Base: 01bf71c28435bb3b81c4b033a78a8f6ae5bbb7f4 (merged PR #126).
drawStrokeLine submitted a path whenever the stroke width was positive,
including fully transparent colors. drawSolid always submitted its default
transparent fill. The fix skips these invisible submissions but preserves path,
corner and FILL-paint preparation: rounded image backgrounds and PressAdjuster
reuse that state. Visible/translucent colors, gradients and the historical
zero-gradient-endpoint fallback must remain unchanged.
Regression tests record commands from attached, measured views, including alpha 0/1/128/255, zero-width strokes, repeated frames, pressed-state resize/corners, image-background fills and transparent/visible setter transitions. Command counts are not a measurement of GPU overdraw layers or frame-time improvement.
The screenshot shows E/SuperTextView-onDraw(L:342): canvas.h = ....
d55309a^ has exactly LogUtils.e("canvas.h = " + canvas.getHeight()) at line 342,
and its LogUtils defaults DEBUG to true. Public JitPack 3.1.6.2 and v3.1.6.2
AAR bytecode corroborates the call and line number, rather than merely assuming
that today’s default-off DEBUG makes the old report invalid.
d55309a commented out this log, dbb843a removed it, and 58ae627 separately
changed DEBUG to false. Downloaded 3.2.3 and 3.2.5 AARs have no canvas.h
class constant. Current device tests repeat drawing 100 times and the host checks
logcat for the exact historical diagnostic. This does not promise silence from
all Android/system logs or real exceptional/fatal conditions.
| Artifact | ARMv7 libGifLib.so size |
ZIP CRC |
|---|---|---|
Public 3.2.3 / v3.2.3 AAR |
124768 bytes | valid |
Public 3.2.5 / v3.2.5 AAR |
0 bytes | valid |
The empty 3.2.5 library is also present in its Git tag. A valid ZIP checksum does not establish a valid native payload. The current library has four nonempty, architecture/16KB/RELRO-checked ABIs, already loaded in the separate native suite.
New protection is shared by the legacy library and standalone regression builds:
gradle/native-validation.gradle checks committed libraries before preBuild,
and validates the actual AAR after bundle* AAR tasks. The CLI’s --aar mode
checks the ABI set, duplicate native entries, ZIP CRC, empty/truncated ELF,
architecture, LOAD bounds/alignment and RELRO, and prints sizes/SHA-256.
CI also covers version-tag pushes and a packaged four-ABI AAR.
Source builds now require Python 3 (-PstvPython=/path/to/python3 overrides the
executable); consuming a published AAR does not execute this source-build guard.
The independent build exercises the same gate, but the legacy AGP 3.3/JCenter
project and a real JitPack publication are not migrated or published here.
The original report names 3.2.3. Today’s served artifact is nonempty and its Last-Modified is 2026-04-08: it cannot establish the reporter’s historical cache. Closing that original failure still needs the resolved dependency/AAR hash, final APK or splits, device/API/ABI and the installed library size/hash, to find the first mismatch along AAR → merged JNI → APK → installed file. No old tag or published artifact is rewritten.
SuperTextView delegates the supplied attributes to TextView and does not replace
Typeface. New XML fixtures compare TextView with SuperTextView inheriting
monospace + bold|italic through a style/textAppearance; normal, outlined,
gradient and combined drawing must preserve the typeface. The real-device suite
also compares actual framework font metrics.
Robolectric legacy Paint does not faithfully model reset/default style, alpha or
glyph metrics. Host tests therefore observe the original draw Paint’s Typeface
and meaningful path properties; pixel/fallback checks belong to the device suite.
This is protection for a verified working configuration, not proof that the
reporter’s unspecified font file/layout/system version has been reproduced.
System family names and pre-API-26 @font loading are different cases.
Separate binary-XML fixtures load PNG, JPEG, vector, shape, selector and GIF resources in both image slots; static images must produce pixels and GIFs must return real JNI metadata. A rounded PNG background additionally checks its opaque center and transparent corner. These exercise the production View and resources, not a demo substitute or a mocked native implementation.
Both original reports contain only the outer InflateException; the innermost cause and original layouts/images are still absent. Tests of current master do not identify the original root cause, and passing API 35 does not substitute for the reported API 31 environment. Do not call these original reports fixed merely because a nearby native/signature issue was addressed by PR #125/#126.
# JDK 11, Gradle 7.5 and Android platform 28; see regression-tests/README.md.
gradle -p regression-tests testDebugUnitTest assembleRelease assembleDebugAndroidTest
python3 -m unittest discover -s tools/tests -v
python3 tools/check_native_binaries.py --aar \
regression-tests/build/outputs/aar/SuperTextView-regression-tests-release.aar
# The library's instrumentation APK is self-contained; select a test device explicitly.
adb -s <serial> install -r \
regression-tests/build/outputs/apk/androidTest/debug/SuperTextView-regression-tests-debug-androidTest.apk
adb -s <serial> shell am instrument -w -r \
com.coorchice.library.test/com.coorchice.library.DeviceRegressionInstrumentation
sysconf=16384, application smaps KernelPageSize=16 kB.
Repeated drawing produces no historical canvas.h/SuperTextView-onDraw
diagnostic; the device crash buffer is empty.Raw device log, device/APK identity and host test totals. These are native-kernel virtual-device results, not physical OEM/API-31 coverage or a quantified GPU-performance measurement. The original root causes of #91, #113, #77 and the declared 3.2.3 instance of #107 remain unconfirmed.