Two races shared the same cause: pushSamples runs on the audio capture thread
while clear() runs on the Compose main thread.
1. Lost redraw notifications
Both paths did `_revision.value += 1`. That expands to get -> add -> set and is
not atomic. A controlled two-thread probe (20k increments each, five runs)
lost up to 6,402 increments / 16%; using StateFlow.update lost zero. Since
revision is the Canvas's only redraw signal, every lost update can leave the
waterfall showing stale rows. If both writes land on the same number, StateFlow
sees no value change and notifies nobody.
Use `_revision.update { it + 1 }` in both paths.
2. Clear resurrected pre-clear audio
pushSamples copies pending audio under the lock, deliberately performs FFT
outside it, then reacquires the lock to append rows. The exact interleaving:
audio thread: take old audio, start FFT
main thread: user taps Clear -> rows/pending empty
audio thread: old FFT completes -> appends old rows again
The display becomes empty then immediately redraws the audio the user cleared.
A deterministic thread probe reproduced old rows after clear. Add a generation
counter protected by the same lock: pushSamples records it before FFT and drops
the computed rows when clear incremented it meanwhile. Fixed probe remains empty.
Verification: :feature:cw:compileReleaseKotlin + full :core:domain:test BUILD
SUCCESSFUL; grep confirms no non-atomic revision increments remain.
All Chinese comments (//, /* */, KDoc) across core/app/feature/build-
logic translated to English (550 lines, 73 files after FT8 rollback).
Code logic untouched - comment text only. Verified: all modules
compileDebugKotlin BUILD SUCCESSFUL.
aapt2 treats in/b+in as the same locale config as id -> Duplicate
resources build failure. in and id are equivalent in Android's
resource matcher (both normalize to id), so values-in + values-id
is sufficient coverage for Indonesian devices.
values-in-rID did not survive aapt2 linking (normalized away).
Switch to values-b+in (BCP-47) which compiles cleanly and keeps an
explicit "in" language config alongside values-id. Verified: aapt2
compile of values-in/values-id/values-b+in OK.
Root cause: aapt2 merges values-in into values-id (in is the legacy
alias of id), so the APK only carried the (id) config. New devices
report "id" and match; older devices report "in" and find no (in)
config -> fall back to English (friend's report: only system date
showed Indonesian).
Fixes:
- Add values-in-rID (core/presentation + feature/cw) so the APK
keeps a real "in" language config; values-in and values-id both
get translatable="false" on the 27 entries that English marks
(aapt2 rejects those with multiple %-substitutions otherwise)
- Verified: aapt2 compile of values/values-in/values-id/values-in-rID
OK; :core:presentation:mergeDebugResources + :feature:cw:mergeDebugResources
BUILD SUCCESSFUL
Friend's device (system language Bahasa Indonesia) fell back to
English even though values-in exists. Modern Android devices report
the Indonesian locale as "id" (ISO-639-1 current code; "in" is the
legacy alias) and resource matching is strict. Added values-id as a
copy of values-in in core/presentation and feature/cw (both language
directories ship the same translations).
Verified: check_strings OK (9 files incl. values-id);
:app:compileDebugKotlin BUILD SUCCESSFUL. Not released (batch with
pending fixes).
Background: 8 bottom-nav items squeeze long English labels on narrow
screens. 4.5.1 introduces the 5+N pattern: 5 main tabs plus a fixed
6th "More" button that pops a second-level menu (spring bounce) with
the remaining pages.
Changes:
- MainScreen: nav split into main (<=5, screenOrder-driven) + more
(subMenuOrder); More button with popup panel + spring animation;
BackHandler closes the menu before navigating back
- MoreMenuPopup: bottom-end card, current page highlighted, scrim
click to dismiss
- UI Settings: page order card now has two zones (main menu, max 5,
Settings locked last with no drag handle / more menu); move buttons
between zones, drag-to-reorder within zones; new subMenuOrder pref
- Defaults: main = Satellites/Passes/Radar/Map/Settings,
more = Mutual/Roaming/CwDecode; old screenOrder migrates by
classifying pages against the default sub menu
- Hard-coded UI strings localized (Tracking/Lat/Lon/Qth/Connect/
Track/Stop/CW permission prompts) into EN/ZH/TR
- NEW Indonesian locale (values-in, 181 strings + cw module strings)
- user rule: every future release must update EN/ZH/TR/IN
- Version 4.5.1 (451)
Verified: check_strings.py OK (8 files, no bare apostrophes);
:app:compileDebugKotlin BUILD SUCCESSFUL locally.
Background: the transponder panel CW decoder (added by the upstream
fork author) used a lightweight Kotlin Bayesian engine (core/domain/cw,
kept untouched as a fallback). This change makes the panel use the
Morse Expert engine ported in 4.5.0, so both CW entry points share the
same decoder with a live waterfall.
