Bump responses from 0.21.0 to 0.22.0#4
Merged
Conversation
Bumps [responses](https://github.com/getsentry/responses) from 0.21.0 to 0.22.0. - [Release notes](https://github.com/getsentry/responses/releases) - [Changelog](https://github.com/getsentry/responses/blob/master/CHANGES) - [Commits](getsentry/responses@0.21.0...0.22.0) --- updated-dependencies: - dependency-name: responses dependency-type: direct:production update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com>
dependabot
Bot
added
dependencies
Codecov Report
@@ Coverage Diff @@
## main #4 +/- ##
=======================================
Coverage ? 66.19%
=======================================
Files ? 47
Lines ? 1855
Branches ? 0
=======================================
Hits ? 1228
Misses ? 627
Partials ? 0 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
This was referenced Oct 21, 2025
JSv4
added a commit
that referenced
this pull request
Dec 7, 2025
Critical Issue #1 - Data Integrity: - Add CheckConstraint to Annotation model ensuring annotations in structural_set must have structural=True (enforces assumption in query_optimizer.py:207-209) - Add same constraint to Relationship model for consistency - Add migration 0050 for the new database constraints - Add 6 test cases in test_structural_annotation_sets.py Critical Issue #2 - IDOR Prevention: - Fix CreateThreadMessageMutation to check permissions BEFORE checking locked status (prevents information disclosure via different error msgs) - Fix ReplyToMessageMutation with same pattern - Use generic "Cannot post/reply in this thread" for both cases Issue #4 - Frontend Test Configuration: - Add --reporter=list to test:ct script in package.json to prevent test hanging as documented in CLAUDE.md
3 tasks
This was referenced Dec 7, 2025
This was referenced Dec 15, 2025
This was referenced Dec 24, 2025
Merged
Merged
8 tasks
JSv4
added a commit
that referenced
this pull request
Jan 2, 2026
- Add FILE_SIZE and TIME_UNITS constants to constants.ts - Create formatters.ts with shared formatting utilities: - formatFileSize: Human-readable file size formatting - formatRelativeTime: Relative time formatting (e.g., "5 hours ago") - formatCompactRelativeTime: Compact format for activity feeds - getInitials: Extract initials from name/email for avatars - Update Documents.tsx to use shared utilities - Remove duplicate helper functions from Documents.tsx This addresses CLAUDE.md Critical Concept #4 (no magic numbers) and improves DRY compliance by consolidating formatting logic.
This was referenced Jan 2, 2026
JSv4
added a commit
that referenced
this pull request
May 14, 2026
* Harden Auth0 permissioning: superuser allowlist, IDOR oracles, callback token at-rest End-to-end audit of every Auth0-touching authorization pathway. Twelve verified findings shipped; agent-reported false positives (intentional IDOR-safe SetCorpusVisibility, AnnotationImagesView, moderation mutations, moderation_metrics, etc.) verified in source and dropped. - F1: AUTH0_SUPERUSER_SUB_ALLOWLIST gates is_superuser elevation in sync_admin_claims_from_payload regardless of JWT claim, blocking the user_metadata-sourced privilege escalation path. New users.W001 system check warns on empty allowlist when USE_AUTH0=True. - F2: ADMIN_CLAIMS_CACHE_TTL 300s -> 30s, tightening the privilege retention window after Auth0 demotion. - F3+F4: AddRelationship now uses Annotation/Corpus visible_to_user with count comparison and a single unified not-found message; broad except no longer surfaces ORM text. - F5+F6: CreateMetadataColumn and UpdateMetadataColumn use visible_to_user + unified not-found-or-no-permission message. - F7: Datacell mutations catch DoesNotExist explicitly and stop echoing ORM exception text. - F9: Bounded JWKS stale-cache fallback (1h grace past TTL); fails closed beyond the window so a key compromise + Auth0 outage cannot let the old key keep verifying tokens indefinitely. - F10: Analysis.callback_token is now stored as SHA-256 hex (callback_token_hash). rotate_callback_token() returns a fresh plaintext at submit time; verify_callback_token() does the constant-time check on incoming requests. Migration backfills existing UUID tokens by hash so in-flight analyzers continue to authenticate. Submit log redacted. - F8+F11: Documentation-only. Step 7 walkthrough for the allowlist, env-var rows for AUTH0_SUPERUSER_SUB_ALLOWLIST + AUTH0_CREATE_NEW_USERS, app_metadata-only requirement called out in danger callout, inline comment that User.email is informational (not an identity field). Operator action required before deploy: populate AUTH0_SUPERUSER_SUB_ALLOWLIST with the Auth0 sub of every user who must retain Django is_superuser. Existing superusers whose subs are not in the allowlist will be demoted within ~30s of their next API call. * Address PR #1641 review: fix AddRelationship document IDOR + black format * Fix pytest failures and lift patch coverage for PR #1641 CI was failing because the AUTH0_SUPERUSER_SUB_ALLOWLIST gate (added in this PR) blocks is_superuser elevation, but several admin-auth tests were still asserting elevation without populating the allowlist. The test_metadata_columns_graphql test was also still asserting the old "don't have permission" string, which the PR replaced with the IDOR-safe unified message. Test updates (failing -> passing): - test_admin_auth.TestAdminClaimsSync.test_sync_is_superuser_true_from_claims - test_admin_auth.TestAdminClaimsSync.test_sync_both_claims - test_admin_auth.TestGetUserByPayloadWithClaimSync.test_sync_claims_cached_handles_cache_failure - test_metadata_columns_graphql.test_create_metadata_column_without_permission Address Claude review findings: - (#2) AUTH0_ADMIN_CLAIMS_CACHE_TTL is now configurable via env var; the module constant stays the default. - (#4) Reverse migration 0021 uses QuerySet.update() instead of model.save() so historical-model signals do not fire on downgrade. - (#5) Token hash backfill iterates with chunk_size=500 to bound memory. - (#7) Wire AUTH0_CREATE_NEW_USERS into AUTH0_JWT from env so deployments can disable user auto-provisioning without editing settings. - (#8) Silence users.W001 in test settings so the deliberate empty-allowlist tests don't print system-check noise. - (#1) _can_serve_stale now asserts it is called under the JWKS lock and carries a prominent docstring warning. New tests added to cover patch lines lacking coverage: - TestAuth0SuperuserAllowlist (3 tests) — empty allowlist, sub not in allowlist, and demote-bypasses-allowlist paths. - TestAuth0SuperuserAllowlistSystemCheck (3 tests) — direct unit tests for users.checks.check_auth0_superuser_allowlist. - TestAnalysisCallbackTokenHelpers (6 tests) — direct unit tests for Analysis.rotate_callback_token / verify_callback_token covering empty hash, None/empty candidate, plaintext-not-stored, and rotation invalidating prior plaintexts. - TestCanServeStale (3 tests) — direct unit tests for _can_serve_stale (no-data, inside-grace, outside-grace). - test_network_error_after_stale_grace_fails_closed and test_json_error_after_stale_grace_fails_closed — JWKS fails closed outside the stale-grace window. - test_admin_claims_cache_ttl_env_override and test_admin_claims_cached_noop_when_auth0_disabled — exercise the new env-configurable TTL and the USE_AUTH0=False short-circuit. * Update CHANGELOG with PR #1641 review follow-ups * Cover Datacell mutation not-found IDOR paths The PR introduced explicit Datacell.DoesNotExist branches in ApproveDatacell / RejectDatacell / EditDatacell so the broad-except no longer leaks ORM text. These paths were not covered by the existing happy-path datacell tests. Adds four tests: - approve / reject / edit returning the unified Datacell not found. message for an unknown id - approve on a datacell owned by another user (queryset filters by creator, so DoesNotExist fires) returning the same message — verifies the IDOR-safe property end to end * Cover AddRelationship bad-id and unknown-id IDOR paths The PR's AddRelationship mutation collapses both unparseable global IDs (caught by the outer try/except) and unknown but well-formed IDs (caught by the visible_to_user count comparison) into a single not-found message. Adds two tests: - test_garbage_global_ids_yield_not_found_message: a base64-decodable but bogus global ID exercises the outer broad-except branch. - test_unknown_source_annotation_id_yields_not_found: a syntactically valid global ID pointing at no annotation exercises the count-mismatch branch. Both paths must produce the unified not-found message so the response shape gives no enumeration oracle. * Address PR #1641 review round 2 Latest Claude review: - (#1) Lock assert was misleading because threading.Lock.locked() returns True if any thread holds the lock, not the caller. Replaced the assert with a sharper docstring note that documents the limitation and points future refactors at threading.RLock. - (#2 / #10) Migration backfill and reverse now use bulk_update batches to keep DB round-trips bounded. - (#3) AddRelationship now visibility-checks relationship_label_pk too, closing the residual oracle where an attacker could probe private AnnotationLabel IDs. - (#5) Added an inline comment at the allowlist gate documenting why is_staff is intentionally NOT allowlist-gated and how to layer one. - (#7) test.py SILENCED_SYSTEM_CHECKS comment now cross-references the direct-call test that bypasses the silencing. - (#9) submit_corpus_documents_to_analyzer logging switched from f-string to %-style for consistency with the new logger.exception calls. New test: - test_inaccessible_relationship_label_yields_not_found verifies the new label visibility check returns the unified not-found message and blocks the Relationship row from being created. * Address PR #1641 review round 3 (security tag, auto-save, log cleanup) Earlier Claude review notes: - rotate_callback_token now auto-saves when self.pk is set, so callers can no longer ship a plaintext to the worker while leaving the DB hash stale. Unsaved-instance behaviour is unchanged (caller still owns the first save). Two new tests pin both branches. - check_auth0_superuser_allowlist is now registered