yorickr/plezy
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore: clean up code comments

Browse Source
This commit is contained in:
edde746
2026-08-10 20:28:41 +02:00
parent 5611c6785a
commit 69fadc220d
170 changed files with 324 additions and 1765 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.maestro/regression_flows/03_tv_library_focus.yaml
+4 -5
View File
@@ -22,8 +22,8 @@ tags:
- pressKey: "Remote Dpad Up"
- pressKey: "Remote Dpad Right"
- pressKey: "Remote Dpad Right"
# The grid row must be matchable before the sort sheet opens, otherwise the
# occlusion assertion below would pass for the wrong reason.
# Wait for the row before opening the sheet so occlusion is tested against the
# grid, not an absent or stale match.
- extendedWaitUntil:
visible: "(?s).*Alpha Archive.*"
timeout: 15000
@@ -31,9 +31,8 @@ tags:
- extendedWaitUntil:
visible: "Title"
timeout: 10000
# The host renders the grid, the barrier, and the sheet in one Stack and never
# unmounts the grid, so before 8e1904dd wrapped the barrier in BlockSemantics
# an occluded row still read as visible to accessibility and to Maestro.
# The grid remains mounted under the Stack; BlockSemantics is required for
# occluded rows to disappear from Maestro's accessibility tree.
- assertNotVisible: "(?s).*Alpha Archive.*"
- pressKey: "Remote Dpad Center"
- pressKey: "Remote Dpad Left"
.maestro/regression_flows/04_tv_player_keys.yaml
+7 -15
View File
@@ -16,13 +16,9 @@ tags:
visible: "(?s)^(Play|Resume).*$"
commands:
- pressKey: "Remote Dpad Center"
# The TV player opens with its chrome down so the OSD and timebar never sit
# over the opening seconds of the picture (#1765), so the Pause button can no
# longer stand in for "the player has finished loading". Wait for the detail
# screen to go first: the labelled spinner is necessarily up the moment the
# player route owns the frame, and it only clears once playback reports its
# first frame or gives up, so this can never advance before the media opens.
# A give-up lands on the transport assertions below, which is where it belongs.
# TV playback starts with chrome hidden (#1765), so wait for its loading label
# rather than treating the Pause button as readiness. A failed open reaches the
# transport assertions below.
- extendedWaitUntil:
notVisible: "Overview"
timeout: 30000
@@ -30,18 +26,14 @@ tags:
notVisible: "Loading video"
timeout: 30000
- assertNotVisible: "(?s)^(Play|Pause)$"
# Hardware transport keys drive playback without raising the chrome (#1676):
# the player answers with a centred disc (pause) or a side readout (seek), never the
# full chrome, so subtitles stay readable.
# Pause first — it also freezes the clip, so the seek assertions below cannot
# race the end of a short fixture item.
# Transport keys keep chrome hidden (#1676); pause first so seeks cannot race
# the end of this short fixture.
- pressKey: "Remote Media Play Pause"
- extendedWaitUntil:
visible: "Paused"
timeout: 10000
- assertNotVisible: "(?s)^Play$"
# D-pad seeking reports through the skip badge, never the scrub bar, and
# consecutive presses in one direction stack into a running total.
# D-pad seeks use the cumulative skip badge, not the scrub bar.
- pressKey: "Remote Dpad Right"
- extendedWaitUntil:
visible: "(?s)^Seek forward 10 seconds$"
@@ -56,7 +48,7 @@ tags:
visible: "Playing"
timeout: 10000
- assertNotVisible: "(?s)^Pause$"
# Select stays the deliberate way to bring the chrome back.
# Select is the explicit way to restore the chrome.
- pressKey: "Remote Dpad Center"
- extendedWaitUntil:
visible: "(?s)^(Play|Pause)$"
.maestro/regression_flows/05_tv_next_episode_back.yaml
+5 -12
View File
@@ -7,10 +7,8 @@ tags:
