rick/LEDMatrix
1
0
Fork 0
mirror of https://github.com/ChuckBuilds/LEDMatrix.git synced 2026-10-04 14:25:08 +00:00
Code Issues Packages Projects Releases Wiki Activity
2,101 Commits 10 Branches 20 Tags
c6701ac00d4c16a726d2da80f9ca023c3832823b
Commit Graph
8 Commits
Author SHA1 Message Date
ChuckandClaude Opus 5.5 6bc13a8934 fix(display): Vegas resumes after live priority, and six smaller runtime fixes (#644) 
- Vegas: a live-priority pause was only lifted from inside run_frame(),
  which returns before that check while paused, so the ticker never came
  back until a restart. run_iteration() now resumes it (the controller
  only calls it when nothing preempts Vegas); start()/stop() clear the
  pause state. Iteration length is timed with the monotonic clock.
- Dim schedule: a per-day disabled day now updates the minute-gate cache,
  so brightness no longer flips back to dim within each minute.
- On-demand: a second request no longer overwrites the rotation resume
  index with the first request's mode.
- Render pipeline: reset() drops the prepared group and deferred queue,
  and a prefetch in flight across a reset discards its result.
- Sync: stop() removes the status file (and the controller's cleanup now
  calls it), standalone removes a stale one at startup, and writes use a
  unique mkstemp temp file.
- render_gate.swap_releases_gil() delegates to frame_timing.
- Stale docstrings/comments corrected.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-09-28 08:25:26 -04:00
ChuckandClaude Opus 5.5 7f9c73e9aa fix(vegas): smooth Vegas scroll pacing -- whole pixels per refresh, measured refresh, off-thread preview writes (#628) 
Vegas scrolls a whole number of pixels per panel refresh, locked to SwapOnVSync, against the refresh the panel really holds (measured from swap gaps), instead of blending sub-pixel positions against the refresh cap. The web preview PNG is encoded off the render thread while scrolling, with writes ordered and retried. On hdpi, late frames fell from 6.3% to 0.7%. See docs/SCROLL_PERFORMANCE.md.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-09-24 19:38:22 -04:00
ChuckandClaude Opus 5.5 b9416ef803 fix(display): Vegas teardown and 240s default; remove dead Vegas buffer code (#637) 
* fix(display): tear down Vegas mode on controller cleanup

DisplayController.cleanup() never called VegasModeCoordinator.cleanup(),
so the Vegas teardown (stop, pipeline/stream reset, adapter cache drop)
was unreachable. Call it before the display manager is cleaned up, and
skip it when Vegas was never created.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(vegas): default max_cycle_duration to the documented 240s

The template, the web UI help, CONFIG_REFERENCE and the controller all
say 240, but the code defaulted to 600 in two places, so a config
without the key ran Vegas iterations 2.5x longer than documented.

from_config now falls back to the dataclass field defaults instead of
repeating each one, so the two copies can no longer drift, and the
controller's follower scroll-speed default reads VegasModeConfig's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(display): let run.py -d show display_manager's DEBUG output

display_manager pinned its logger to INFO at import, overriding the root
level, so debug mode never showed its DEBUG lines. Use get_logger() from
src.logging_config like the rest of the core and leave the level to the
logging setup.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(display): run each startup validation check once

StartupValidator.validate_all() ran twice at boot, before and after the
plugin manager was created, so every config, cache, display and
systemd-unit warning was logged twice. The second pass now runs only the
plugin checks. Drop the commented-out raise_on_errors line.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(vegas): one INFO line per plugin-list refresh

StreamManager logged "=" * 60 banners and a line per plugin (INCLUDED,
SKIPPED, FETCHING CONTENT, SEGMENT CREATED) at INFO on every refresh and
fetch, i.e. at each cycle start and every 30s. Log one INFO summary of
the rotation per refresh and move the per-plugin detail, the weighting
breakdown and "no content this cycle" to DEBUG (the adapter still warns
when every content path fails).

Also drop the check/cross marks from the controller's log messages.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(vegas): drop the per-iteration static-mode plugin scan

run_iteration() rebuilt _static_mode_plugins on every iteration, asking
every plugin for its display mode and logging the set at INFO, but
nothing ever read it: static pauses are triggered by
_check_static_plugin_trigger() from the next segment. Delete it, the
coordinator's get_ordered_plugins() that only it used, and the
write-only _static_pause_plugin / _static_pause_start.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(vegas): remove the staging buffer that was never filled

StreamManager and RenderPipeline carried a double-buffer design that
nothing used: _staging_buffer was only ever cleared or swapped, so
swap_buffers() never did anything and should_recompose()'s
staging_count > 0 branch was dead, and _active_scroll_image,
_staging_scroll_image, _is_rendering, _last_frame_time and
_frame_interval were written but never read. Delete the machinery and
rewrite the docstrings around what actually carries updates:
_pending_updates, consumed by process_updates() in swap mode and
invalidate_pending_updates() in continuous mode.

should_recompose() no longer builds a buffer-status dict every frame.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(display): tidy the display controller without changing behaviour

- Import VegasModeCoordinator locally instead of through module globals
  (there is no circular import to avoid).
- Drop hasattr() checks on attributes PluginManager.__init__ always sets
  (plugin_executor, plugin_last_update, get_plugin_lock,
  run_scheduled_updates*, stop_update_worker) and the dead "older
  manager" fallbacks; keep the health_tracker None checks, now via
  _health_tracker().
- Extract _display_once() for the per-frame display call both render
  loops copied, _advance_on_demand() for the two on-demand rotations,
  _reset_on_demand_fields() for the error and clear paths, and
  _timezone() / _in_window() for the two schedule checks.
- Remove always-true conditions and the unreachable non-plugin else
  branch in run(), and read _was_display_active / _last_published_mode /
  vegas_coordinator directly now that __init__ declares them.
- Declare the follower render state in __init__, name its tuning
  constants, add _follower_sign(), and share the 90/s sync send
  interval with the render pipeline (SYNC_SEND_INTERVAL).
- Delete history narration and the "Opt #N" labels, fix the comment
  that called _scroll_speed constant (hot reload updates it), and drop
  a startup timing log that measured nothing.
- render_pipeline / plugin_adapter: read display_manager.width/height
  as the properties they are, drop an empty TYPE_CHECKING block, an
  aliased threading import and a duplicated `if result and
  self.sync_manager:`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(display): trim dead code from display_manager