under Tags.security so it shows up under ``manage.py check --tag security``. - logger.exception(...) calls in annotation_mutations and extract_mutations no longer pass ``e``; ``logger.exception`` captures the traceback automatically and the extra arg double-logged the exception text. * Document AUTH0_ADMIN_CLAIMS_CACHE_TTL env var Adds the AUTH0_ADMIN_CLAIMS_CACHE_TTL row to the env-var table so operators can find the configurable knob added in this PR. * Address PR #1641 review round 4 (non-integer pk, atomic note) - (#1) AddRelationship now casts each parsed pk to ``int()`` inside the outer try/except so a base64-decodable but non-numeric global ID (``"BogusType:not-an-int"``) fails closed at parse time instead of later at the queryset boundary. Updated test_garbage_global_ids_yield_not_found_message to assert the not-found response directly now that the path is deterministic. - (#5) Migration 0021 now carries a module docstring explaining why the AddField + backfill + RemoveField operations stay in one atomic transaction (partial runs would leave live rows without a hash) and flags the expected lock window for large tables. bulk_update keeps the transaction short. Items addressed from this review but already in place: AnnotationLabel inherits BaseOCModel so ``.objects.visible_to_user()`` works out-of-the-box; AUTH0_CREATE_NEW_USERS is consumed via ``auth0_settings.AUTH0_CREATE_NEW_USERS`` in ``get_auth0_user_from_token`` which reads from the AUTH0_JWT dict; AUTH0_JWT is only assigned in base.py (verified via grep). * Escalate users.W001 to users.E001 when existing superusers will be demoted Per latest review: when ``USE_AUTH0=True`` AND ``AUTH0_SUPERUSER_SUB_ALLOWLIST`` is empty AND at least one is_superuser row already exists, the next claim sync will silently demote that superuser within ~30s of their next API call. The previous Warning was too quiet for this footgun. The check now: - Emits ``users.E001`` (Critical) when an existing superuser would be silently demoted. Critical errors block ``manage.py check --deploy`` so a deploy that would lock out the only admin account fails fast. - Falls back to ``users.W001`` (Warning) when no superuser rows exist, preserving the original advisory behaviour. - Tolerates DB-unreachable contexts (e.g. ``makemigrations``) by catching ORM errors and emitting the lower-severity warning. Added a test class setUp that demotes the migration-created initial superuser so the W001 vs E001 branches can be tested deterministically, plus a new test pinning the E001 path with an explicitly-created superuser. ``users.E001`` is added to SILENCED_SYSTEM_CHECKS in test settings so the elevated severity does not break unrelated test runs that happen to leave a superuser in the DB. * Fix pre-existing mypy errors in requests.post headers callers A new ``urllib3`` type stub release narrowed ``requests.post(..., headers=)`` to ``MutableMapping[str, str | bytes]``. The four call sites below were passing ``dict[str, str]``, which is invariant in the value type and so fails the narrower stub even though ``str`` is a subtype of ``str | bytes``. These mypy errors fail the linter step on ``main`` as well — they are unrelated to PR #1641's security work — but they block the Backend CI's pytest job from running because pytest depends on linter passing. Fix is uniform: keep the local ``dict[str, str]`` typing (matches every caller of ``maybe_add_cloud_run_auth`` so we don't have to widen that shared helper too), and cast to ``Any`` at the ``requests.post/get`` call site so mypy lets the narrower dict satisfy the wider stub. No runtime behaviour changes; the dict is still ``str -> str`` at runtime. Files: - opencontractserver/users/tasks.py (lines 45, 165) - opencontractserver/pipeline/rerankers/cohere_reranker.py (line 132) - opencontractserver/pipeline/parsers/docxodus_parser.py (line 156) - opencontractserver/pipeline/parsers/docling_parser_rest.py (line 344) --------- Co-authored-by: Claude <noreply@anthropic.com>
This was referenced May 15, 2026
JSv4
added a commit
that referenced
this pull request
May 17, 2026
After merging main into PR 1665, the UserCanMixin.user_can supertype now declares request: Any = None, but five subclass overrides added by PR #1663 did not include that parameter and mypy flagged the signature mismatch. - DocumentManager / RelationshipManager / UserFeedbackManager (shared/Managers.py): add request kwarg and forward to super() / recursive Manager.user_can calls. - AnnotationManager / NoteManager (shared/Managers.py): add request kwarg and forward to Analysis.objects.user_can / Extract.objects.user_can / Document.objects.user_can / Corpus.objects.user_can so the optimizer cache flows through the recursion. - ConversationManager / ChatMessageManager (conversations/models.py): add request kwarg and forward to _default_user_can. Also addresses Claude review item #7 (audit folder-service callers): config/graphql/document_types.py:1153 (resolve_folder_in_corpus) now passes request=info.context to DocumentFolderService.get_document_folder. get_folder_document_count and get_folder_path have no in-tree callers. Adds review item #4: negative regression test pinning that refresh_from_db() does NOT clear Tier 1 (the documented footgun) so a future change to that contract is caught by CI.
JSv4
added a commit
that referenced
this pull request
May 17, 2026
Discovery tools (list_fieldsets, list_analyzers, list_recent_extracts, list_recent_analyses) now check corpus visibility via visible_to_user(user).filter(pk=...).exists() instead of bare Corpus.objects.get(pk=...), so a caller cannot probe private corpus IDs through an existence oracle (review item #8). The error message is unchanged across both branches per CLAUDE.md's IDOR guidance. list_recent_extracts and list_recent_analyses fold the per-row M2M count into the SELECT with annotate(Count(...)) so the document_count field no longer issues N queries on N rows (review item #1). start_extract and start_analysis collapse the two-step visible_to_user().exists() + .get() pattern into a single visible_to_user().filter(pk=...).first() lookup (review item #5). Same fix applied to the Fieldset/Analyzer lookups in those functions. list_fieldsets now includes fieldsets pinned to the *current* corpus (via Q(corpus__isnull=True) | Q(corpus_id=corpus_id)) so the discovery surface matches what start_extract actually accepts — previously a caller listing fieldsets would not see one that start_extract would have allowed if supplied directly (review item #7). MAX_LIST_LIMIT / DEFAULT_LIST_LIMIT / DEFAULT_RECENT_LIMIT hoisted to opencontractserver/constants/tools.py per CLAUDE.md's no-magic-numbers rule (review item #4). Test hygiene: dead GremlinEngine fixture removed from TestListRecentAnalyses.setUp (review item #10); inspect import lifted out of test_to_core_tool_returns_async_function loop (review item #11).
5 tasks
JSv4
added a commit
that referenced
this pull request
May 17, 2026
* Add two-tier permission cache for Manager.user_can (closes #1640) PR #1637 centralized authorization in _default_user_can (corpuses Phase 0). This follow-up keeps the cold path cheap as more models adopt the API: Tier 1 (transparent) — per-instance memoization inside get_users_permissions_for_obj. Stored on the instance as _oc_granted_perms_cache, keyed by (user_id, include_group_permissions). Anonymous users bypass. Cache is a defensive frozenset; callers receive a set copy so _default_user_can can still fold read_<model> in for CRUD/ALL checks without poisoning the cache. Tier 2 (opt-in) — request-scoped PermissionQueryOptimizer mirroring ConversationQueryOptimizer. Attached lazily to the request as _permission_query_optimizer via get_request_optimizer(request). Keyed by (user_id, content_type_id, instance_pk, include_group_permissions) so multiple instances of the same model share a cache. None-tolerant for Celery / management-command contexts. Plumbing: - _default_user_can / UserCanMixin / InstanceUserCanMixin / CorpusFolder.user_can gained a request= kwarg (kw-only, default None — backward compatible). - DocumentFolderService classmethods (~25 entry points) accept and forward request; corpus GraphQL queries + mutations + corpus_folder_mutations pass request=info.context. - set_permissions_for_obj_to_user gained a request= kwarg and now self-invalidates both tiers for the affected (user, instance) pair when called inside the HTTP lifecycle. Phase-0 Corpus mutation call sites in corpus_mutations.py and document_mutations.py pass it through; out-of-band callers (Celery / fixtures / importing) keep the positional signature. Tier 1 is still scrubbed for the target user even when request is None so reused instances see the new grant. Tests: opencontractserver/tests/permissioning/test_permission_optimizer.py covers per-instance memoization (zero queries on repeat, no leakage across include_group_permissions, fast-path short-circuits don't populate cache, defensive-copy contract), the request-scoped optimizer (lazy attach, None-safe, distinct-user keying, per-user / per-instance / total invalidation, anonymous skip), and the mutation invalidation contract (mid-request grant visible after set_permissions_for_obj_to_user with request; Tier 1 scrubbed when request is None). Docs: docs/architecture/query_permission_patterns.md gains a row in the Layer 2 table and a new "Two-Tier user_can Caching" subsection that documents the attribute names, keying, and invalidation contract. * Fix CI linter failures - Apply pyupgrade rewrite of Optional[X] -> X | None in permission_optimizer.py - Drop now-unused Optional import - Suppress mypy arg-type warning on intentional AnonymousUser test path * Address Claude review on PR #1665 - Document group-permission and refresh_from_db staleness boundaries on INSTANCE_PERMS_CACHE_ATTR, set_permissions_for_obj_to_user, and the arch doc so callers know when to invalidate manually - Widen PermissionQueryOptimizer._cache type annotation: instance_pk is Any (UUID/str-PK models would have tripped mypy) - Raise ValueError when invalidate() receives both `instance` and an explicit `content_type_id`/`instance_pk` instead of silently