Segfault - null ptr deref in what appears to be strdup about jsc-android-buildscripts HOT 6 CLOSED

btw, I built my own binaries, changing --release to . --debug . and --enable-debug=yes.

However, I still wasn't able to get a better backtrace out of logcat. Any tips would be appreciated.

from jsc-android-buildscripts.

Just remove this line https://github.com/SoftwareMansion/jsc-android-buildscripts/blob/master/common.sh#L139 to preserve symbols in the build. You can also replace -s with -g to include GDB debug information.

Note that installArchives also strips debug symbols, so if you are using it you have to run adb logcat | $ANDROID_NDK/prebuilt/darwin-x86_64/bin/ndk-stack -sym . in a directory with libjsc.so copied directly from lib/distribution-?/jsc/lib/?/

from jsc-android-buildscripts.

Thanks. I think it was that 2nd piece that I was missing. I have encountered this trace on a few different occasions and the symptom went away when we fixed some type mismatch/coersion bug in javascript. I wonder if a later version of jsc has found and fixed this.

from jsc-android-buildscripts.

ok.. here is the latest trace with the x86 so :

Stack frame #00 pc 0001c7c2  /system/lib/libc.so (strlen+18)
Stack frame #01 pc 00084360  /system/lib/libc.so (__strlen_chk+32)
Stack frame #02 pc 000731aa  /system/lib/libc.so (strdup+42)
Stack frame #03 pc 008ef58c  /data/app/com.mobilemessenger-1/lib/x86/libjsc.so: Routine formatLocaleDate at /bitrise/src/target/webkit/Source/JavaScriptCore/runtime/DatePrototype.cpp:255 (discriminator 1)
Stack frame #04 pc 008ea291  /data/app/com.mobilemessenger-1/lib/x86/libjsc.so: Routine formatLocaleDate at /bitrise/src/target/webkit/Source/JavaScriptCore/runtime/DatePrototype.cpp:316
Stack frame #05 pc 008ea0e4  /data/app/com.mobilemessenger-1/lib/x86/libjsc.so: Routine dateProtoFuncToLocaleString at /bitrise/src/target/webkit/Source/JavaScriptCore/runtime/DatePrototype.cpp:582 (discriminator 1)

from jsc-android-buildscripts.

This seems to directly indicate that somewhere in our libraries we are using dateLocale operations and should therefore use the icu build. Interesting that it was able to get to this point in the code, though, I would have guessed that HAVE(LANGINFO_H) would have been false.

from jsc-android-buildscripts.

pointing android/app/build.gradle to the ICU build as prescribed in the README did remove the SEGV. It might be worth either documenting that toLocaleString() might crash in the documentation, or figure out how to actually make it work like toString in all cases. I can try to provide some help there if needed.

from jsc-android-buildscripts.