- Add _new_canvas() for the image/draw/fontmode="1" setup that was
  copied six times.
- Call resolve_double_sided() and compose_pixel_mapper_config() directly
  instead of through a module alias and a passthrough method, and replace
  the comment that said the passthrough read class attributes.
- Delete the unused _initialized flag and _ORIENTATION_ROTATE_DEGREES
  alias (no core or monorepo reader; tests stop resetting the flag), the
  test pattern's unreachable no-matrix branch (it only runs once the
  matrix exists), `del old_image  # help GC` (a no-op on a local), a
  duplicated early return in process_deferred_updates, and stale
  comments.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(vegas): remove unread fields and test-only helpers, fix docstrings

- ContentSegment: drop total_width, fetched_at, is_stale, image_count and
  is_static, none of which is read.
- StreamManager: drop _current_index (never advanced) and the test-only
  get_all_content_for_composition() and has_pending_updates();
  VegasModeConfig: drop the test-only is_plugin_included().
- geometry.find_blank_cut() has had no production caller since the crop
  moved to item boundaries; delete it and its tests.
- PluginAdapter: the _finalize docstring described separator_width
  between every image, and _crop_to_budget's said cuts snap to the
  nearest blank column; both now describe what the code does.
- Coordinator: the static-pause interrupt log no longer blames follower
  mode for every interrupt, and set_update_callback names the callback
  the controller actually wires.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(scroll): correct ScrollHelper comments and drop dead branches

- Four comments said the strip always starts with display_width of
  blank; it does only when lead_gap is None (Vegas passes its own).
- Delete the "Width calculation mismatch" warning: the image is created
  at the calculated width, so the two can never differ.
- Remove the two scroll_delay <= 0 fallbacks (which disagreed with each
  other): set_scroll_delay clamps it to at least 0.001 and nothing in
  core or the plugin monorepo assigns it directly.
- Trim the scipy history from the blend docstring.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(run): drop a redundant comment

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(display): log set_scrolling_state only when it changes

Vegas and scrolling plugins set the scrolling state every frame, so once
display_manager's DEBUG output became visible in debug mode it printed
"Scrolling state set to: True" about 120 times a second. Log only when
the value differs from the previous one; the state, activity timestamp
and frame hold still update on every call.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* docs(changelog): display-vegas

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-09-24 17:37:15 -04:00
ChuckandClaude Opus 5.5 84afa9d64f refactor: delete dead Python code in the core (and stop storing Wi-Fi passwords) (#608) 
* refactor(plugins): remove the no-op PluginHealthMonitor

Its monitor loop did nothing (`if callbacks: pass`), register_health_check
had no callers and api_v3.health_monitor was never read by any route. The
live health data comes from PluginHealthTracker, which is untouched.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(store): drop the never-set uninstall tombstones

Nothing in production called mark_recently_uninstalled, so the
reconciler's was_recently_uninstalled check was always False. The
persistent uninstall registry is what actually stops resurrection; the
reconciler test now exercises that gate instead.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(common): delete unused config/display/game helpers, utils and error_handler

Nothing in core, the web UI, scripts or the plugin monorepo imports
config_helper, display_helper, game_helper, utils or error_handler; only
their own tests did. The error_handler re-exports leave src.common's
__all__; APIHelper, TextHelper, ScrollHelper, LogoHelper and the adaptive
layout exports are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(config): drop ConfigService's unused versioning and save API

ConfigVersion, get_version/get_version_history/get_version_config,
rollback, save_config, reload, get_plugin_config and the backward-compat
load_config/get_config_path/get_secrets_path had no callers. The display
controller only uses get_config, subscribe, unsubscribe and shutdown,
plus the file watcher. Change detection now compares against the
current checksum instead of the last history entry.

The subscriber tests asserted `callback.called or True`; they now
reload the way the watcher does and assert the notification.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(plugins): drop unread plugin state history and callbacks

plugin_state.PluginStateManager kept a bounded per-plugin transition
history that only get_state_history (tests only) read; get_state_info
reports a separate lifetime count, which stays. set_error_info and
record_display had no callers, and set_state_with_error's `error`
argument only fed the history.

The web-side state_manager.PluginStateManager loses
subscribe_to_state_changes, _notify_callbacks, set_plugin_error and
get_state_version, none of which had callers; with no subscribers the
old-state copy in update_plugin_state went with them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(plugins): remove unused PluginManager methods and attribute guards

update_all_plugins was only called by a test (the display loop uses
run_scheduled_updates); get_plugin_health_metrics,
get_plugin_resource_metrics and get_plugin_state had no callers; and
plugin_modules was written but never read. plugin_directories is now
initialised in __init__, so the hasattr() guards around it go.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(plugins): remove unused executor, loader, store and package helpers

- PluginExecutor.execute_safe: no callers.
- PluginLoader._parse_semver: only its own tests; compatibility.parse_semver
  is the live copy and test_compatibility.py already covers it.
- PluginStoreManager.get_installed_plugin_info: no callers.
- PluginResourceMonitor._local: never read.
- src.plugin_system.get_store_manager and __api_version__: no importers in
  core, scripts or the plugin monorepo.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(wifi): stop storing Wi-Fi passwords in wifi_config.json

WiFiManager appended every joined network's SSID and password, in
plaintext, to saved_networks in config/wifi_config.json, and nothing
(web UI, backup restore, scripts) ever read them back: NetworkManager
keeps its own credentials. The writes are gone, and loading the config
now drops any saved_networks key and rewrites the file, so passwords
already on disk are scrubbed.