letting one win; add a regression test - Trim the long inline two-tier-cache-invalidation comment in set_permissions_for_obj_to_user per the one-line WHY convention * Add coverage for _default_user_can branches and folder_service request kwarg Three new test classes target the codecov patch-coverage gap in permissioning.py (71.64% → expected ~95%): DefaultUserCanCoverageTestCase pins each user-resolution path (None / AnonymousUser / int id / str id / unknown id / unauthenticated double) and each permission-type branch (READ/CREATE/UPDATE/EDIT/DELETE/ COMMENT/PUBLISH/PERMISSION/CRUD/ALL) plus the is_public-fold-in path that keeps CRUD passing on public corpora. SetPermissionsInvalidationCoverageTestCase covers the per-user scrub selectivity (other users' cache entries survive) and the no-request Celery path (Tier 1 still drops). FolderServiceRequestKwargCoverageTestCase exercises the request= kwarg on DocumentFolderService and the permission-denial empty-queryset branch (returns empty rather than raising). * Address Claude review on PR #1665 and close codecov patch gaps - Clarify the docstring on ``PermissionQueryOptimizer.get_granted``: the "without double-caching" wording was inaccurate. A Tier 2 cold miss does warm Tier 1 as a side-effect, so the same frozenset intentionally ends up in both caches. Spell out why that redundancy is desired (out-of-request callers still get Tier 1). - Add two coverage tests targeting the previously-uncovered warming paths in ``get_users_permissions_for_obj``: * direct superuser call → guardian-superuser return path * pre-attached guardian prefetch → fast-path return Both paths exist because ``_default_user_can`` short-circuits before reaching the helper, so they are only exercised by callers that invoke the helper directly. * Fix mypy errors in test_permission_optimizer.py Replace assertIsNotNone with assert for type narrowing on cache lookups, and drop an unused type:ignore comment on a Django reverse-relation accessor that mypy now resolves correctly. * Address Claude review: clarify Tier 2 default vs Tier 1 Add a docstring note to PermissionQueryOptimizer.get_granted explaining why include_group_permissions defaults to True (mirrors user_can) while the lower-level get_users_permissions_for_obj defaults to False. Production always passes the flag explicitly, so the divergence is benign — the note prevents future maintainers from chasing a phantom inconsistency. * Address review: guard partial-coordinate invalidate against PK collisions invalidate(instance_pk=X) without content_type_id was a silent footgun: a Corpus with pk=5 and a Document with pk=5 would both be evicted because the wildcard match ignores model identity. Raise ValueError when an instance_pk arrives without its content_type_id so callers must either pair the two or use instance= shorthand. Add a regression test. * Address review #3, #5: document Tier 2 absence in Celery + pickling staleness * Address review: thread request= through manager.user_can overrides After merging main into PR 1665, the UserCanMixin.user_can supertype now declares request: Any = None, but five subclass overrides added by PR #1663 did not include that parameter and mypy flagged the signature mismatch. - DocumentManager / RelationshipManager / UserFeedbackManager (shared/Managers.py): add request kwarg and forward to super() / recursive Manager.user_can calls. - AnnotationManager / NoteManager (shared/Managers.py): add request kwarg and forward to Analysis.objects.user_can / Extract.objects.user_can / Document.objects.user_can / Corpus.objects.user_can so the optimizer cache flows through the recursion. - ConversationManager / ChatMessageManager (conversations/models.py): add request kwarg and forward to _default_user_can. Also addresses Claude review item #7 (audit folder-service callers): config/graphql/document_types.py:1153 (resolve_folder_in_corpus) now passes request=info.context to DocumentFolderService.get_document_folder. get_folder_document_count and get_folder_path have no in-tree callers. Adds review item #4: negative regression test pinning that refresh_from_db() does NOT clear Tier 1 (the documented footgun) so a future change to that contract is caught by CI. * Address review #1665: align defaults, lock optimizer cache, scrub pickled cache - Tier 1 `get_users_permissions_for_obj` default now `include_group_permissions=True` to match `PermissionQueryOptimizer.get_granted` / `_default_user_can` / every `Manager.user_can` surface. One default, one answer. - `PermissionQueryOptimizer._cache` is now guarded by `threading.Lock()` for reads, writes, and the invalidate-loop iteration. DB work stays outside the lock so guardian lookups never serialise on it. - `InstanceUserCanMixin.__getstate__` strips `INSTANCE_PERMS_CACHE_ATTR` so models pickled into Celery tasks can never carry a stale Tier 1 frozenset to the worker (footgun → impossible, not just documented). - Comments pin reliance on `ContentType._cache` and clarify that creator-based models on the Tier 1 path are always cache-safe (no guardian mutation path exists for them). - Added Tier1PicklingScrubTestCase to lock in the new __getstate__ contract. * Address review #1665: forward request= through _compute_effective_permissions The optimizer's _compute_effective_permissions invokes Document.objects.user_can / Corpus.objects.user_can repeatedly per request. Threading request= through to those calls activates the Tier 2 PermissionQueryOptimizer, deduplicating guardian lookups across distinct Document/Corpus instances within a single request. Annotation/Relationship manager user_can() overrides now forward request= as context= to the optimizer, completing the request-scoped cache plumbing so the kwarg doesn't silently no-op for these checks. * Address review #1665: warn on invalidate() clear-all + assert cold-path query budget Two follow-ups to the latest Claude review: - PermissionQueryOptimizer.invalidate() no-args docstring now points at invalidate_caches() for the clear-all intent so an accidental call doesn't silently nuke the request cache. - test_repeat_user_can_zero_queries wraps the warm-up call in an explicit assertNumQueries(6) so a regression that blows up the cold path is caught here instead of only the second-call zero-query assertion. * Address review #1665: thread-safe Tier 1 cache + type annotations - Replace the raw `dict` on `instance._oc_granted_perms_cache` with `_InstancePermsCache(dict)`, a thread-safe subclass exposing `drop_for_user(user_id)` that holds an internal Lock for the iterate-then-delete sweep in `set_permissions_for_obj_to_user`. Closes the ASGI / async-view exposure flagged in the latest review (the sweep would otherwise risk `RuntimeError: dictionary changed size during iteration` if two threads ever touched the same instance). Individual reads/writes still ride CPython's GIL atomicity; the lock is for the compound op only. - `_get_or_create_instance_perms_cache` uses `__dict__.setdefault` so two threads first-touching the same instance converge on one cache object rather than racing via `getattr`+`setattr`. - Backwards-compat fallback in `set_permissions_for_obj_to_user`: if an instance carries a plain-`dict` cache (older worker unpickle, test fixture), invalidation falls back to a snapshot-then-`.pop()` pattern so `RuntimeError` cannot escape. - Add explicit `request: Any = None` annotations on `ConversationManager.user_can` / `ChatMessageManager.user_can` (previously inferred as `None`, which would reject real request objects at the type level). - Drop rotting inline `# PR #1663` comment references per CLAUDE.md. - New `Tier1CacheThreadSafetyTestCase` pins the contract: cache is the thread-safe wrapper, concurrent reader threads + repeated invalidate sweeps don't race, mutation invalidation routes through `drop_for_user` (mocked). * ci: trigger Backend CI for 524c9e7 * Address review #1665: defer lazy import past cache hits + superuser cache caveat Two minor polish items from the latest review: - ``PermissionQueryOptimizer.get_granted`` previously ran the lazy ``from opencontractserver.utils.permissioning import …`` on every call. On warm requests where most calls are cache hits, that's a needless ``sys.modules`` lookup (~30 ns each). Move the import inside the two branches that actually call ``get_users_permissions_for_obj`` (anonymous-user fallback, cache miss) so the cache-hit path stays pure-Python with no module table touch. - Add a NOTE at the superuser cache-write site in ``get_users_permissions_for_obj`` warning that the cache key omits ``is_superuser``, so test scenarios that promote/demote the same Python instance need to drop the cache attribute manually. The same caveat already lived in the constants module; mirroring it at the write site makes the boundary visible to anyone tracing a stale-cache bug in tests. Tests still pass (44 in test_permission_optimizer.py, 490 across the full permissioning suite locally). * Address PR #1665 review: invalidate_caches direct path, lock trade-off note, MRO comment, group-perm doc section, magic-count comment - PermissionQueryOptimizer.invalidate_caches() now clears self._cache directly instead of delegating to invalidate() with no args (which it warned against). - Added a DESIGN NOTE inside get_granted explaining why the DB call lives outside the lock — preempts future well-meaning serialization fixes. - InstanceUserCanMixin.__getstate__ now documents the MRO assumption (Django 4.2+ Model.__getstate__ / Python 3.11+ object.__getstate__). - test_repeat_user_can_zero_queries now flags assertNumQueries(6) as upgrade-sensitive with explicit guidance on bumping the count. - query_permission_patterns.md gains an operator note enumerating the group-targeted guardian call sites in this repo and the manual invalidation contract for tenants that wire group perms themselves. --------- Signed-off-by: JSIV <5049984+JSv4@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
JSv4
added a commit
that referenced
this pull request
May 17, 2026