Changes:
- New mini layout cw_panel_main.xml (waterfall 80dp + decoded text,
status line hidden but ID kept for controller lookup)
- MainActivity.onCreate overload with applyImmersive flag; panel binds
with false so the host window system bars are not touched
- CwDecoderPanel now embeds the mini layout via AndroidView and drives
the MainActivity controller: start/stop/reset map to the engine,
lifecycle follows panel expand (start) / collapse (release mic)
- radar module now depends on feature:cw (+ constraintlayout 2.2.1,
same as cw) for layout + controller reuse
- What's new rewritten in EN/TR/ZH for this release only
Verified: :feature:radar:compileDebugKotlin and :app:compileDebugKotlin
BUILD SUCCESSFUL locally; check_strings.py OK (7 files, no bare
apostrophes).
The waterfall showed mirrored/upside-down garbage because the FFT
never produced a valid spectrum:
1. g3.c.f() (high-precision sin) had its quadrant-0 case mangled by
jadx into a nested-if that returned NaN for small angles - the
twiddle factor table ended up with 329/1024 NaN entries.
Restored the smali switch: case0->g(), case1->c(), case2->-g(),
case3->-c().
2. i3.d.k() routed the runtime FFT (a5==0, single-thread path) into
the else of if(a5!=1) instead of if(a5!=0), so the FFT never ran
and the output stayed in the time domain (peak at bin 357 for a
669Hz tone instead of bin 86).
Verified with a JVM harness (static-block tables + pure-tone inputs):
200Hz->bin26, 400Hz->bin51, 669Hz->bin86, 1000Hz->bin128, all exact.
Twiddle table NaN count: 329 -> 0.
Decode page showed stats but empty waterfall and no decoded text:
the ported i3/d.k() FFT dispatch was broken by a jadx structure
misplacement - the runtime path (k=1 -> a5=0 -> cond_22 single-thread
FFT) was replaced by a hallucinated `throw null` else-branch while the
real cond_22 code sat in a dead else. Verified against smali
(7030-7263): j3.c.q forward FFT + post-processing loop + tail + small-
array branch now live in the a5==0 branch.
Also:
- UiSettingsCard now lists CwDecode (toggle + drag-reorder) between
Roaming and Map
- unknown screenIds in persisted screenOrder fall back to
defaultScreenOrder position (CwDecode lands between Roaming and Map
for existing users instead of trailing after Settings)
Verified: :feature:cw + :feature:settings + :app compileDebugKotlin
BUILD SUCCESSFUL.
Three release-breaking issues found in the v4.5.0 APK (app crashed on
launch, label showed "Morse Expert", dex shrank to 282KB vs 3.5MB):
1. app_name: the ported app_values.xml shipped a "Morse Expert"
app_name string which overrode Look4Sat Pro's label during resource
merging - removed (no other string collisions).
2. R8 stripped nearly all code: the in-app sun.misc.Unsafe/Cleaner
stubs clashed with android.jar library classes. Moved stubs to
com.rtbishop.look4sat.feature.cw.suncompat and updated k3.d/s/r
imports (k3.r keeps the reflective Class.forName("sun.misc.Unsafe")
string, which returns null on Android hidden-API limits).
3. proguard-rules.pro added (AGP 9 variant-level
CanProduceConsumerProguardFiles): keep pas.** (JNI RegisterNatives
resolves by class name) plus all ported CW classes.
4. app-level ndk abiFilters forced to armeabi-v7a: the ported
libnativedecoderjni.so is v7a-only, so a multi-ABI APK would crash
with UnsatisfiedLinkError on arm64 devices.
Verified: :feature:cw:compileDebugKotlin BUILD SUCCESSFUL.
Compose integration of the ported CW decoder engine:
- CwDecodeScreen: AndroidView embedding the ported activity_main.xml,
lifecycle delegated to the ported MainActivity controller (onCreate ->
onResume, onDispose -> onPause/onDestroy), RECORD_AUDIO runtime
permission flow (with permanent-denial -> app settings), original
options_menu actions as a top button row (pause/clear/save/record/
settings), double-back-to-exit preserved
- CwSettingsDialog: message_type (general_text/ham_radio_qso),
text_font_size (7-99), and the 9 color keys (bg_color/text_color/...)
reading/writing the same prefs keys as the original app
(getPackageName()+"_preferences"), colors sourced from I2.b tables
- Navigation: Screen.CwDecode ("CwDecode") placed between Roaming and
Map in the default order; defaultScreenOrder updated; ic_cw morse icon;
nav_cw strings (en/zh/tr); app depends on :feature:cw
Verified: :feature:cw:compileDebugKotlin + :app:compileDebugKotlin
BUILD SUCCESSFUL (first pass, no errors).
mckero