Also removes _check_dnsmasq_conflict (never called) and _detect_trixie,
whose result only reached one log line, along with the
NM_CONNECTIONS_PATHS constant only it used.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(display): remove unreachable and unused DisplayController code

- _follower_rebuild_scroll_image: never called.
- mode_duration (never read) and last_mode_change (write-only).
- The `chosen_cap <= 0` branch: chosen_cap is either the minimum of
  caps already filtered to > 0 or DEFAULT_DYNAMIC_DURATION_CAP (180).
- The `max_duration < min_duration` branch directly after
  `max_duration = max(min_duration, max_duration)`.
- The circuit-breaker branch's `display_result = False` and
  `manager_to_display = None`: the first is overwritten a few lines
  later, the second is already None there.
- The bool-to-bool conversion of execute_display's result, which is
  always a bool.
- The `loaded_plugins` lookup in _update_modules: PluginManager has no
  such attribute, so it always fell through to `plugins`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(vegas): remove unused config update, boundary finder and refresh

VegasModeConfig.update had no callers outside its own tests (the
coordinator rebuilds the config with from_config on a change);
geometry.find_item_boundary and StreamManager._refresh_plugin_content
had no callers at all.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* refactor(run): drop the debug block that pretended to import the plugin system

In debug mode run.py put src/plugin_system itself on sys.path and printed
"Plugin system import successful" without importing anything. Nothing
imports plugin_system modules by bare name, so the path entry did
nothing either.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test: delete tests that test nothing

- test/plugins/test_{basketball_scoreboard,calendar,clock_simple,
  odds_ticker,soccer_scoreboard,text_display}.py skip everywhere the named
  plugins are not installed, including CI (LEDMATRIX_PLUGINS_DIR holds only
  the fixture plugin); test_plugin_matrix.py already covers every
  discovered plugin. Their PluginTestBase and the fixtures only it used
  (plugins_dir, mock_display_manager, mock_cache_manager,
  mock_plugin_manager, base_plugin_config in test/plugins/conftest.py) go
  with them.
- test_plugin_system.py: test_discover_plugins (body was `pass`) and
  test_dependency_check (a comment), plus the test_plugin_manager fixture
  only the former requested.
- test_display_manager.py: test_draw_image asserted that an image it had
  just assigned was not None.
- test_display_controller.py: the rotation and schedule-override tests
  re-implemented the run-loop arithmetic inline and asserted on their own
  result without calling the controller.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test: expect one plugin_last_update success stamp after update_all_plugins

EveryStampRecordsACompletion required at least two success-path stamps;
the second was update_all_plugins, removed as test-only. The worker and
synchronous paths share the remaining stamp in _execute_update_now, and
the check that every stamp calls _note_update_completed is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-09-23 12:36:26 -04:00
ChuckandClaude Opus 5 bdb9a94033 refactor(api-v3): split the 10,469-line blueprint into a package (#553) 
* refactor(api-v3): split the 10,469-line blueprint into a package

web_interface/blueprints/api_v3.py held 111 routes, 56 helpers and 181
functions in one module -- 9% of the core by line count and three times the
next largest file. It becomes a package of nine route modules grouped by path
segment, plus __init__.py for the shared imports, constants, Blueprint and
helpers.

Every route module decorates the SAME api_v3 Blueprint object, so endpoint
names stay api_v3.<function>, the URL map is unchanged and app.py is untouched.
Verified: 111 routes before, 111 after, byte-identical rules, endpoints and
methods, and every endpoint still on the one blueprint.

  plugins   3,867   config    1,178   starlark  692   system  619
  fonts       452   misc        398   wifi      361   display 326   backup 212
  __init__  1,787 (imports, constants, Blueprint, 56 helpers)

Two things the URL-map check could not catch, both found by running the suite:

1. PROJECT_ROOT = Path(__file__).parent.parent.parent. Moving the code one
   directory deeper made that resolve to web_interface/ instead of the project
   root. Nothing failed at import; it surfaced as ~110 tests failing with 404s
   and "installation script not found", because every path built from it was
   one level too shallow. Now parents[3], and test_api_v3_url_map.py asserts
   PROJECT_ROOT/run.py exists so the next move cannot repeat it.

2. Module-attribute patching. Tests do
   monkeypatch.setattr(api_v3_module, "_BACKUP_EXPORT_DIR", ...) and a route
   module that binds such a name by value never sees the patch. The shared code
   therefore stays in __init__.py rather than moving to a _common submodule --
   it has to live on the module the tests patch -- and the eleven names tests
   patch are read back through the package (_pkg.X) instead of bound by value.
   Those eleven were found by AST-scanning every setattr in the test tree, not
   by guessing; "time" is among them, used to drive a fake clock through the
   second-resolution credential-backup filenames.

Test changes are confined to what genuinely moved: patch targets that now name
the owning route module, imports of helpers, and six tests that scan the api_v3
source as a file and now read the package directory.

Full suite: 4,278 passed, 68 skipped, 0 failed.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014RRtqXDCnvnY6EQwhT5CV9

* fix(api-v3): address CodeRabbit findings from the blueprint-split review

Fixes to the api_v3 package split (PR #553), one per finding verified
against the actual code:

- __init__.py: _redact_credentials only blanked scalar values under a
  credential-named key; a bare list of secrets under such a key (e.g.
  tokens: ["a", "b"]) passed through untouched, since the list branch
  recursed with no memory that its key looked like a credential. Nested
  dicts still walk normally (a documented, tested behaviour -- a container
  like secrets: {api_key: ..., note: ...} is a section name, not a value to
  blank outright), but any value reached under a credential-shaped key is
  now actually blanked.

- __init__.py: the OAuth helper script's raw stderr/stdout went to
  logger.error unredacted (CWE-532) right next to a comment claiming this
  was deliberate; the HTTP response already used the existing redact_text
  helper. Routed the log line through the same helper.

- __init__.py / starlark.py: the standalone Starlark manifest fallback
  (used when the plugin instance isn't loaded) read-modified-wrote
  manifest.json with no lock, unlike StarlarkAppsPlugin._update_manifest_safe
  (plugin-repos/starlark-apps/manager.py), which already holds an flock for
  the same file when the plugin is loaded. Added _starlark_manifest_lock,
  mirroring that pattern, and wrapped every standalone read-modify-write
  call site in it. The app-config update route also wrote config.json and
  the manifest as two separate, non-transactional writes (a second,
  distinct finding at the same call site); config.json is now rolled back
  if the manifest write that follows it fails.

- backup.py: restore options used bare bool() on values from the request,
  so {"restore_secrets": "false"} restored secrets anyway (bool("false") is
  True). Switched to the existing _coerce_to_bool helper already used for
  this exact purpose elsewhere in the package.

