE_CONTEXT, components/rust_handlers.py, templatetags/_forms.py); add on_table_prev / on_table_next handlers (clamped to [1, table_total_pages]); call refresh_table() from sort/search/filter/page/prev/next handlers (selection handler deliberately exempt — UI state). 15 regression cases in TestDataTableEmitToHandlerCrossReference, TestPaginationHandlersExist, TestRowAffectingHandlersCallRefresh (python/djust/tests/test_data_table_handler_contracts.py). Subclasses that overrode table_X_event class attrs are unaffected — only the bare-default path was broken.

@action no longer re-raises after recording exception state. Closes #1276. The decorator's docstring promised templates could read {{ <name>.error }} after an exception, but the implementation re-raised — the dispatcher's exception-frame path then bypassed the re-render and the template never saw the recorded error field. Fix: catch Exception (not BaseException), record state, log at ERROR level via logger.exception, return None. BaseException subclasses (KeyboardInterrupt, SystemExit, GeneratorExit) still propagate by Python convention. 8 new regression cases in TestActionExceptionDoesNotPropagate + TestActionSuccessRecordsState

• TestActionLazyInitializesActionState (python/djust/tests/test_action_decorator_contract.py); 6 existing tests in test_action_decorator.py updated to the new contract. Docstring at decorators.py:262-272 rewritten to match. Behavior change: code that wraps @action calls in try/except to handle the re-raise now sees a clean return. Mirror the old behavior by re-raising explicitly inside the handler.

Lifecycle Coverage Audit + Decorator/Tag Contract Audit (docs/audits/lifecycle-2026-05.md, docs/audits/decorator-contract-2026-05.md). Two companion audit docs modeled on the v0.9.2-4 VDOM audit. Document the canonical state-type × lifecycle-hook matrix and the decorator/tag-name dispatch contract, surfaced from 10 downstream consumer bug reports (#1267, #1273-#1281). The lifecycle audit catalogues 8 ranked weaknesses including the central control-flow gaps in mount() (#1280, #1281). The decorator/tag audit catalogues 8 weaknesses including the data_table tag emitting 23 event names that don't match any DataTableMixin handler (#1275 generalized). Each audit ships with a 4-phase improvement roadmap, test gaps, strategic observations, and a companion canon update for CLAUDE.md / PR-checklist. Pre-staged issues filed for each not-yet-tracked weakness (#1283-#1291). Audit-driven Phase 1 fixes blocking v0.9.2 stable will land in the v0.9.2-5 drain bucket; Phase 2/3 fixes targeted for v0.9.3.

Production Deployment guide extended with Tier 1/2/3 patterns (docs/website/guides/deployment.md). Adds 8 new sections to the canonical deployment guide based on patterns surfaced from real-world djust deployments:

• Channel Layer (cross-process push) — separate concern from DJUST_STATE_BACKEND, required when any view uses push_to_view, presence, or cursor tracking.
• Database Connection Pooling — three-layer guidance (CONN_MAX_AGE, PgBouncer, RDS Proxy), with the LISTEN/NOTIFY caveat for transaction-mode pooling.
• Celery Integration — broker choice (Redis vs SQS), pool choice (prefork vs gevent), beat-singleton invariant, gevent monkey-patch gotcha, queue-depth-based worker auto-scaling.
• Static and Media Files — cloud-agnostic CDN options, S3 + CloudFront config, ASGI_SERVE_STATIC=False opt-out for offloading static-file serving from the ASGI server.
• WebSocket stickiness on AWS ALB — the simpler "stick on Django sessionid" pattern as an alternative to a custom application-set cookie.
• Sizing and Scaling Tiers — concrete vCPU/RAM recommendations indexed to concurrent active users (≤50, 50-500, >500), with explicit "when to escalate" triggers.
• "What's Already Production-Ready in djust" — anti-recommendation list (Redis state, channels_redis, sync_to_async, transaction.on_commit, Origin check, HSTS) so users don't re-evaluate canonical patterns on every deployment.
• Extended Gunicorn+Uvicorn workers section with concrete production CMD + flag rationale (-w sizing, --timeout 120, --keep-alive 5). Cloud-agnostic where possible; AWS as canonical example with PgBouncer / GCS / Cloudflare R2 noted in parallel.

• permanent proptest seed in fuzz_test.proptest-regressions.