- playback
---
- runFlow: ../subflows/onboard_jellyfin_tv.yaml
# Reach Search with the D-pad, the way 10_tv_settings_navigation.yaml does. A
# tap here flips InputModeTracker into pointer mode, which collapses the rail
# to icons and leaves no "Search" label to hit — and a percentage coordinate
# does not survive the jump from a phone in forced TV mode to a 4K television.
# Reach Search by D-pad: taps switch to pointer mode and hide the TV rail label,
# while percentage coordinates are not portable across TV sizes.
- pressKey: "Remote Dpad Left"
- extendedWaitUntil:
visible: "Search"
@@ -38,11 +36,8 @@ tags:
- waitForAnimationToEnd:
timeout: 5000
- tapOn: "(?s)^Play S1E1$"
# The TV player opens with its chrome down (#1765), so the Pause button can no
# longer stand in for "the player has finished loading". Wait for the detail
# screen to go first: the labelled spinner is necessarily up the moment the
# player route owns the frame, and it only clears once playback reports its
# first frame or gives up, so this can never advance before the media opens.
# The player starts with chrome hidden (#1765); wait for its loading label
# rather than using Pause as readiness.
- extendedWaitUntil:
notVisible: "(?s)^Play S1E1$"
timeout: 30000
@@ -51,9 +46,7 @@ tags:
timeout: 30000
- pressKey: "Remote Media Fast Forward"
- pressKey: "Remote Media Fast Forward"
# Gate on the prompt's own Cancel action, the label the iOS branch below taps.
# "Next Episode" is the credits skip button at this point in the episode, so
# waiting on it would let Back fire before the prompt ever opened.
# Wait for Cancel, not Next Episode (the credits-skip action), before pressing Back.
- extendedWaitUntil:
visible: "(?s)^Cancel$"
timeout: 20000
.maestro/regression_flows/06_playback_recovery.yaml
+2 -4
View File
@@ -8,10 +8,8 @@ tags:
---
- runFlow: ../subflows/onboard_jellyfin.yaml
- tapOn: "(?s)^Libraries.*"
# Every seeded alphabet title carries the same dateadded, so which of them the
# "Recently Added" rail returns is not stable across scans. This flow only
# needs some playable movie, so accept any of them rather than pinning one and
# failing whenever the tie-break lands elsewhere.
# All seeded titles share dateadded, so accept any playable Recently Added movie
# instead of relying on an unstable tie-break.
- extendedWaitUntil:
visible: "(?s).*(?:Alpha Archive|Bravo Beacon|Charlie Circuit|Delta Drive|Echo Engine|Foxtrot Frame|Gamma Garden|Hotel Horizon|India Index|Juliet Junction|Kilo Key|Lima Loop|Mike Matrix|November Node|Oscar Orbit|Papa Pipeline|Quebec Queue|Romeo Relay|Sierra Signal|Tango Track|Uniform Update|Victor View|Whiskey Widget|Xray XML|Yankee Yield|Zulu Zone).*, movie, .*"
timeout: 15000
.maestro/regression_flows/07_sheet_back_dismiss.yaml
+6 -21
View File
@@ -7,20 +7,9 @@ tags:
- settings
- sheets
---
# Guards 8e1904dd and c48cbf70 together.
#
# On a phone, Settings is a pushed route. Before c48cbf70 it carried no
# OverlaySheetHost, so Manage Libraries fell back to showModalBottomSheet and a
# single Back tore down the sheet *and* Settings. Before 8e1904dd the host
# answered the platform pop with a TV-only dedup marker that could already be
# spent, stranding the sheet open with no way out. Either defect fails the
# "Manage Libraries is still there after one Back" assertion below.
#
# The same commit blocks semantics behind the barrier, so the occlusion
# assertions pin that an open sheet actually hides the rows underneath.
# Pristine state on purpose. The TV regressions in the same CI group leave
# "Force TV mode" enabled, and this flow asserts the handheld Settings route,
# so it must not inherit whichever layout the previous flow left behind.
# Guards 8e1904dd and c48cbf70: phone Settings must host the sheet so one Back
# dismisses only the sheet, while BlockSemantics hides rows behind the barrier.
# Start clean because preceding TV flows may leave Force TV mode enabled.
- runFlow: ../subflows/onboard_jellyfin.yaml
- runFlow: ../subflows/open_settings.yaml
- assertVisible: "(?s)^Services.*"
@@ -29,10 +18,8 @@ tags:
- extendedWaitUntil:
visible: "Maestro Movies"
timeout: 15000
# The barrier swallows every pointer event, so the rows behind it must leave
# the semantics tree too. Without BlockSemantics an occluded row still reads as
# visible and a tap on it lands on the barrier instead. "Manage Libraries" is
# excluded on purpose: it is also the sheet's own title.
# The barrier removes covered rows from semantics; the sheet title remains
# visible because it belongs to the sheet.
- assertNotVisible: "(?s)^Services.*"
- assertNotVisible: "(?s)^Video Playback.*"
- assertNotVisible: "(?s)^Appearance.*"
@@ -41,15 +28,13 @@ tags:
- waitForAnimationToEnd:
timeout: 3000
- assertNotVisible: "Maestro Movies"
# One Back closed only the sheet: Settings is still the current route.
- extendedWaitUntil:
visible: "(?s)^Manage Libraries.*"
timeout: 10000
- assertVisible: "(?s)^Services.*"
- assertVisible: "(?s)^Video Playback.*"
- assertNotVisible: "Discover"