- config.py: an automated import-rewrite mangled four user-facing
  validation strings and their neighbouring comments -- "Invalid start
  time" had become "Invalid start _pkg.time" (and likewise for "end time")
  in both the schedule and dim-schedule per-day validation paths.

- display.py: `import _pkg.time as time_module` -- _pkg is a local alias
  for the package, not a real importable module, so this raised
  ModuleNotFoundError whenever a caller restarted an already-running
  display service via /display/on-demand/start, after the on-demand
  request was already written to cache. Fixed to `import time`. Audited
  the rest of the package for the same `_pkg.<module>` import mistake;
  every other `_pkg.` reference is a legitimate attribute read-through
  (`_pkg.time.time()`, `_pkg._get_starlark_plugin()`, ...), not a broken
  import statement.

- fonts.py: validate_file_upload's max_size_mb parameter is silently
  unused by that helper (it only checks filename/extension) -- the font
  upload route saved arbitrarily large files as a result. Added the same
  seek-and-check pattern already used for the sibling .star upload.

- wifi.py: two ad hoc, inconsistent bool coercions. POST
  /wifi/ap/auto-enable used bare bool(), so a JSON string "false" enabled
  it. POST /wifi/radio's enabled/force parsing recognized real bool and
  some strings but not int 1/0 (1 is True is False in Python). Factored one
  small _parse_bool_ish helper local to this file and used it at all three
  sites.

Not changed: the "unknown/misspelled restore option keys default to True"
half of the backup.py finding -- the file's own comment documents that a
missing key deliberately means "restore everything," matching the
already-existing JSON-parse-failure guard a few lines above it; only the
bool-coercion defect was a real bug.

Added or extended regression tests for every fix, following each area's
existing test conventions. Full suite: 4328 passed, 62 skipped, 2 failed
on both this branch and origin/main (missing tzdata package breaks two
timezone-alias tests in test_onboarding_checklist.py, unrelated to this
change) -- no new failures.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3bPMESe2TfrGvbs1ef9c5

* fix(api-v3): reject unknown restore option keys

CodeRabbit's review of the blueprint split (#553) asked that
POST /backup/restore reject option keys outside RestoreOptions'
known set. The follow-up commit fixed the bool("false")-is-True
bug with _coerce_to_bool but never added the key check: a typo'd
or renamed key (e.g. "restoreSecrets") is silently ignored by
opts_dict.get(key, True), so the flag stays at its True default
and secrets get restored despite the caller's request saying
otherwise -- with no indication anything was wrong.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vmcwf5vMgYqdt8bJTZtiwb

* fix(api-v3): address CodeRabbit findings on the blueprint split

- _redact_credentials: blank scalar descendants of objects reached
  through a credential-owned list (e.g. tokens: [{"value": "secret"}])
  regardless of field name -- the existing name-based walk only
  protected direct dict values under a credential key, not list items.
- wifi.py: reject enabled/force/auto_enable_ap_mode values
  _parse_bool_ish can't recognize (400) instead of silently treating
  them as False, which could disable Wi-Fi or the radio itself.
- Starlark manifest locking: lock a stable manifest.json.lock sidecar
  instead of manifest.json itself, in both the standalone route path
  (_starlark_manifest_lock) and the plugin path
  (StarlarkAppsPlugin._save_manifest / _update_manifest_safe).
  manifest.json is replaced by an atomic rename on every write, which
  swaps in a fresh inode; a lock held on the old inode does not
  exclude a second locker that opens the path afresh right after the
  rename and gets the new inode, so two writers could race despite
  each holding "a lock". A sidecar that no write ever touches always
  resolves to the same inode for every locker.

Skipped as stale: the "serialize the complete manifest
read-modify-write" finding at api_v3/__init__.py -- every standalone
handler that calls _write_starlark_manifest is already wrapped in
_starlark_manifest_lock() on this branch.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(api-v3): re-check reconciliation findings by the reconciler's own rules

Both CodeRabbit findings on the merge commit, verified against the code first.

Major, plugins.py: the stale-findings filter derived its own notion of "in
config" and "on disk", and both were looser than the reconciliation module's.
set(load_config()) also contains system keys, the secrets-file keys load_config()
merges in, and non-dict values; and any directory holding a manifest.json
counted as installed even when that manifest does not parse. Either looseness
clears a finding that is still true -- and a secrets key read as a plugin is the
precise bug the filter exists to stop reporting, so reintroducing that asymmetry
while re-checking was the wrong way round.

The two extractions now live in state_reconciliation.py as config_plugin_ids()
and disk_plugin_ids(), with ignored_config_keys() and secrets_top_level_keys()
alongside. _get_config_state() and _get_disk_state() use them too, so there is
one definition rather than two that can drift. _get_disk_state() re-reads each
manifest for version/name after taking membership from the shared extractor;
that costs one extra small read per plugin on a path that runs once per boot.

Minor, the new test: the fixture assigned api_v3.config_manager and
api_v3.plugin_manager directly. Those live on a module-level blueprint
singleton, so the mocks leaked into every later test that imports api_v3 --
pointing at a tmp_path already deleted. Both now go through monkeypatch.setattr,
which restores them. This is the same pollution class that made an earlier test
in this session break seven unrelated ones, so it is worth getting right.

Five cases added for the parity itself: a secrets key, a system key and a
non-dict value must not clear an "installed but missing from config" finding,
and neither an unparseable manifest nor a .standalone-backup- directory may
count as installed. All five fail against the looser version.

Linux CI on the preceding commit: Core unit tests, plugin harness, CodeQL and
CodeRabbit all pass. Codacy reads action_required on every commit of this
branch including the first, so it is pre-existing and not from this work.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
 2026-09-11 10:07:38 -04:00
ChuckandClaude Opus 5 ca26c1b83b fix(vegas): stop the width cap emitting fragments and stale windows (#446) 
* fix(vegas): stop the width cap emitting fragments and stale windows

Two defects in the rotation that narrows an oversized plugin to its
width budget. Both were found while investigating "cut off early /
starts in the middle" reports and are the reason the cap is no longer
on by default; they still bite anyone who sets one.

A rotation's last window was whatever happened to be left over. Windows
are placed by walking forward from the previous one, with nothing
looking at the remainder, so a 1,840px stocks ticker against a 1,536px
budget split 1,492 + 348 -- every other appearance showed seven seconds
and cut. Absorb a remainder below half a budget into the window before
it. That overruns the budget by at most half, which is the better trade:
the budget guards against one plugin holding the panel for minutes, not
against a 20% overshoot. The floor is measured against the budget rather
than the panel because snapping to item boundaries already lands an
ordinary window short of it -- a 512px budget over 182px-pitch items
yields 348px windows, so an absolute floor merges windows that were
never fragments.

The stored offset also outlived the content it was recorded against. It
was a pixel column, reused verbatim after the plugin re-rendered, so
once anything ahead of it changed width the window pointed at unrelated
items -- observed as news refreshing 9,793px -> 9,505px mid-rotation.
Track the rotation as an index into the strip's item boundaries instead,
since the Nth boundary survives a digit appearing in a price, and record
alongside it what the offset indexes into: a row list, a boundary list,
or a column in a gapless image. A mismatch restarts the rotation rather
than reinterpreting the number, which also closes the case where one
plugin's row index was read back as a pixel column after its content
changed from several rows to one wide strip.

Replaying the four plugins that actually hit the cap on a live 512px
panel: no window is now a fragment, none exceeds 1.5 budgets, and every
rotation still covers the whole strip.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Udr6MfaFLUPhX5Fgo67Jf5

* fix(vegas): apply the runt floor to the multi-row rotation too

The floor only guarded the single-image path. I had reasoned the row
path could not produce a runt because it wraps, which is wrong: wrapping
only helps when the row wrapped to actually fits. Rows of 450, 450 and
100 against a 512px budget give the 100 a pass of its own -- two seconds
against nine, which is the symptom this branch exists to remove.

Reproduced before changing anything:

    pass 1: 450px    pass 2: 450px    pass 3: 100px

A window may now overrun the budget while it is still shorter than the
floor, bounded at the same 1.5 budgets the single-image path allows, so
the short row is carried with its neighbour instead of standing alone.

    pass 1: 450px    pass 2: 450px    pass 3: 550px

A next row too wide to absorb within that cap still leaves a short
window standing -- rows of 900 and 100 keep alternating. Merging them
would mean a window of nearly two budgets, and the rule that always
shows an oversized first row already makes the same trade.

Three regression tests: the reported shape, that the overrun stays
bounded when a row cannot be absorbed, and that absorbing never drops a
row from the rotation. The single-image path is untouched -- the four
plugins that actually hit the cap on a live panel replay identically.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Udr6MfaFLUPhX5Fgo67Jf5

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
 2026-08-10 09:41:25 -04:00
ChuckandClaude Opus 5 6287acd591 fix(vegas): stop capping plugin width by default (#445) 
* fix(vegas): stop capping plugin width by default

Vegas plugins read as "cut off early" or "starting in the middle". That
was the per-plugin width budget, not the scroll engine:
overflow_mode=rotate is designed to resume mid-content on each
appearance, so the symptom was the feature working as specified.

Measured over a 17-plugin fleet on a 512px panel, the cap was a bad
trade. Only four plugins were ever wide enough to hit the 3.0 default --
leaderboard 11,518px, news 10,021px, odds-ticker 4,643px, hockey
1,508px. Weather is 650px and flights 512px; the cap never touched them
or the other eleven. So it bought nothing on thirteen plugins while
costing two visible faults on four: content entering mid-item (a news
ticker started at column 6027 of its own strip), and a final rotation
window of whatever happened to be left -- 348px of an 1,840px stocks
ticker, seven seconds of panel time.

Default max_plugin_width_ratio to 0 (uncapped), so every plugin
contributes all of its content and is always entered at its beginning.
The cap remains available, and vegas_max_width_screens still caps an
individual plugin -- which is where the knob belongs, since a genuinely
long ticker is a property of that plugin rather than of the fleet.

Verified on a live 512px device: 327 budget crops in the preceding six
hours, none after.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Udr6MfaFLUPhX5Fgo67Jf5

* test(vegas): cover the width cap still working when asked for

Defaulting the cap off must not quietly remove it. Asserts that an
explicit ratio is honoured and validates, alongside the existing check
that omitting it means uncapped.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Udr6MfaFLUPhX5Fgo67Jf5

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
 2026-08-07 16:17:28 -04:00
ChuckandClaude Sonnet 5 5b45f35888 Vegas mode: reclaim dead space and pace the rotation (#423) 
* Vegas mode: reclaim dead space and pace the rotation

On a wide panel Vegas mode spent much of its time showing black. At 50px/s
on a 512px display, one display width of blank is 10.2 seconds, which makes
several long-standing behaviours expensive:

- ScrollHelper prepended a full display width of black as an "initial gap",
  charged once per cycle — 10.2s of black at the start of every rotation.
- Plugins without get_vegas_content() are captured off a full-display canvas,
  so their blank margins entered the ticker too. Measured: of-the-day drew
  35px of "No Data" on a 512px canvas (92% blank), youtube-stats 142px of
  content with 185px of black either side. Only the scroll_helper path had
  any trimming.
- Cycle transitions deliberately pushed a blank frame and then recomposed
  synchronously: 84ms at best, 4.8s at worst, every millisecond of it black.
- buffer_ahead doubled as the cycle size, so a 21-plugin install showed 3
  plugins per cycle and took ~7 cycles to come around.
- separator_width was applied between every image rather than at plugin
  boundaries, so a per-row ticker like the F1 scoreboard (116 images, which
  it renders 4px apart internally) got a 32px chasm between each row — and
  the width budget didn't count those gaps, so the plugin quietly occupied
  far more of the panel than intended.

Changes:

- src/vegas_mode/geometry.py: numpy column-ink primitives shared by the
  trimmer and the audit tool, so the number reported is the number acted on.
  A Python per-column loop over a 17,000px strip is far too slow for the
  render path.
- PluginAdapter trims every content path, not just scroll_helper. Only outer
  edges are cropped: interior blank columns are the plugin's own layout
  (logo left, score right) and closing them would corrupt the design. A
  plugin on a non-black background is inherently unaffected.
- ScrollHelper.create_scrolling_image takes an explicit lead_gap, still
  defaulting to display_width so the many standalone-ticker callers are
  unchanged. Vegas passes lead_in_width (default 0).
- Cycle end holds the last rendered frame instead of blanking, turning the
  recompose into a brief freeze rather than the panel switching off.
- plugins_per_cycle (default 6) is split from buffer_ahead, which goes back
  to being only a prefetch low-water mark.
- max_plugin_width_ratio (default 3x display width) caps one plugin's share
  of a cycle. Overflow is deferred, not discarded: a rotation offset advances
  each fetch so later rows appear on subsequent cycles. Single oversized
  images are cropped at a blank column so the cut misses glyphs.
- Composition groups images by plugin: rows are joined by intra_plugin_gap
  (default 8) and separator_width applies only between plugins. The width
  budget now counts those gaps.
- Plugin data updates no longer run on the Vegas render path.

All new settings are user-configurable in Display -> Vegas Scroll, including
min/max cycle duration and dynamic duration, which previously existed in code
but were reachable only by hand-editing config.json.

Measured with scripts/dev/vegas_audit.py on a 512x64 panel:

  mean ink coverage    42.7% -> 69.4%
  fully blank           5.9% -> 0%
  reads as empty        13.6% -> 0%
  worst blank stretch    4.8s -> 0s
  full rotation          414s -> 123s
  plugins per cycle         3 -> 6

Note the metric choice: a "fully blank" scan (>=95% black viewport) reported
only 0.4% and badly understated the problem, because two full-width segments
with mid-canvas content never fully blank the viewport — they hold it at ~28%.
window_coverage_stats grades every viewport position by how much ink it
carries, which is what tracks perceived dead time.

Known remaining: cycle transitions still freeze ~3.5s while the next cycle is
fetched. Fixing that needs background prefetch, which is deferred because the
fallback-capture path mutates the shared display_manager.image and racing it
against the render loop risks torn frames.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Drop unused Optional import from the vegas audit script

Flagged by Codacy (F401). Any, Dict and List are all still used.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Align Vegas API bounds with validate(), fix audit config plumbing

Both from review feedback on #423.

The web API's accepted ranges disagreed with VegasModeConfig.validate(),
which is what actually gates Vegas starting:

  scroll_speed      1-100  -> 1-200   (a slider value of 150 returned 400)
  separator_width   0-500  -> 0-128
  target_fps        1-200  -> 30-200
  buffer_ahead      1-20   -> 1-5

The three loose ones were the dangerous direction: the value saved with a
200, then VegasModeCoordinator.start() failed validation with only a log
line, so the ticker silently never ran. The UI already matched validate() in
all four cases, so the API was the odd one out.

test_vegas_api_bounds_match_validate parses the numeric_fields map out of
api_v3 and asserts every bound against validate(), plus that validate()
accepts both endpoints and rejects just outside them, so these cannot drift
apart again. That test immediately caught a missing upper bound on
min_plugin_width, now added — unbounded it would drop every segment and
leave a blank ticker.

Separately, vegas_audit.py constructed PluginAdapter without the config, so
it fell back to VegasModeConfig() defaults and would report trimming and
width-budget behaviour that differed from the user's config.json. It now
passes the loaded config exactly as the coordinator does. This is the same
class of drift the explicit lead_gap and grouping arguments already guard
against. Output is unchanged on a rig whose config matches the defaults.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Vegas mode: render plugins narrower, space rows by measured separation

Trimming reclaims blank margins but cannot compact a layout that genuinely
spans the display — a five-column forecast, a progress bar drawn at 100%
width, a stat block with the panel's whole width between its elements. Those
need the plugin to make different layout decisions, which means telling it the
screen is narrower while it renders.

DisplayManager.render_size() presents a smaller logical canvas for the
duration of a Vegas content fetch, reusing the same _LogicalMatrix
indirection double-sided mode already relies on so plugins see a consistent
size from every accessor. Plugins that size themselves from matrix.width need
no changes at all; one that wants to be explicit can read the new
BasePlugin.get_vegas_render_width().

Width is a percentage so a single setting travels across panel sizes:
vegas_scroll.render_width_pct globally, or vegas_width_pct in an individual
plugin's config. Measured on a 512x64 panel with real data:

  ledmatrix-weather   1536px -> 576px   (forecast becomes narrow cards)
  youtube-stats        353px -> 199px   (2% blank left, so genuinely compact)
  geochron             453px -> 153px   (ink density rises to 100%)
  ledmatrix-flights    950px -> 740px

The youtube-stats figure is the clearest evidence the layout itself changed
rather than being cropped: at full width the content had to be trimmed from
512px to 353px, whereas at 40% it arrives with almost no blank to reclaim.

Row spacing is now measured rather than added. A flat gap gets it wrong in
both directions at once — content drawn flush to its own edges ends up nearly
touching (reported for recent sports scores, which sat 8px apart), while
content already carrying wide margins gets pushed even further out.
separation_gap() measures the blank each pair already has and adds only the
shortfall, up to min_content_separation (default 24). intra_plugin_gap stays
as a floor applied regardless.

Two tests shipped in the previous commit encoded the old flat-gap arithmetic
and are updated to the measured semantics, including one renamed to reflect
that zero intra_plugin_gap alone no longer butts rows together.

Also fixes a real bug found while testing: the harness display manager had no
render_size(), and because the adapter catches broadly that surfaced as "no
content" rather than an error, silently dropping five plugins. Added the
context to VisualTestDisplayManager for parity, and _render_at() now degrades
to a no-op on any display manager lacking it, so a third-party or older
harness loses the narrowing rather than the content.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Vegas mode: end cycles before the wrap, keep the width budget honest

Three fixes, the first a regression from lead_in_width defaulting to 0.

get_visible_portion wraps: once scroll_position + display_width passes the end
of the strip it fills the right of the frame from the *head* of the same strip.
So the final display_width of travel showed the cycle's first plugin re-entering
on the right while its last plugin exited on the left, and the recompose that
followed replaced both at once. On a 512px panel at 50px/s that was 10.2s of
two plugins on screen at once, ending in a hard cut — reported as the ticker
"switching mid-scroll" from F1 to news.

That used to be invisible because the strip began with a full display_width of
blank, so the wrapped-in region was black. Removing that blank (it was 10s of
dead panel per cycle) exposed the wrap. Cycles now end one display width
earlier, before any wrapped content appears, clamped for strips no wider than
the display so they don't complete instantly and spin the recompose loop.

Verified on hardware: a 3936px strip now completes at 68.5s, exactly
(3936 - 512) / 50.

Second, auto_trim=False also skipped the width budget, which is an unrelated
concern — turning off margin cropping should not let one plugin hold the panel
for minutes. Seen in the field: the F1 scoreboard contributed 116 images and
14,848px untouched, giving a 33,821px cycle (11 minutes of content). The budget
now applies regardless of trimming; with it restored that cycle is 6,362px.

Third, the budget accounted for row gaps using the flat intra_plugin_gap while
the compositor had moved to measured separation, so it under-counted by up to
(min_content_separation - intra_plugin_gap) per row and a many-row plugin
overran its cap. Both now use the same separation_gap() rule, and a test
asserts the composed block fits the budget end to end rather than trusting the
two paths to agree.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Fix IndexError in find_blank_cut when the cut lands on the image edge

A cut position after the last column is legitimate — _crop_to_budget asks for
min(start + budget, img.width), which equals the width whenever the remaining
strip is shorter than the budget. find_blank_cut clamped target to width but
then walked leftwards starting at target itself, so ink[width] raised
IndexError.

Caught on hardware: it killed the ledmatrix-stocks fetch, and because
_fetch_plugin_content catches broadly that surfaced as the plugin silently
contributing nothing for the cycle.

Only reachable on the second or later pass of the rotating window over a single
oversized image, which is why the existing tests missed it — they all exercised
the first pass, where start is 0 and start + budget is comfortably inside the
image. Added TestRotationAcrossMultipleCycles, which walks the window round
several times and asserts content is never lost, plus direct coverage of
find_blank_cut at and beyond the image edge.

Both bounds now stop at width - 1 so neither direction can index past the end.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Only cut oversized segments at real gaps between items

The width-budget crop snapped to the nearest blank column, and in rendered text
the gap between two characters is a single column. So a cut routinely landed
inside a word: the cycle showed "Wednesda" and the orphaned "y" turned up as a
lone floating letter in the next cycle, positioned after whatever plugin
happened to precede it.

Measured on the clock-simple segment to confirm: its blank runs are
[1, 1, 1, 1, 1, 8, 8] — five single-column letter gaps, every one of which
find_blank_cut would happily have chosen.

Cuts now only land in a run of at least min_cut_gap blank columns (default 6),
which excludes letter spacing while still finding the gaps plugins put between
items (the stocks ticker uses 32px, baseball 48px). Where no boundary falls
inside the budget the cut waits for the next one and overruns, because
splitting an item is worse than a slightly long segment.

Continuous content is treated differently on purpose: an image with no internal
gaps is a map or a chart, where any column is as good as another, so it is still
cut to the budget exactly. The gap rule protects discrete items; letting a solid
image escape the cap in its name would be wrong.

blank_runs() is vectorised — 48ms for a 17,000px strip, against seconds for a
per-column Python loop.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Hold capture_mode for every plugin render, not just narrowed ones

The native content path only entered capture_mode when it was also narrowing
the canvas, so at full width — which is every plugin without a vegas_width_pct
override, i.e. most of them — a plugin calling update_display() while building
its Vegas content wrote straight to the hardware. That is a visible flash
mid-scroll, and it lines up with the flash reported at cycle transitions, when
several plugins are fetched back to back.

Suppression is now unconditional; the narrowing context stays separate because
it is already a no-op at full width.

Both contexts are reached through helpers that degrade to nullcontext when the
display manager lacks them. That matters more than it looks: the adapter's
handlers are deliberately broad, so an AttributeError from a missing context
does not surface as an error — it surfaces as the plugin contributing nothing.
Making the call unconditional without this turned 44 tests red for exactly that
reason, all of them reporting lost content rather than the real cause.

The test double now provides capture_mode and render_size too, so tests
exercise the real contexts instead of silently taking the degraded path.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Vegas mode: one continuous strip instead of swapping cycles

A cycle used to be a discrete strip that got replaced: motion stopped, every
pixel was substituted at once, and the next group started with the viewport
already full. That is the freeze, the flash and the jump.

The strip is now extended rather than replaced. ScrollHelper gains
append_content(), which adds items on the right without touching
scroll_position or total_distance_scrolled, so motion continues and the next
group simply arrives from the right. Because completion is measured against
total_scroll_width, extending also defers completion — there is no longer a
cycle boundary to see.

drop_scrolled_prefix() reclaims what has gone past, keeping the strip bounded
however long Vegas runs (observed 5,000-11,000px against an unbounded strip
otherwise). It shifts total_distance_scrolled and total_scroll_width together so
the completion arithmetic is unchanged, and refuses to run while the viewport is
wrapping: wrapping reads the head of the strip into the right of the frame, so
trimming the head there would visibly change the picture. A test caught that.

Groups are prepared off the render thread. The constraint is that the canvas and
the matrix proxy are process-wide mutable state, so narrowing or capturing
through them from another thread would corrupt the frame the render loop is
pushing. get_content() therefore takes offscreen_only: the background thread uses
only paths that avoid the canvas, and anything needing it is marked and picked up
on the render thread. That puts the expensive work (native renders of leaderboard
and baseball cards, seconds each) in the background and leaves the cheap work
(display capture, 40-600ms) in the foreground.

DisplayManager's capture flag is now thread-local. As a shared flag, a background
capture would have suppressed the render loop's own frame pushes for its
duration, freezing the panel precisely when the point was to avoid a freeze.

Canvas-bound plugins are drained one at a time rather than as a batch: six at
once held the render thread for 1.75s. Drains are also spaced by two seconds
while the lookahead is healthy, since taking them back to back turns one long
stall into a run of short ones. When the strip is genuinely running short the
throttle is ignored, because content matters more than smoothness there.

Measured on hardware: zero cycle-complete swaps, drains landing 2-4s apart,
lookahead holding at 1,200-3,500px, no errors.

Set continuous_scroll false to restore the swap behaviour; the old path is intact.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Pace the Vegas frame loop adaptively: 31.5 -> 78.7 fps

The loop slept a fixed frame_interval on top of however long the frame took, so
at a measured 31.6ms per frame a flat 8ms of that was pure idle — a quarter of
the budget spent not rendering. It now sleeps only the remainder of the budget.

