0.2.62 (unreleased)

  • Set TERM=xterm-256color inside tox test environments so local runs launched from Ghostty do not print missing xterm-ghostty terminfo warnings.

  • Point the test settings’ extra static files directory at the shipped src/cast/static tree so tox migrations no longer emit Django’s staticfiles.W004 warning.

  • Allow editor API post and episode draft PATCH requests to send the optimistic concurrency token as a strict If-Match: "<revision id>" header instead of JSON base_revision_id.

  • Avoid deep-copying Podlove and DOTe transcript JSON during public speaker sanitization when all speaker labels are already public or when the transcript contains no speaker metadata.

  • Fix Blog.last_build_date crashing with IndexError on blogs and podcasts without any published posts; it now falls back to the page’s first publication date or the current time, so feeds for empty blogs render.

  • Fix the podcast feed’s RSS <category> element: it now emits the first entry of the blog’s keywords field when keywords are set (the previous guard checked a nonexistent categories attribute and always produced an empty category).

  • Declare explicit permission_classes = (AllowAny,) on the public API views (Podlove detail, player config, facet counts, theme list, theme update) so intended anonymous access no longer depends on the DRF project default and cannot be broken by a host site setting restrictive default permissions.

  • Remove dead tooling files (bootstrap.md, .travis.yml, setup.cfg, runtests.py, javascript/jest.config.js, command_lines.txt); the commands.py test and mypy subcommands now invoke pytest and the configured mypy directly.

  • Stop pinning the pytest-randomly seed so the test order actually randomizes, and fix the two test-isolation bugs this surfaced (a shared API client leaking session cookies between tests, and a monkeypatched USE_THREADEDCOMMENTS baking the wrong comment form base class when the forms module was first imported under the patch).

  • Clean up packaging and type-check metadata: drop the contradictory MIT license classifier and a duplicated classifier, check types against Python 3.11 (the support floor), and remove the legacy mypy django-stubs plugin table.

  • Make Post.get_description side-effect free: the feed/social-card body template is now passed as a rendering argument instead of being stored on the instance, so a post or episode rendered after its description was generated no longer leaks the post_body.html partial into the full page render.

  • Wrap Video.save poster generation in a transaction like Audio.save, so a failed poster derivation no longer leaves a half-saved video row.

  • Add sync_media and create_renditions keyword arguments to Post.save (both default True) so bulk operations and migrations can skip media-derivation side effects explicitly.

  • Move the HtmxHttpRequest typing class to cast.http_types; it stays importable from cast.views and cast.views.htmx_helpers for backwards compatibility.

  • Eliminate per-row queries when building blog index snapshots for mixed post/episode listings: page specialization and episode audio now load in bulk, so the query count no longer grows with the number of episodes, and a regression test guards the flat query count.

  • Deduplicate the audio/video/transcript Wagtail-admin media views: the previously triplicated index/add/edit/delete/chooser/ chosen/chooser_upload views now share one parametrized implementation in cast.views.media. URL names, templates, contexts, messages, and chooser modal workflows are unchanged.

  • Fix the video chooser upload view ignoring the CAST_CHOOSER_PAGINATION setting: it paginated with a hardcoded page size of 10 instead of CHOOSER_PAGINATION like every other chooser view (invisible under the default of 10, wrong for sites that override the setting).

  • Extract the transcript domain into a new cast.transcripts package: the WebVTT/Podlove/DOTe rewrite logic, speaker-sample selection, voice-reference derivation, and known-speaker handling now live in Django-free per-format modules plus a service layer, and the Transcript model keeps its public API as thin delegates. Public names such as time_to_seconds, convert_dote_to_podcastindex_transcript, and the KNOWN_SPEAKER_* constants remain importable from cast.models.transcript.

  • Refactor the transcript admin edit view to dispatch its POST actions through a declarative action→handler map, and deduplicate the episode_from_latest_revision helper that was copied between the transcript and Voxhelm admin views. Messages, redirects, and form behavior are unchanged.

  • Centralize the static CAST_* setting defaults in a single registry in cast.appsettings: the scattered inline getattr(settings, ...) defaults now resolve through one accessor (still read at call time, so override_settings and runtime changes keep working), and the cast.E001 type check derives from the registry instead of a separately maintained table. Defaults and check coverage are unchanged.

  • Document all previously undocumented settings in the settings reference, including CAST_AUDIO_PLAYER, CAST_EDITOR_SCOPES, CAST_POST_BODY_BLOCKS, the comment form CSS-class and moderator settings, and the remaining styleguide tunables.

  • Split the Voxhelm integration into a cast.voxhelm subpackage with dedicated settings (the site → Django setting → environment precedence chain), client (HTTP transport), task_refs, and service (orchestration) modules. The transcript-generation status helpers move to cast.transcripts.generation_status, which removes the circular import between cast.models and the Voxhelm code (a static test now guards that nothing imported during model initialisation depends on cast.voxhelm). Every previously public name remains importable from cast.voxhelm.

  • Show the “Generate transcript” action on the episode and audio edit pages only when Voxhelm is actually configured (API base and key resolvable via site settings, Django settings, or environment variables) for the request’s site. Transcript generation status stays visible regardless, and direct POST requests keep reporting missing configuration as friendly error messages.

  • Deduplicate feeds.py: the shared item_description/item_link/ item_pubdate methods and the Atom stylesheet write() now live in one place, the iTunes/Podcast-Index generator mixins are statically typed (no more # type: ignore or defensive try/except), and podcast feed XML output is unchanged.

  • Blog (non-podcast) RSS and Atom feeds now emit a stable per-entry identifier and publication dates: RSS items gain <pubDate> and a <guid isPermaLink="false"> set to the post’s UUID, and Atom entries gain <updated> and a UUID <id>. Feed readers may show existing blog entries as new once, after which the UUID identifier survives slug and URL changes.

  • Document cast.api.views as a frozen legacy API surface (including the bare-text POST /api/upload_video/ response); new integrations should use the editor API for structured errors, scoped authorization, and If-Match revision handling.

  • Stop shipping test-only settings in the installed package. The former cast.settings module (committed SECRET_KEY, tests.urls root URLconf, MD5 hasher) now lives entirely in tests/settings.py, and dev_settings.py is renamed to dev_tools.py to reflect its role as a feature-flag resolver rather than a settings module. django-environ is dropped from the dependencies. This is an internal packaging change with no effect on configuring a real django-cast site.

  • Give the repository cache serialization an explicit type discriminator instead of guessing a cached object’s class from which keys are present. Blog/podcast and post/episode entries now record their type and deserialize by it, with a fallback that keeps entries cached by earlier versions working.