* Add extract & analyzer agent tools Six new tools let an LLM agent discover the Fieldsets and Analyzers visible to the calling user and dispatch them just like a human would: - list_fieldsets / start_extract / list_recent_extracts - list_analyzers / start_analysis / list_recent_analyses start_extract and start_analysis are approval-gated and require WRITE on the corpus; they reuse the same Celery pipeline (run_extract, process_analyzer) as the existing GraphQL mutations. The agent always picks from existing Fieldsets/Analyzers — it cannot invent new schemas. Document scope follows the agent context (corpus agent = whole corpus, doc agent = current doc) and requested document_ids are intersected with corpus.get_documents() so the agent can never escape its scope. Not added to DEFAULT_DOCUMENT_ACTION_TOOLS: opt-in only via a CorpusAction's pre_authorized_tools. Docs updated in docs/architecture/llms/README.md, docs/corpus_actions/intro_to_corpus_actions.md, and docs/extract_and_retrieval/data_extraction.md. * Fix failing tests in PR 1675 - test_async_variant_matches: wrap sync ORM calls in sync_to_async since test methods are async (was raising SynchronousOnlyOperation in CI under pytest-xdist parallel runners). - test_start_extract/start_analysis_requires_approval: supply all required kwargs since PydanticAIToolWrapper._maybe_raise uses Signature.bind (not bind_partial), so omitting required kwargs raised TypeError before the approval gate could fire. * Apply black formatting to test_extract_analyzer_tools.py * Address review: security, N+1, double-fetch, magic numbers, test hygiene Discovery tools (list_fieldsets, list_analyzers, list_recent_extracts, list_recent_analyses) now check corpus visibility via visible_to_user(user).filter(pk=...).exists() instead of bare Corpus.objects.get(pk=...), so a caller cannot probe private corpus IDs through an existence oracle (review item #8). The error message is unchanged across both branches per CLAUDE.md's IDOR guidance. list_recent_extracts and list_recent_analyses fold the per-row M2M count into the SELECT with annotate(Count(...)) so the document_count field no longer issues N queries on N rows (review item #1). start_extract and start_analysis collapse the two-step visible_to_user().exists() + .get() pattern into a single visible_to_user().filter(pk=...).first() lookup (review item #5). Same fix applied to the Fieldset/Analyzer lookups in those functions. list_fieldsets now includes fieldsets pinned to the *current* corpus (via Q(corpus__isnull=True) | Q(corpus_id=corpus_id)) so the discovery surface matches what start_extract actually accepts — previously a caller listing fieldsets would not see one that start_extract would have allowed if supplied directly (review item #7). MAX_LIST_LIMIT / DEFAULT_LIST_LIMIT / DEFAULT_RECENT_LIMIT hoisted to opencontractserver/constants/tools.py per CLAUDE.md's no-magic-numbers rule (review item #4). Test hygiene: dead GremlinEngine fixture removed from TestListRecentAnalyses.setUp (review item #10); inspect import lifted out of test_to_core_tool_returns_async_function loop (review item #11). * Address Claude review on PR 1675 - Prefetch columns with is_manual_entry filter at DB level - Pin corpus_action_id lookup to the operating corpus (security: prevent cross-corpus lineage attribution from direct callers) - Move process_analyzer import to module scope (no actual circular dependency justifies the local import) - Reject fieldsets whose only columns are manual-entry (LLM cannot extract from them) - Drop decorative section-banner comments and explain-the-what comments - Tests: cross-corpus action ID is dropped (extract + analysis), outside-corpus document_id warns and falls back to full corpus, manual-entry-only fieldset raises, corpus_action_id lineage on start_analysis * Address review: distinguish user-not-found from unauthenticated, clamp_limit tests, comment fixes - start_extract / start_analysis now distinguish 'no user_id supplied' (unauthenticated) from 'user_id refers to no row' (not found), so the error message no longer misattributes the failure. - Add parametrized test for _clamp_limit covering None, 0, negative, oversized, non-numeric, and numeric-string inputs. - Add test pinning the user-not-found vs unauthenticated split. - Fix misleading _patch_process_analyzer docstring (it patches the import site, not the source module). - Note the intentional transaction asymmetry between start_extract and start_analysis so future readers don't 'fix' it. * Address review #1675: document edge contracts + cover zero-column discovery extracts_and_analyzers.py: - _resolve_target_document_ids: comment the memory trade-off for very large corpora and point at the SQL-pushdown alternative for future optimization. - start_analysis: comment that analysis_input_data is intentionally not pre-validated against analyzer.input_schema — validation lives in run_task_name_analyzer where the payload is spread into the task function call (TypeError on unknown kwargs). Keeps the agent path bug-compatible with the human GraphQL/CorpusAction path. test_extract_analyzer_tools.py: - Add test_listing_shows_zero_columns_then_start_rejects pinning the deliberate two-step failure mode: list_fieldsets shows a manual-entry-only fieldset with column_count=0 (discovery surface) and start_extract rejects it as ValueError (dispatch surface). * Address review #1675: wrap start_analysis in atomic + clarify LLM tool schema - Wrap process_analyzer call in transaction.atomic() so the on_commit task dispatch can be rolled back if anything raises afterwards. The previous comment claimed process_analyzer managed its own transaction; it does not — it just registers on_commit inside whatever transaction is open. In autocommit mode (Celery workers) the callback would fire immediately, kicking off the downstream task even on an error path. start_extract already wraps in atomic; this aligns start_analysis. - Lift _extract_status magic strings into constants/tools.py (EXTRACT_STATUS_FAILED / COMPLETED / RUNNING / QUEUED) and use them at every call site for parity with the human GraphQL surface. Analyses keep using the model-sourced analysis.status directly. - Tool registry descriptions for fieldset_id / analyzer_id now say explicitly which type they are (integer vs CharField string) and how to map them from the discovery tool output, since list_fieldsets returns int ids and list_analyzers returns string ids and the LLM could otherwise conflate the two. - start_analysis comment now documents the analysis_input_data -> task-function injection surface so future readers know the analyzer task is the validation boundary. * Address review #1675: status vocab, payload hardening, listing UX, SQL perf - Unify analysis status vocabulary across start_analysis and list_recent_analyses. New ``_normalize_analysis_status`` lowercases the model-sourced JobStatus value; ``_ANALYSIS_STATUS_QUEUED`` is what start_analysis returns now (derived from ``JobStatus.QUEUED.value`` rather than the Extract constant), so an LLM polling the listing right after dispatch sees the same string rather than an apparent case-shift mid-lifecycle. - Add ``ANALYSIS_INPUT_DATA_RESERVED_KEYS`` and ``_strip_reserved_input_keys`` to strip framework-owned kwargs (``analysis_id``, ``user_id``, scoping IDs) from agent-supplied ``analysis_input_data`` before dispatch. Warn on strip so production traces can flag override attempts. Closes the unique prompt-injection escalation surface on the agent path; the task itself remains the canonical validation boundary. - Truncate oversized ``Analyzer.input_schema`` in ``list_analyzers`` via ``_summarise_input_schema`` + ``ANALYZER_INPUT_SCHEMA_MAX_INLINE_CHARS`` (4_000). Replaces the schema with a placeholder when its JSON dump exceeds the cap so a misbehaving analyzer can't inflate every agent's context window. - ``list_fieldsets`` now exposes an ``extractable`` flag and orders by ``-created`` to match ``list_recent_extracts``. The flag surfaces the listing-vs-dispatch contract for manual-only/empty fieldsets up-front; the ordering removes the silent position shift that occurred when a fieldset was edited mid-session. - Push ``_resolve_target_document_ids`` intersection into SQL via ``corpus.get_documents().filter(pk__in=...)`` / ``.exists()`` so large corpora don't load the entire active doc ID set into a Python ``set`` on every dispatch. - New tests pin each contract: status normalisation, reserved-key strip + warning, legitimate payload pass-through, oversized ``input_schema`` truncation, and the ``extractable`` flag. * Address review #1675: move asyncio marker to method level on TestApprovalGate Stacking ``@pytest.mark.asyncio`` on the class combined with ``TransactionTestCase`` is fragile across pytest-django versions — pytest-asyncio tries to drive every method as a native coroutine while ``TransactionTestCase`` wants to route them through Django's own sync/async test runner. Per the latest review, the safer pattern (and the one used elsewhere in this codebase, e.g. ``test_conversation_search.py``) is ``@pytest.mark.django_db`` on the class + ``@pytest.mark.asyncio`` on each async test method. The reviewer also flagged the file's five separate ``from opencontractserver.constants.tools import …`` blocks. Confirmed non-actionable: this project's ``isort`` settings (no ``combine_as_imports``) split ``as`` aliases into their own statements, so consolidating fails ``pre-commit``. Leaving the structure as-is. 53 tests pass after this change. * Address review #1675: raise on scope mismatch, slim list_fieldsets, lift coverage - _resolve_target_document_ids now raises ValueError when a document agent's injected document_id is not in the corpus, instead of silently expanding to the full corpus scope (which a single-doc agent should never reach). Test updated to assert the new contract. - list_fieldsets returns column names by default; full column definitions (query/match_text/output_type/instructions) gated behind a new include_columns=true flag. Bounds the discovery payload for corpora with many fieldsets of many long-form columns. Tool description and tests updated. - list_analyzers ordered by -created (was order_by("id"), lexicographic over CharField IDs and unpredictable as new analyzers are registered). - start_extract / start_analysis descriptions now warn that omitting document_ids on a corpus agent dispatches against the whole corpus, so admins working over very large corpora know to bound it. - Added tests: limit pagination clamp end-to-end for list_fieldsets and list_analyzers; all four _extract_status branches (queued / running / completed / failed) in list_recent_extracts. - Removed redundant @pytest.mark.django_db on TestExtractAnalyzerRegistryIntegration (TransactionTestCase already provides DB isolation). - CHANGELOG: rewrote the "extract/analyzer review follow-ups" entry to describe the behaviour changes rather than the review process. * Address review #1675: lift CRUD grant into the analyzer/extract framework Mirrors the existing create_and_setup_analysis chokepoint with a new create_and_setup_extract helper so every Extract created through a user-attributable flow gets guardian CRUD on the creator without the implementer having to remember. start_extract (agent tool) and process_corpus_action's get_or_create path now route through (or inline) the framework grant. Docstrings on both helpers pin the contract so a future refactor can't quietly drop it. Other review follow-ups in the same sweep: - Add `created` to list_recent_analyses output so an LLM polling right after start_analysis sees a timestamp instead of three nulls. - list_fieldsets builds rows with stable key order (common fields first, payload last) for readability. - Consolidate redundant `from opencontractserver.constants.tools` blocks. - Clarify list_analyzers description: the corpus_id is a visibility check, not a corpus-affinity filter. - Update include_columns description to list extract_is_list. - _get_user_or_none gets a docstring explaining why the return type stays inferred (User = get_user_model() is a runtime variable, not a valid mypy type). Tests: - test_create_and_setup_analysis_grants_creator_crud — pins the Analysis chokepoint's CRUD contract. - test_extract_utils.py — full coverage for the new Extract helper (creation, CRUD grant, document linking, started timestamp, corpus_action lineage, name defaulting). Closes #1674 --------- Signed-off-by: JSIV <5049984+JSv4@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
5 tasks
This was referenced May 22, 2026
JSv4
added a commit
that referenced
this pull request
May 25, 2026
…Markup note Addresses the latest claude-bot review on PR #1781: #1 — overflowMenuItems.ts audit drift fixed - Earlier commit added "About cite → /about" to the audit comment but did not add the link to OVERFLOW_MENU_LINKS, leaving the comment misleading. Add the About entry to the array so the always-on overflow stays in parity with the Footer (essential links remain reachable from long-scroll views where the in-flow Footer is unreachable). - tests/NavMenu.ct.tsx: assert the About menuitem in both desktop + mobile, anonymous + authenticated. Rewrite the arrow-key wrap tests to walk every item dynamically so future audit additions or removals don't require touching them again. #3 — renderInlineMarkup index-as-key annotated - The reviewer flagged the bare index key as fragile. Reality: the segment array is the deterministic output of `String.split(regex)` on static JSON content, so positions never reorder. Added a comment explaining why composite keys would be a stable-for-stable swap rather than an improvement. #4 + #5 — magic colour / font-family literals extracted to OS_LEGAL_* - OS_LEGAL_COLORS gains the cite-brand neutrals: `ink` (#0f172a, the navy primary button background), `inkHover` (#1e293b, paired hover state — same value as textPrimary but kept as a semantic alias so future divergence is cheap), `warmPaper` (#FAFAF7, brand page background). Each carries a docstring explaining intent vs raw slate-scale position. - OS_LEGAL_TYPOGRAPHY.fontFamilySerif upgraded from the legacy Georgia-first stack to the cite Source Serif 4 / Source Serif Pro chain (Georgia stays in the fallback list — strictly additive, no regression for existing serif consumers). This is the reviewer's suggested fix path: update the constant once, not the call sites six times. - Refactored every touched component to reference the tokens instead of inlining the values: CallToAction.tsx (2 colors, 2 serif stacks) CookieConsent.tsx (6 warmPaper, 2 ink + 2 inkHover, 1 accent, 1 sans stack, 2 serif stacks) Footer.tsx (1 serif stack) Login.tsx (1 accent, 1 ink, 1 inkHover, 1 warmPaper) NavMenu.tsx (CiteMark bracketColor + nodeColor via tokens, navbar brand-name override via fontFamilySerif) NewHeroSection.tsx (2 serif stacks) About.tsx (4 serif stacks) CiteWordmark.tsx (variant fill ternary now uses textPrimary / warmPaper; inline SVG text style uses fontFamilySerif) renderInlineMarkup.tsx (1 serif stack) - CiteWordmark.test.tsx: compare rendered fill against OS_LEGAL_COLORS.{textPrimary, warmPaper} rather than typed hex literals so brand-palette edits never require a parallel test edit. #6 (font dedup) and #7 (CiteMark decorative aria) noted only — the sans fallback chain was retained verbatim per the CookieConsent Windows-rendering case, and the empty-ariaLabel → aria-hidden path is already correct (covered by an existing unit test). PR description updated separately to keep VERSION_TAG = v3.0.0.rc1 and reframe this PR as the rebrand release-candidate (not the GA cut) — matches the constants.ts comment and the releaseScreenshot calls that lock the v3.0.0.rc1 surface in amber.
JSv4
added a commit
that referenced
this pull request
May 25, 2026
…1781) * brand(v3): rebrand to [cite] — landing, about, and chrome wordmark Brings in the cite brand system (handoff in /home/jman/Downloads/ cite-rebrand-handoff): the project ships v3 as cite — the citation layer underneath the public record. Scope is the public surface — assets, landing, About, and the wordmark/favicon/meta that flow from them. The deeper product UI chrome and the wider OpenContracts naming inside the code stay untouched in this pass. Changes: - Add CiteMark + CiteWordmark React components (inline SVG, font fallback chain) plus the source SVGs under src/assets/brand/. - Swap PNG favicon for SVG icon mark (favicon.svg, favicon-32.svg, favicon-16.svg) — modern browsers pick up SVG, .ico kept as alternate fallback. Update manifest.json + index.html meta and JSON-LD to use the cite name, taglines, and cite.opensource.legal URL. Switch theme color to warm paper #FAFAF7. - Pull Source Serif 4 in alongside Inter from Google Fonts. - Bump VERSION_TAG from v3.0.0.rc1 → v3.0.0 (rebrand is the v3 GA milestone per migration plan). - DiscoveryLanding hero: rewrite as "[•] The citation layer / underneath the public record." with Source Serif headline, teal second line, and the cite mark at cap-height. Subhead, search placeholder, GetStarted action list, and StatsSection labels follow the home_page.md copy. - CallToAction: replace the gradient/Rocket "Ready to dive in?" block with an editorial-voice About-cite paragraph and a quiet sign-in / read-more pair, per the brand voice rules. - /about: new route + page rendering the four-section long-form copy from about.md (Why cite exists / Why it's broken / What cite is / Why we think we can do this). Linked from the new footer. - NavMenu: switch the os_legal_128 PNG logo for the inline cite icon mark; restyle the brand-name slot to Source Serif 4 at 22px with -0.5px tracking so [cite] reads as a typographic wordmark instead of a UI label. - Footer: replace the os_legal_128 PNG with the stacked opensource.legal · [cite] lockup; tighten the copy and add About + GitHub links. - Login: drop the legacy PNG/turquoise theme for the cite icon mark + Source Serif title; navy primary button. - Component tests: update NavMenu Branding and CallToAction assertions to the new wordmark / copy. Refresh the landing--call-to-action--anonymous doc screenshot. Verification: tsc --noEmit + vite build clean; targeted Playwright component tests pass (CallToAction Component x2, NewHeroSection x2, NavMenu Branding x3). Closes #N/A (rebrand prep). * docs(cite): README rebrand, new About/Login CT screenshots, RC1 version - README.md: rewrite around the cite framing while keeping the GitHub repo name OpenContracts (per migration plan — repo rename is a coordinated cut after launch). Lead with the opensource.legal · [cite] lockup, restate the "why cite / why broken / what cite is" arc, add a domain-migration table, and fix the cite product tagline throughout. Add docs/assets/images/brand/ with the wordmark, lockup, icon mark, and inline marker SVGs so the README and other docs can reference the production marks. - v3.0.0.rc1: revert VERSION_TAG from v3.0.0 back to v3.0.0.rc1 — the rebrand ships as RC1, GA drops the suffix. Bump the release screenshot key to v3.0.0.rc1 to match. Release-screenshot regex (^v\d+\.\d+\.\d+) still accepts the suffixed form. - DiscoveryLanding CT: replace the now-stale "The open platform for" assertion with the new cite hero text + capture landing--discovery-page--anonymous.png (regenerated) and the v3.0.0.rc1 release landing-page snapshot. - New CT tests + doc screenshots: * NewHeroSection close-up → landing--hero--cite-headline.png * GetStarted card with cite [•] bullets → landing--get-started--cite-bullets.png * Full About page (asserts headings + named incumbents) → about--full-page--anonymous.png → releases/v3.0.0.rc1/about-page.png * Login card (cite mark + wordmark + tagline) → login--cite-brand--anonymous.png → releases/v3.0.0.rc1/login-page.png Verification: tsc --noEmit clean; 17/17 rebrand-related Playwright component tests pass (CompactLeaderboard ×3, CallToAction ×2, DiscoveryLanding ×1, FeaturedCollections ×3, NewHeroSection ×3, GetStarted ×2, About ×2, Login ×1). * brand(cite): rebrand CookieConsent modal — beautiful on mobile Every information element from the OpenContracts-era cookie modal is preserved verbatim (demo-system caveat, cookie-usage explainer, data we collect, data you agree to share, analytics breakdown when PostHog is configured, MIT-style warranty disclaimer). The visual language is rewritten against the cite brand system. Brand alignment: - Source Serif 4 modal title ("Cookies and terms" — sentence case) with a small [•] PRIVACY eyebrow above it that puts the brand atom on the surface that introduces the project. - Warm paper #FAFAF7 throughout (header, body, footer) instead of pure white; thin slate borders define structure. - Demo banner: drops the warning-yellow #fefce8/#854d0e palette (off-brand — teal is the only saturated color) for a quiet slate-on-paper card with a 3px teal left stripe and slate AlertTriangle icon. Same information, matter-of-fact voice. - Lead paragraph set in Source Serif with the word *cite* italicized, per the brand voice rule. - Section eyebrows in muted slate, 10px / 0.6px tracking, sentence case ("Cookie usage", "Data we collect", "Data you agree to share") — within the brand's "ALL CAPS only in eyebrow labels" carve-out. - Cards: white, 1px slate border, 6px radius, no shadow on desktop (brand: shadows subtle to absent). On mobile, list rows pick up a surfaceLight tint so each item stays scannable. - Disclaimer: ALL CAPS warranty text is preserved (it's a legal convention, not marketing) but quieted to 12px muted slate inside a brand-correct white card with its own "Warranty disclaimer" eyebrow. - Primary CTA: navy chrome #0F172A button — brand spec puts navy on primary buttons and reserves teal for accent / links / active states. Full-width 48px touch target on mobile. Implementation note: the @os-legal/ui Modal renders via createPortal(content, document.body), so a styled-components wrapper around <Modal> can't reach the portaled DOM with descendant selectors. Replaced StyledModalWrapper with an inline <style> tag plus className="cookie-consent-modal" / overlayClassName="cookie-consent-overlay" props on <Modal> — same pattern NavMenu uses for its .oc-navbar overrides. This is what finally lets the navy button override and Source Serif title actually paint. Tests + screenshots: - CookieConsent.ct.tsx updated for the new copy ("Cookies and terms", "Accept and continue", "Cookie usage", etc.) and the added "Warranty disclaimer" eyebrow. Both desktop and mobile variants now also capture a v3.0.0.rc1 release screenshot. - cookie-consent--modal--default.png and cookie-consent--modal--mobile.png regenerated. - New release screenshots: releases/v3.0.0.rc1/ cookie-consent-modal.png (desktop) cookie-consent-modal-mobile.png (iPhone 13 viewport) Verification: tsc --noEmit clean; 2/2 CookieConsent CT tests pass. * config(landing): JSON-drive Discover + About; ship default and public-record variants The Discover landing page and the /about long-form page are now driven by a typed JSON content pack. Two variants ship in the repo, switchable at runtime via REACT_APP_LANDING_VARIANT: - default → "The citation layer for agentic workflows." World-facing pitch aimed at developers, AI labs, research teams, and anyone landing on the repo or self-hosting cite against their own corpus. This is the repo default and the framing used in README, index.html OG tags, and the bundled Discover/About screenshots. - public-record → "The citation layer underneath the public record." End-user pitch with the named-incumbent argument (Westlaw, Lexis, JSTOR, USPTO, Wheaton v. Peters, Public.Resource.Org, Free Law Project). Lands well on cite.opensource.legal where the audience is end-users of a specific corpus of public-domain documents. A deployer flips between them by setting REACT_APP_LANDING_VARIANT in frontend/public/env-config.js — no rebuild, no fork. Unknown variants fall back to default rather than crashing. Adding a new pitch is a three-step process documented in README: drop a JSON file in src/config/landingContent/, register it in index.ts, set the env var. Architecture: - src/config/landingContent/ types.ts — typed schema (HeroContent, StatConfig, GetStartedContent, CallToActionContent, AboutContent + footer links). Stat keys are constrained to a CommunityStatKey union so a typo in a variant fails at build. default.json — world-facing pitch. publicRecord.json — JSv4's deployment pitch. index.ts — variant registry + useLandingContent() hook backed by useEnv().REACT_APP_LANDING_VARIANT with useMemo memoization. renderInlineMarkup — tiny inline-italic renderer that wraps `*…*` spans in Source Serif italic <em>. The brand rule restricts italic to (a) the cite product name and (b) publication/platform names in body copy — both are flat inline spans, so a full Markdown parser would be overkill. __tests__/ — vitest checks both bundled variants pass the type contract, diverge on hero copy (guards against a copy-paste collapse), and reference only stable GraphQL stat keys. Component refactor: - NewHeroSection, StatsSection, GetStarted, CallToAction, About all read from useLandingContent() and stop hard-coding copy. The components stayed visually identical — same layout, same brand treatment — only the strings/lists are now external. Italics inside body copy route through renderInlineMarkup so the *cite* product name picks up its Source Serif italic treatment automatically. Tests: - landing-components.ct.tsx and About.ct.tsx import the same JSON the components do and assert against `defaultLandingContent.hero.primary`, `…callToAction.headline`, every Get Started action label, every About section heading, etc. Future copy edits to the default JSON keep the tests green automatically. Screenshots: - landing--discovery-page--anonymous.png, landing--hero--cite-headline.png, landing--call-to-action--anonymous.png, about--full-page--anonymous.png regenerated to show the world-facing pitch (the repo default). - releases/v3.0.0.rc1/landing-page.png and about-page.png replaced (write-once was holding the old public-record framing — rc1 ships the world-facing default, so the snapshots needed to match). Docs + meta: - README rewritten with the "Why cite" arc in the agentic-workflows voice. New "Customizing the landing and About copy" section documents the variant mechanism and the three-step recipe for adding deployment-specific pitches. - index.html <title>, OG tags, and JSON-LD updated to the repo default ("citation layer for agentic workflows"). JSv4's deployment will need to swap these too on the production build (separate concern — meta is build-time-baked, not runtime). - EnvConfig gains REACT_APP_LANDING_VARIANT (default "default") so the type system enforces the contract for deployers reading from window._env_. Verification: tsc --noEmit clean; 6/6 new vitest variant checks pass; 17/17 Playwright CT tests pass (CompactLeaderboard ×3, CallToAction ×2, DiscoveryLanding ×1, FeaturedCollections ×3, NewHeroSection ×3, GetStarted ×2, About ×2, Login ×1). * copy(landing): emphasize human/agent ground-truth collaboration; speak to bots too Tweaks the default (world-facing) landing + About content pack so the human/agent collaboration loop is the lede, not a subordinate clause — and so an LLM crawling the surfaces actually finds the affordances it needs. About page: - Title: "The citation layer that agents and humans can both read." - Lede sharpened around the ground-truth framing: "Humans curate the ground truth, agents reason over it, and the graph compounds." - "What cite is" extended with the protocol-surface line that answers "how do I use this?" in one breath: "Same graph, two interfaces: a GraphQL and REST API for humans and applications, a Model Context Protocol endpoint for agents. The substrate is identical; only the surface you call differs." - New "How the graph gets built" section between "What cite is" and "Who it's for" — describes the three-step loop concretely: Humans curate. → Agents reason. → Humans review. → (compound) Closing fourth paragraph addresses an LLM reader directly so the capability information survives chunking and shows up cleanly in retrieval: "If you are an LLM-based agent reading this: the corpus is exposed to you over Model Context Protocol at /mcp/. You can list documents, search annotations, follow citation edges across the graph, and propose new edges by writing annotations of your own. Cite the document and span you pulled from in every answer …" Hero subhead: - Verbs sharpened from abstract ("humans curate / agents reason over") to concrete agent-meaningful action ("humans annotate, agents traverse, both propose new edges"). One ground truth, two collaborators — the framing humans appreciate; verbs an agent's planner can match against its action space. CTA body: - Closing line: "Humans and agents work the same graph; the ground truth is shared." README: - "Why cite" arc mirrors the in-product framing: same graph / two interfaces / agents propose, humans review / graph compounds. - New direct-address blockquote at the end of the section for LLMs parsing the README, naming /mcp/, /llms.txt, and /.well-known/mcp.json so a crawler finds the discovery surfaces in one hop. Screenshots: - landing--discovery-page--anonymous.png, landing--hero--cite-headline.png, landing--call-to-action--anonymous.png, about--full-page--anonymous.png regenerated for the sharpened copy. - releases/v3.0.0.rc1/landing-page.png + about-page.png replaced (write-once was holding the prior wording — rc1 ships this copy). Verification: tsc --noEmit clean; 6/6 vitest variant checks pass; 16/16 Playwright CT tests in the rebrand suite pass. * test: cover cite-brand components and Footer for codecov patch target The PR introduced CiteMark, CiteWordmark, renderInlineMarkup, and the new Footer lockup with no direct test coverage, dragging codecov/patch to 79.27% (auto target). Add focused unit + CT tests: - src/components/brand/__tests__/CiteMark.test.tsx — strokeFor branches for all four size bands, prop pass-through, aria semantics. - src/components/brand/__tests__/CiteWordmark.test.tsx — both variants, default sizing, prop pass-through. - src/config/landingContent/__tests__/renderInlineMarkup.test.tsx — empty, plain, single italic, multi-italic interleaved with text. - tests/Footer.ct.tsx + tests/FooterTestWrapper.tsx — three viewport breakpoints (>1000px, compact, <400px) exercising the $compact and $small ternaries plus two new docScreenshots (layout--footer--full-width, layout--footer--compact). Test wrapper split is required because Playwright CT cannot mount components