# Reopening proves the host did not strand the one-shot dedup marker: a second
# dismissal has to work exactly like the first.
# Reopening verifies the one-shot pop deduplication marker was not stranded.
- tapOn: "(?s)^Manage Libraries.*"
- extendedWaitUntil:
visible: "Maestro Movies"
.maestro/regression_flows/08_track_choice_survives_pending_pass.yaml
+4 -14
View File
@@ -7,20 +7,10 @@ tags:
- playback
- subtitles
---
# Guards 468d6804. When a source advertises tracks the native list has not
# produced yet, TrackManager arms an automatic selection pass with a 5s first
# attempt and a 25s deadline. A user pick made inside that window used to only
# persist the preference, leaving the pass armed, so the choice snapped back to
# the server/profile preferred track up to ~30s later. The pick now retires the
# pending pass, and the wait below is long enough to outlive both timers.
#
# Fixture: Codec H264 EAC3 Multisub advertises three E-AC-3 audio streams
# (Hindi and Japanese flagged default, English not) and 36 SRT subtitle
# streams, so both picks below are genuine changes away from the automatic one.
# Pristine state on purpose, for two reasons: the TV regressions in the same
# CI group leave "Force TV mode" enabled, and a remembered track selection
# from an earlier run would pre-select the very rows this flow picks, making
# every assertion below vacuous.
# Guards 468d6804: a user pick must retire TrackManager's pending automatic
# selection pass, or the server preference can overwrite it up to 30s later.
# The codec fixture supplies non-default audio and subtitle choices, and clean
# state prevents remembered selections from making the assertions vacuous.
- runFlow: ../subflows/onboard_jellyfin.yaml
- runFlow:
file: ../subflows/open_codec_sample.yaml
.maestro/regression_flows/09_language_picker_locales.yaml
+4 -13
View File
@@ -7,16 +7,9 @@ tags:
- settings
- i18n
---
# Guards 7677d159 and 100d7729. AppLocale is switched over exhaustively in
# appearance_settings_screen.dart, so a missing arm is a compile error, but a
# locale whose JSON never reached the generator ships a picker entry that falls
# back to English. Selecting Turkish and reading root-level navigation labels
# proves the generated locale is wired end to end.
#
# This flow mutates a persisted pref, and the app language decides every
# selector after the tap, so it starts from a cleared install rather than
# ensure_onboarded: a run that dies mid-flow must not leave the next flow
# hunting for English strings in a Turkish tree.
# Guards 7677d159 and 100d7729: selecting Turkish and reading root navigation
# labels proves the generated locale is wired end to end. Start clean because
# the language preference persists and affects every later selector.
- runFlow: ../subflows/onboard_jellyfin.yaml
- runFlow: ../subflows/open_settings.yaml
- tapOn: "(?s)^Appearance.*"
@@ -54,9 +47,7 @@ tags:
- assertVisible: "Keşfet"
- assertNotVisible: "(?s)^Libraries.*"
- assertNotVisible: "Discover"
# Walk back through the Turkish tree and restore English. The pref outlives a
# plain launchApp, so leaving the device in Turkish would break every later
# flow in the suite.
# Restore English before leaving the device; the locale preference outlives launchApp.
- repeat:
times: 3
while:
.maestro/regression_flows/10_tv_settings_navigation.yaml
+5 -15
View File
@@ -7,21 +7,11 @@ tags:
- tv
- settings
---
# Guards ef310459 and the Apple-only gates touched by 6c14049e, and pins TV
# settings navigation generally.
#
# It deliberately does NOT guard the 15b54e2e density change. That commit took
# every settings row from ~80dp to ~61dp, but the rows below the fold stay in
# the semantics tree either way, so this flow passes identically on both sides
# of it — measured, not assumed. A real guard would need a pixel `height`
# assertion, and Maestro has no DPR-relative form, so pinning one here would
# only hold for the box it was written on. Row geometry belongs to
# test/widgets/setting_tile_test.dart.
#
# Every step uses D-pad keys, never taps. InputModeTracker flips to pointer
# mode on the first touch event and the TV number spinner only builds in
# keyboard mode, so a single tapOn anywhere above would silently swap the
# spinner for a plain text field.
# Guards ef310459 and the Apple-only gates touched by 6c14049e.
# This flow intentionally does not pin 15b54e2e's row density: semantics stay
# stable across that change, and Maestro has no DPR-relative height assertion.
# Use D-pad only; a tap switches InputModeTracker to pointer mode and removes
# the TV number spinner.
- runFlow: ../subflows/onboard_jellyfin_tv_device.yaml
- pressKey: "Remote Dpad Left"
- extendedWaitUntil: