Repository navigation
WaEnhancer Community 0.0.6
WaEnhancer Community 0.0.6
SHA-256: 0d3a7f427505d0012f87118b22eb737a0de90bf0fc3be4e4df0c100d7270bdf7
Changes since v0.0.5.
What's changed
- #56 fix: stop nullable prop hook exception storm and publish R8 mappings
- #52 fix: make diagnostics report runtime coverage honestly
- #58 Fix floating bar icon-only labels
Full changelog
This section compiles the descriptions of the pull requests included in this release.
#56 — fix: stop nullable prop hook exception storm and publish R8 mappings
Merged by contribution from @igorcv88.
Problem
WhatsApp 2.26.33.76 with Community 0.0.5 can produce an extreme LSPosed exception storm from an R8-obfuscated XC_MethodHook callback:
NullPointerException: Attempt to invoke virtual method 'int java.lang.Integer.intValue()' on a null object reference
The installed APK was verified against the official v0.0.5 SHA-256, so this is not the pre-PR #44 SeenTick bug.
The host resolver cache identifies the relevant methods as:
- Boolean props:
X.00D.A0A(X.00F, X.00D, Boolean, int) - Integer props:
X.00D.A01(X.00F, X.00D, Integer, int)
The regression was introduced by e85848f when Others.hookProps() changed from scanning populated runtime Integer arguments to precomputing the first int/Integer parameter index. For A01, that selects the nullable boxed Integer at index 2 instead of the primitive int field id at index 3. Direct (int) casting then auto-unboxes null on a very hot startup/property path, producing the log flood.
Changes
- add
HostArgCompat.findPreferredIntParameterIndex():- prefer a unique primitive
intslot; - fall back to a unique boxed
Integeronly when no primitive exists; - reject ambiguous signatures instead of guessing;
- prefer a unique primitive
- add
HostArgCompat.integerAt()so callbacks fail open on null or incompatible runtime values rather than auto-unboxing; - harden both Boolean and Integer property hooks:
- validate the host signature once when installing the hook;
- skip the individual hook with one diagnostic if no unambiguous property-key slot exists;
- read only the validated slot on the hot path;
- remove the direct nullable unboxing and runtime argument scan;
- add regression tests for the exact WhatsApp 2.26.33.76 shape
(..., Integer, int)with the boxed slot null, boxed-only compatibility, and ambiguous signatures; - replace the separate CI/release workflows with one manual Build APK Release workflow;
- run wrapper validation, release-note generator validation, unit tests and the signed release build in the same runner invocation;
- verify signing certificate, application id, versionName/versionCode and SHA-256 before publishing;
- publish the release R8
mapping.txtalongside the APK/checksum so future obfuscated frames such asZ.t81can be retraced directly; - write the release URL and verification summary into the workflow summary.
Workflow consolidation
.github/workflows/ now contains only build-apk-release.yml.
There is no longer an automatic PR/push CI workflow and no separate debug APK assembly. The manual release workflow executes:
./gradlew clean :app:testWhatsappDebugUnitTest :app:assembleWhatsappRelease ...
so tests and the APK release build share the same runner/Gradle invocation and cache. This avoids paying for a second workflow/build solely to repeat checks before a release.
Expected behavior
For X.00D.A01(X.00F, X.00D, Integer, int), the property key resolves to index 3. A null index-2 Integer is never touched by the hook, so the Integer.intValue() exception storm disappears without wrapping the callback in a blanket try/catch or suppressing LSPosed logs.
If WhatsApp later changes the accessor to an ambiguous integer signature, WAEnhancer leaves that accessor untouched instead of risking a semantically wrong override.
Separate compatibility issue
loadNewMessageMethod() remains an independent optional compatibility issue: the current resolver still depends on the removed extra_payment_note anchor. This PR intentionally does not guess a replacement accessor without host-APK evidence. FMessageWpp already treats that capability as optional and fails open to the media accessor; a semantic replacement resolver should be handled separately after identifying the actual 2.26.33 method.
#52 — fix: make diagnostics report runtime coverage honestly
Merged by contribution from @igorcv88.
Problem
The functional diagnostics report still looked static after the runtime snapshot fix because the catalog contains many more features than currently emit callback-level diagnostic telemetry.
Only these features currently call diagnosticTriggered():
- HideSeen
- AntiRevoke
- ViewOnce
- DownloadViewOnce
- FreezeLastSeen
Everything else was shown as NOT_EXERCISED, which incorrectly implied that the user could make the state change by toggling/exercising the feature. For entries such as SeenTick and FloatingBottomBar there is currently no runtime callback telemetry at all, so no user action can make them become TRIGGERED yet.
The report also displayed behavior NOT VERIFIED for every catalog entry marked manual even though DiagnosticsActivity only exposes manual confirmation controls for the required smoke-suite features.
Fix
- explicitly track which catalog entries have real runtime callback probes
- report uninstrumented entries as
NO_RUNTIME_PROBEinstead ofNOT_EXERCISED - show callback-probe coverage at the top of the report
- explain that changing a setting cannot change an uninstrumented diagnostic entry
- add exact exercise instructions for each currently instrumented feature
- only show manual behavior verification for the required smoke-suite controls that actually exist in DiagnosticsActivity
- only list a required manual test as pending after its runtime callback has actually triggered
Expected behavior
A report should now make the distinction explicit:
TRIGGERED: an instrumented hook callback executed during the active sessionNOT_EXERCISED: the feature has a runtime probe but the required callback has not happened yetNO_RUNTIME_PROBE: the feature is cataloged for regression testing but runtime callback telemetry has not been implemented yet
This PR does not pretend that all catalog features are validated. It makes the current coverage truthful so future instrumentation work can be added feature by feature without misleading the user.
Validation
./gradlew :app:testWhatsappDebugUnitTest :app:assembleWhatsappDebug#58 — Fix floating bar icon-only labels
Merged by contribution from @igorcv88.
Makes Text size = 0 an explicit icon-only mode for the Floating Bottom Bar.
- expands the text-size slider range to include 0
- removes label height and icon-label spacing from shared geometry at 0sp
- lets the existing runtime
labelVisiblepath set WhatsApp tab labels toGONE, instead of leaving invisible label layout behind - keeps all existing presets/defaults unchanged
- adds JVM coverage for icon-only geometry and measured-label override handling
This is aimed at the last-page/Calls active-indicator deformation seen when labels are visually absent but still participate in the tab item layout.
Other changes
These commits reached master without an associated merged pull request:
fb623eaci: consolidate release checks into one workflow8ce98b8ci: remove superseded release workflowa55c20fci: remove redundant automatic CI workflowea80ef5docs: retarget build badge to consolidated release workflow43b680ffix: add semantic host resolver fallbacksa291f47fix: probe message accessor through compat resolver0ce080efix: recover FMessage text accessor on 2.26.33a7d8982fix: resolve conversation delegate through host compat59d2ecbfix: make LSPosed marker best effort9824f18fix: adapt profile status hook to current host UI417447cfix: harden semantic resolver against DexKit API drift37bb929fix: keep audio type behavior unchanged629ba85test: cover delegate interface storage resolution3957b7cfix: tighten host delegate fallback matching2696dd8test: harden delegate fallback coverage9fe8c70ci: compress R8 mapping release assete4f904eEnable GitHub Sponsors funding