defined inside the test file (per CLAUDE.md split-import rule). MemoryRouter wrap is needed because the CT root provides Jotai/Apollo/Theme but no router context. * review: address PR #1781 feedback (broken /contact, AGPL→MIT, a11y, dead code) Addresses 10 of the issues raised across the three claude-bot reviews: Bugs / broken refs - Footer.tsx: drop the broken /contact Link (no route registered in App.tsx). overflowMenuItems.ts audit comment refreshed to reflect that the Footer now matches the overflow's omission. - index.html: JSON-LD license fields flipped from AGPL to MIT to match the README + CookieConsent copy. - index.html: og:title now reads "[cite]" with brackets to match the brand spec (the wordmark IS the brackets). - index.html: SVG <link rel="icon"> entries switched from pixel sizes="32x32" etc. to sizes="any" (the spec-correct value for scalable formats). - index.html: dead apple-touch-icon link to the no-longer-shipped /logo192.png removed; comment notes brand-correct PWA icons are intentionally deferred. Accessibility - CiteMark.tsx: when ariaLabel is empty (decorative usage in GetStarted, CallToAction, About eyebrow, CookieConsent), the SVG now renders aria-hidden="true" and drops role="img" so screen readers skip it instead of announcing it as an unlabelled image. New unit test pins the behavior. Dead code / variant coupling - Login.tsx: replace the hardcoded styled <Title>[cite]</Title> + variant-specific subtitle with the existing CiteWordmark component + variant-neutral "Sign in to continue." copy. This both wires up the previously-unused CiteWordmark and stops the Login screen from shipping public-record copy regardless of REACT_APP_LANDING_VARIANT. - src/assets/images/os_legal_128*.png: delete three unused old-brand PNGs (no remaining imports in src/). Test updates - Login.ct.tsx: assert against the new CiteWordmark SVG + neutral subtitle instead of the removed <Title> + public-record tagline. Docs - CHANGELOG.md: add an "Added" entry under [Unreleased] for the v3 cite surface rebrand, including the explicit out-of-scope list (codebase-wide string scrub, PWA install icons, OG image refresh). * brand: generate PWA + OG PNG assets for the cite rebrand The v3 rebrand stripped the old logo192.png / logo512.png from the public dir without replacements; the resulting manifest only carried the SVG favicon, which is not reliably accepted by Chrome for Android's "Add to Home Screen" install criteria, and the OG/Twitter card still linked to the OpenContracts product screenshot. - frontend/scripts/generate-brand-pngs.js (new): renders the cite icon mark SVG through headless Chromium at the exact target pixel dimensions. Chosen over librsvg / sharp / inkscape because Playwright is already a dev dependency for the CT suite — no new install footprint. Outputs four files: public/cite-192.png (192×192, transparent, purpose: any) public/cite-512.png (512×512, transparent, purpose: any) public/cite-maskable.png (512×512, brand bg, mark inside the central 80% safe area for Android adaptive-icon shape masking) public/OpenContractsScreenshot.png (1200×630 OG card with the cite wordmark + public-record tagline + opensource.legal eyebrow; legacy filename retained so the existing <meta og:image> reference resolves to the new card) - frontend/public/manifest.json: add the three PNG icon entries alongside the existing SVG and ICO, with the correct purpose:any / purpose:maskable split. - frontend/index.html: re-introduce <link rel="apple-touch-icon"> pointing to cite-192.png so iOS picks up a brand-correct home screen icon (iOS does not reliably support SVG apple-touch-icon). Replace the deferred-PWA-icons comment with one that points readers at the generator script. - CHANGELOG.md: drop the deferred-PWA-icons / OG image bullets from the "Out of scope" list and call out the generator + new assets under a dedicated "PWA + OG assets" bullet under the rebrand entry. * review: extract cite-brand tokens + fix overflow audit + renderInlineMarkup note Addresses the latest claude-bot review on PR #1781: #1 — overflowMenuItems.ts audit drift fixed - Earlier commit added "About cite → /about" to the audit comment but did not add the link to OVERFLOW_MENU_LINKS, leaving the comment misleading. Add the About entry to the array so the always-on overflow stays in parity with the Footer (essential links remain reachable from long-scroll views where the in-flow Footer is unreachable). - tests/NavMenu.ct.tsx: assert the About menuitem in both desktop + mobile, anonymous + authenticated. Rewrite the arrow-key wrap tests to walk every item dynamically so future audit additions or removals don't require touching them again. #3 — renderInlineMarkup index-as-key annotated - The reviewer flagged the bare index key as fragile. Reality: the segment array is the deterministic output of `String.split(regex)` on static JSON content, so positions never reorder. Added a comment explaining why composite keys would be a stable-for-stable swap rather than an improvement. #4 + #5 — magic colour / font-family literals extracted to OS_LEGAL_* - OS_LEGAL_COLORS gains the cite-brand neutrals: `ink` (#0f172a, the navy primary button background), `inkHover` (#1e293b, paired hover state — same value as textPrimary but kept as a semantic alias so future divergence is cheap), `warmPaper` (#FAFAF7, brand page background). Each carries a docstring explaining intent vs raw slate-scale position. - OS_LEGAL_TYPOGRAPHY.fontFamilySerif upgraded from the legacy Georgia-first stack to the cite Source Serif 4 / Source Serif Pro chain (Georgia stays in the fallback list — strictly additive, no regression for existing serif consumers). This is the reviewer's suggested fix path: update the constant once, not the call sites six times. - Refactored every touched component to reference the tokens instead of inlining the values: CallToAction.tsx (2 colors, 2 serif stacks) CookieConsent.tsx (6 warmPaper, 2 ink + 2 inkHover, 1 accent, 1 sans stack, 2 serif stacks) Footer.tsx (1 serif stack) Login.tsx (1 accent, 1 ink, 1 inkHover, 1 warmPaper) NavMenu.tsx (CiteMark bracketColor + nodeColor via tokens, navbar brand-name override via fontFamilySerif) NewHeroSection.tsx (2 serif stacks) About.tsx (4 serif stacks) CiteWordmark.tsx (variant fill ternary now uses textPrimary / warmPaper; inline SVG text style uses fontFamilySerif) renderInlineMarkup.tsx (1 serif stack) - CiteWordmark.test.tsx: compare rendered fill against OS_LEGAL_COLORS.{textPrimary, warmPaper} rather than typed hex literals so brand-palette edits never require a parallel test edit. #6 (font dedup) and #7 (CiteMark decorative aria) noted only — the sans fallback chain was retained verbatim per the CookieConsent Windows-rendering case, and the empty-ariaLabel → aria-hidden path is already correct (covered by an existing unit test). PR description updated separately to keep VERSION_TAG = v3.0.0.rc1 and reframe this PR as the rebrand release-candidate (not the GA cut) — matches the constants.ts comment and the releaseScreenshot calls that lock the v3.0.0.rc1 surface in amber. * review: address PR #1781 polish — a11y, keys, memoization, variant tests About.tsx - composite-index keys for sections + paragraphs (no longer rely on section.title uniqueness) - aria-label on external footer links so screen readers announce the new-tab navigation CookieConsent.tsx - DemoBanner gets role="note" + aria-hidden on the decorative icon (role="alert" would double-announce on top of the dialog open event) - corrected the CSS-scoping comment block — selectors are qualified by cookie-consent-modal / cookie-consent-overlay classes, not the stale data-cookie-consent attribute that was never wired up NewHeroSection.tsx - hardcoded 1023px / 767px breakpoints replaced with the shared TABLET_LANDSCAPE_BREAKPOINT / TABLET_BREAKPOINT constants CiteMark.tsx, CiteWordmark.tsx - wrap in React.memo so the chrome (NavMenu, Footer, About, Login, CTA) doesn't re-render SVGs on unrelated parent state changes renderInlineMarkup.tsx - reframe the index-as-key comment around the real invariant (static immutable JSON content) instead of "split() is deterministic" - re-export renderInlineMarkup from landingContent/index.ts so callers can import from the package root generate-brand-pngs.js - hoist BRAND_COLORS, MASKABLE_FRAME, MASKABLE_SAFE_AREA, OG_WIDTH, OG_HEIGHT to the top so a future brand-color or spec change is a one-line edit useLandingContent.test.tsx (new) - direct hook coverage: default-variant fallback when env is unset, public-record selection, variant switching diverges on hero/about copy, unknown-key falls back to default * test(useLandingContent): mock useEnv instead of (window as any)._env_ The previous version cast through `any` to satisfy the strict ambient `Window._env_` type. That breached the project's any-baseline gate (scripts/check-any-baseline.js) by +7. Switch the test to vi.mock the `useEnv` hook directly — same coverage, no `any` cost, no global window manipulation. Mirrors the pattern in utils/__tests__/analytics.test.ts. * style: prettier-format useLandingContent.test.tsx --------- Signed-off-by: JSIV <5049984+JSv4@users.noreply.github.com>
JSv4
added a commit
that referenced
this pull request
May 25, 2026
- Replace hardcoded ``z-index: 2`` in NativeLinkLayer with new ``Z_INDEX.PDF_NATIVE_LINK_LAYER`` constant + sibling ``PDF_SELECTION_LAYER`` entry, removing the magic-number rule violation flagged in review. - Switch ``NativeLinkLayer`` from ``React.FC`` (which the project's ``"jsx": "react-jsx"`` transform leaves out of scope unless React is explicitly imported) to a named ``FC`` import from "react" — same pattern recommended by review. - Drop the ``as any`` cast in NativeLinkLayer.test.tsx for a proper ``as unknown as PDFPageProxy`` so the any-baseline lint check is back to clean. - Slim down the CHANGELOG entry per review minor #4 — the existing entry doubled as a design doc; the format guidance asks for a short summary with file paths + a coverage pointer. Also drop the duplicate "Discovery endpoints rebranded" entry that local-main drift surfaced here (PR #1785 already owns that entry on main).