Measured on hardware: 31.5 fps to 78.7 fps sustained, with CPU going *down* from
150% to 127%. Scroll speed is unchanged at 49.9px/s against a configured 50,
because motion is derived from elapsed time rather than frame count — this buys
smoothness, not speed.

Worth recording what the bottleneck was not: the per-frame render path measures
0.34ms in total (0.18ms for the numpy slice, 0.17ms for the dirty-tracking
digest), which is a theoretical 2900 fps. Optimising any of that would have been
wasted effort. The frame was idle, not busy.

Also nices the prefetch thread. Its work is PIL and numpy that releases the GIL,
so the scheduler can act on the priority, and without it the prefetch competes
for the same cores as the render loop.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Sub-pixel scrolling: motion at the frame rate, not the pixel rate

With integer positioning the number of distinct frames per second equals the
scroll speed in px/s, however fast the loop renders. Measured at 50px/s and
78.7fps, 36% of frames were byte-identical: the extra frames cost work and
bought no motion, and what was left was 50 discrete 1px steps a second.

Two things were wrong with the pre-existing sub-pixel support. get_visible_portion
never consulted sub_pixel_scrolling — it always took the integer path, so the flag
and _get_visible_portion_subpixel were dead code. And that implementation needed
scipy.ndimage.shift, which is not installed on the target devices (HAS_SCIPY is
False there), so it would not have interpolated even if reached. Verified both:
positions 1000.0 and 1000.5 produced identical frames either way.

Blending is now wired up and implemented with numpy. Two details make it
affordable: slice cached_array directly instead of building two PIL images only
to convert them straight back (the naive version measured 15x the integer path),
and use fixed-point uint16 multiply-add rather than float32, which suits the Pi's
cores and gives finer weighting than the panel can resolve. Result 0.939ms
against 0.237ms — 0.70ms added per frame, a 1065fps ceiling.

Measured on hardware: 81.2 fps with blending on, against 78.7 with it off, so no
cost within noise — and every frame is now a distinct position rather than one in
three being a repeat.

The trade is a slight horizontal softening of text, since each frame blends two
positions. Set smooth_scroll false for maximum crispness.

Also benchmarked and cleared as non-issues: extending the strip costs 9.4ms on an
11,000px strip and trimming 2.5ms, both under one frame at this rate.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Add overflow handling: keep ordered content whole instead of rotating a window

The width budget split any oversized plugin by advancing a window each cycle.
That is right for interchangeable items — news headlines, odds, stock prices —
but wrong for ordered content: a league table showed ranks 1-6, then resumed at
7 two rotations later, which reads as out of order and out of context. Nobody
needs rank 23 in a ticker; they need the top of the table, every time.

overflow_mode chooses between them:

  rotate   — advance a window each cycle so everything is seen eventually
             (unchanged default)
  truncate — always show the start and drop the rest, keeping ordered content
             coherent. Records no window state, so every pass starts at the top.

Per-plugin vegas_overflow overrides the global setting, since one install has
both kinds of plugin. Also adds per-plugin vegas_max_width_screens, so content
that must stay whole can be given more room — or uncapped with 0 — without
lifting the cap on every ticker.

Applied on the test rig: f1-scoreboard and ledmatrix-leaderboard set to
truncate, and baseball given 4.5 screens because it was showing 8 of 9 games
when the whole slate needed only a little more room. Verified: F1 now reports
"the first 10 of 116 ... the rest are not shown", baseball has dropped out of
the budget log entirely, and stocks, odds-ticker and stock-news still rotate.

Also corrects the crop log, which claimed "window advances next cycle"
unconditionally and so misreported truncated crops. A test now pins the
behaviour behind the message: truncate must leave no offset recorded.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Stop Vegas mode showing last night's games as if they were live

A game that was live in the evening was still being drawn as live the next
morning. Two faults combined to freeze plugin visuals indefinitely.

PR #291 added a call to plugin_adapter.invalidate_plugin_scroll_cache() so
a plugin's own cached scroll image would be rebuilt from fresh data. That
method was never implemented. hot_swap_content() wraps the call in a broad
except, so every hot swap has raised AttributeError and been swallowed
silently ever since — which is why the visuals it was meant to keep fresh
never were.

Continuous scrolling then removed the only path that reached it at all:
should_recompose() and hot_swap_content() are called from the
non-continuous branch of run_frame(), and continuous_scroll defaults to
True. So on a default install the pending-update flags were set by the
update tick, never consumed, and grew without bound.

Together these froze content completely, because refetching is not enough
on its own: the sports plugins' get_vegas_content() regenerates only "if
the cache is empty", so take_next_group() kept receiving the same picture
however often it asked.

Fixed by:

- Implementing invalidate_plugin_scroll_cache(). It covers both layouts —
  a helper directly on the plugin (stocks, news, odds-ticker) and one
  owned by a scroll-display manager (the sports scoreboards, which is the
  shape that produced this bug) — and clears cached_image and
  cached_array together, since the array is the image's numpy mirror.

- Adding StreamManager.invalidate_pending_updates() and calling it from
  the continuous branch. It only drops the caches; the plugin recomposes
  when it next comes round in the rotation. process_updates() is wrong
  here: it refetches synchronously and merges into the active buffer that
  continuous mode bypasses, and hot_swap_content() rebuilds and
  repositions the whole strip, which is the freeze-and-jump this mode
  exists to avoid.

Tests assert the fix rather than the implementation: 14 of the 17 new
tests fail without it. Includes the wiring itself, since the regression
was a call that was simply absent, and a check that the scroll position is
untouched so this cannot regress into the swap's visible jump.

All Vegas suites pass (355 tests). test_display_controller_vegas_tick.py
still cannot be collected off-device for want of rgbmatrix, identically
with and without this change.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

* Fix two CodeRabbit-flagged test assertions in vegas density tests

test_prepared_group_is_used_without_refetching had a tautological final
assertion; now checks stream.calls directly. test_no_partial_letter_at_either_edge
required both crop edges to be blank, but the left edge here is always the
crop's start position with no lead-in gap in word_strip, so it legitimately
carries ink — only the right edge is an actual cut and needs the check.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

---------

Co-authored-by: Claude <noreply@anthropic.com>
 2026-07-31 09:40:38 -04:00