This was referenced May 25, 2026
JSv4
added a commit
that referenced
this pull request
May 25, 2026
…on bug Addresses Claude review on #1790 (plus restoring the back button that the PR accidentally hid): Critical (back button gone for sidebar discussion view): - ThreadDetail dropped its compact BackButton whenever ``customOnBack`` was passed. But ``customOnBack`` is the back-navigation HANDLER the parent supplies, not a signal that the parent renders its own button. ``ThreadDetailWithContext`` delegates click handling without showing its own back affordance, so the new gate stranded the user with no way to return to the list — and broke the threads-discussions.spec.ts e2e flow (the "Back to discussions" button locator timed out after 3 retries). Reverted to ``compact &&`` so the button is always shown in compact mode; the handler still respects customOnBack. CLAUDE.md rule #4 (no magic numbers): - MessageBadges.tsx hoisted TOOLTIP_HEIGHT_EST + TOOLTIP_MAX_WIDTH to module scope, swapped its local ``GAP = 8`` and ``VIEWPORT_PADDING = 8`` for the existing shared ``POPOVER_GAP`` and ``CONTEXT_MENU_VIEWPORT_PADDING`` from ``frontend/src/assets/configurations/constants.ts``. Stacking-context bug: - TooltipPopup z-index was 1000 — below ModernContextMenu (10000) and the canonical POPOVER_Z_INDEX (100002). Switched to POPOVER_Z_INDEX so the portalled tooltip outranks context menus and modal overlays. Stale-read race: - The JSX was computing ``bottom: window.innerHeight - tooltipPos.top`` at render time, racing with a resize that could fire between computePosition and the next paint (1-frame misalignment). Capture both ``top`` and ``bottom`` inside the effect and read them straight through in the JSX. Orphan tooltip: - When the badge scrolled fully out of the viewport the tooltip stayed fixed at its last computed coords. ``computePosition`` now early- returns + clears state when ``rect.bottom < 0 || rect.top > window.innerHeight``. Test fixture typo: - "What de y'all think?" -> "What do y'all think?" in BadgeClipHarness and ThreadDetailSidebarLayout.ct (both fixture and assertion). Verified: - yarn tsc --noEmit: clean - yarn lint: clean - 8/8 CT (ThreadDetail* + DocumentDiscussionsContent + BadgeClip)
JSv4
added a commit
that referenced
this pull request
May 25, 2026
…#1790) * sidebar discussion view: portal badge tooltip + rework cramped header The badge tooltip in the sidebar discussion view was clipped by the overflow:hidden chain (ContextSidebarContainer / DocumentDiscussionsContent Container / FlexColumnPanel). Render TooltipPopup via createPortal to document.body with fixed coords derived from the badge rect; viewport-edge clamping plus a flip-below-the-badge fallback prevents the same clip near the top of any scroll container. Arrow stays anchored under the badge centre via a --arrow-offset CSS variable. While in the same view, the ThreadDetail compact header packed back-arrow + type badge + title + description + document link + status pills into one flex-wrap row, and a duplicate compact back button shadowed the parent's "Back to List". Reworked so each piece gets its own line (title row / description / context link / meta row separated by a dashed border) and the redundant back button is suppressed when customOnBack is provided. Added Playwright CT coverage (ThreadDetailSidebarLayout.ct.tsx + harnesses) that pins both the new header layout and the no-clip tooltip behaviour, with auto screenshots for the docs surface. * Address review feedback (#1790) - MessageBadges tooltip: re-position on capture-phase scroll + resize while the tooltip is visible. ``position: fixed`` doesn't follow the badge when an ancestor scroll container moves under it; without this listener the tooltip drifts to the wrong place when the sidebar scrolls while the cursor is held still over a badge (review item #1). - Replace the ``["--arrow-offset" as any]`` style cast with a ``Record<string, string>`` spread so the any-baseline gate stays honest (was the new offender that pushed current=438 baseline=437). - ThreadDetailSidebarLayout test: assert structurally that the tooltip is NOT a descendant of the clip-frame (``el.closest('[data-testid=clip-frame]') === null``). The previous "nonzero width/height" assertion could pass even if the portal failed to escape the clip frame (review item: test). * Address PR review on #1790: constants, z-index, stale read, back-button bug Addresses Claude review on #1790 (plus restoring the back button that the PR accidentally hid): Critical (back button gone for sidebar discussion view): - ThreadDetail dropped its compact BackButton whenever ``customOnBack`` was passed. But ``customOnBack`` is the back-navigation HANDLER the parent supplies, not a signal that the parent renders its own button. ``ThreadDetailWithContext`` delegates click handling without showing its own back affordance, so the new gate stranded the user with no way to return to the list — and broke the threads-discussions.spec.ts e2e flow (the "Back to discussions" button locator timed out after 3 retries). Reverted to ``compact &&`` so the button is always shown in compact mode; the handler still respects customOnBack. CLAUDE.md rule #4 (no magic numbers): - MessageBadges.tsx hoisted TOOLTIP_HEIGHT_EST + TOOLTIP_MAX_WIDTH to module scope, swapped its local ``GAP = 8`` and ``VIEWPORT_PADDING = 8`` for the existing shared ``POPOVER_GAP`` and ``CONTEXT_MENU_VIEWPORT_PADDING`` from ``frontend/src/assets/configurations/constants.ts``. Stacking-context bug: - TooltipPopup z-index was 1000 — below ModernContextMenu (10000) and the canonical POPOVER_Z_INDEX (100002). Switched to POPOVER_Z_INDEX so the portalled tooltip outranks context menus and modal overlays. Stale-read race: - The JSX was computing ``bottom: window.innerHeight - tooltipPos.top`` at render time, racing with a resize that could fire between computePosition and the next paint (1-frame misalignment). Capture both ``top`` and ``bottom`` inside the effect and read them straight through in the JSX. Orphan tooltip: - When the badge scrolled fully out of the viewport the tooltip stayed fixed at its last computed coords. ``computePosition`` now early- returns + clears state when ``rect.bottom < 0 || rect.top > window.innerHeight``. Test fixture typo: - "What de y'all think?" -> "What do y'all think?" in BadgeClipHarness and ThreadDetailSidebarLayout.ct (both fixture and assertion). Verified: - yarn tsc --noEmit: clean - yarn lint: clean - 8/8 CT (ThreadDetail* + DocumentDiscussionsContent + BadgeClip) --------- Co-authored-by: Claude <noreply@anthropic.com>
JSv4
added a commit
that referenced
this pull request
May 25, 2026
* Make native PDF link annotations clickable in the viewer Embedded hyperlinks in PDFs were unreachable because the SelectionLayer overlay sits at z-index 1 across the entire page and intercepts every mouse event. PDF.js painted the link rects into the canvas as pixels, but page.getAnnotations() was never called, so no DOM target existed to receive the click. New NativeLinkLayer extracts Link annotations from each page, filters to safe protocols (http/https/mailto), projects each rect through the viewport, and renders transparent <a> elements at z-index 2 (above SelectionLayer, below the SelectionActionMenu portal). The overlay container is pointer-events: none so empty space still falls through to the selection logic; each anchor is pointer-events: auto so the browser handles the click (with hover state, right-click menu, and keyboard focus). Internal dest-based navigation is not yet wired. * Address review feedback (#1788) - Replace hardcoded ``z-index: 2`` in NativeLinkLayer with new ``Z_INDEX.PDF_NATIVE_LINK_LAYER`` constant + sibling ``PDF_SELECTION_LAYER`` entry, removing the magic-number rule violation flagged in review. - Switch ``NativeLinkLayer`` from ``React.FC`` (which the project's ``"jsx": "react-jsx"`` transform leaves out of scope unless React is explicitly imported) to a named ``FC`` import from "react" — same pattern recommended by review. - Drop the ``as any`` cast in NativeLinkLayer.test.tsx for a proper ``as unknown as PDFPageProxy`` so the any-baseline lint check is back to clean. - Slim down the CHANGELOG entry per review minor #4 — the existing entry doubled as a design doc; the format guidance asks for a short summary with file paths + a coverage pointer. Also drop the duplicate "Discovery endpoints rebranded" entry that local-main drift surfaced here (PR #1785 already owns that entry on main). * Address #1788 review: wire SelectionLayer to constant, reset rawLinks, add aria-label - SelectionLayer was still using a literal `zIndex: 1` even though the PR introduced `Z_INDEX.PDF_SELECTION_LAYER` to make the layering relationship between SelectionLayer (1) and the new NativeLinkLayer (2) explicit. Wire it through so the constant is not dead. - NativeLinkLayer now clears `rawLinks` at the start of the page effect so a future recycled component instance cannot briefly paint the previous page's link positions while the new fetch is in flight. - Transparent anchors now expose `aria-label={link.url}` so screen readers (VoiceOver in particular ignores the `title` attribute as an accessible-name source) get a robust name. - Narrow rect type from `number[]` to a 4-tuple, with an explicit `length === 4` guard before the cast, so the `[x1, y1, x2, y2]` destructuring is type-safe. - Strengthen the unit test mock: `getViewport` now honours the passed `scale` (was hard-coded to ×2), and the test asserts `getViewport` was called with the expected scale. Adds an aria-label regression test. --------- Signed-off-by: JSIV <5049984+JSv4@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
4 tasks
This was referenced May 28, 2026
4 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps responses from 0.21.0 to 0.22.0.
Release notes
Sourced from responses's releases.
Changelog
Sourced from responses's changelog.
Commits
bbc4f78release: 0.22.0b3caf4cAdd docs for reponse file loading(#590)3bb694bUse partialmethod for shortcuts (#583)a24d964Add docs for recording beta features (#582)caca43cremove OriginalResponseShim (#585)d4ace43[BETA] replay responses from toml files. (#581)873ddb4add docstrings to OrderedRegistry (#587)240504cadd docstrings to registry methods (#586)60931f1adjust error message when Retry is exhausted (#580)7318dddUpdaterequestsdependency to the version of 2.22.0 or higher (#584)Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting
@dependabot rebase.Dependabot commands and options
You can trigger Dependabot actions by commenting on this PR:
@dependabot rebasewill rebase this PR@dependabot recreatewill recreate this PR, overwriting any edits that have been made to it@dependabot mergewill merge this PR after your CI passes on it@dependabot squash and mergewill squash and merge this PR after your CI passes on it@dependabot cancel mergewill cancel a previously requested merge and block automerging@dependabot reopenwill reopen this PR if it is closed@dependabot closewill close this PR and stop Dependabot recreating it. You can achieve the same result by closing it manually@dependabot ignore this major versionwill close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)@dependabot ignore this minor versionwill close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)@dependabot ignore this dependencywill close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)