Develop Reference

Native Code coverage with android soong build system

I'm trying to generate code coverage report for my native components with AOSP source code using soong build system.
I have extended aosp vhal but unit test cases are same as in below link.
http://androidxref.com/8.1.0_r33/xref/hardware/interfaces/automotive/vehicle/2.0/default/tests/
Tried adding below to cc_test, cc_binary in Android.bp
native_coverage : true,
cflags: [
"-g",
"-O0",
"-fprofile-arcs",
"-ftest-coverage",
],
ldflags : [
"-fprofile-arcs",
"-ftest-coverage",
],
Native binary unit-tests-coverage is generated in out/target/product but I can't find gcno intermediates for this.
Running below command gives me *.gcda files for each test files.
adb shell \
GCOV_PREFIX=/data/local/tmp \
GCOV_PREFIX_STRIP=`echo $ANDROID_BUILD_TOP | grep -o / | wc -l` \
/data/local/tmp/unit-tests-coverage
I have tried below links but not sure how to proceed :(
http://logan.tw/posts/2015/04/28/check-code-coverage-with-clang-and-lcov/
https://android.googlesource.com/platform/bionic.git/+/master-soong
https://android.googlesource.com/platform/build/soong/+/581341d%5E%21/
https://android.googlesource.com/platform/external/e2fsprogs/+/fedfb27%5E%21/
https://android.googlesource.com/platform/development/+/master/scripts/acov#23
http://androidxref.com/9.0.0_r3/xref/bionic/README.md#293
I'm not sure if google's vts framework can be used here to generate native code coverage.
https://codelabs.developers.google.com/codelabs/android-vts-8/#6
"gcnodir" is generated but not sure how to make use of it.
/coverage/data/nativetest64/vehicle-unit-tests-coverage/unit-tests-coverage.gcnodir

Posting answer to my question for other users on SO.
Install coverage tool :
sudo apt-get install lcov (This should install lcov-1.12)
sudo apt-get install gcc-4.6 (Clang generates .gcno approximately equal to gcc 4.2 that aren't compatible
with gcov-4.8. Installing gcc-4.6 to get gcov-4.6 and invoking lcov with '--gcov-tool /usr/bin/gcov-4.6')
Download LLVM 3.8 for llvm-cov to work : http://releases.llvm.org/download.html
All native unit test cases i.e instrumented binary needs to be executed on target. To build and emit clang's instrumentation based profiling. Example: http://androidxref.com/9.0.0_r3/xref/hardware/interfaces/automotive/vehicle/2.0/default/Android.bp#82 (Renamed to vehicle-manager-unit-test for shorter name)
export NATIVE_COVERAGE=true
Add native_coverage: true to test module in Android.bp
Go to: module-name/test
Use mm or make command to build native binary
Ex: For hardware/interfaces/automotive/vehicle/2.0/default/tests/ :
mma or make vehicle-manager-unit-test -j32
Copy coverage enabled instrumented binary to target
adb push out/target/product/product_name/data/nativetest64/vendor/vehicle-manager-unit-test /data/nativetest64/vehicle-manager-unit-test
adb shell chmod +x /data/nativetest64/vehicle-manager-unit-test
Run test cases and generate .gcda files
adb shell \
GCOV_PREFIX=/data/local/tmp \
GCOV_PREFIX_STRIP=echo $ANDROID_BUILD_TOP | grep -o / | wc -l \
/data/nativetest64/vehicle-manager-unit-test
adb shell find -iname *.gcda
adb pull /data/local/tmp/proc/self/cwd/out/soong/.intermediates/hardware/interfaces/automotive/vehicle/2.0/default/vehicle-manager-unit-test/android_x86_64_silvermont_vendor_cov/obj/hardware/interfaces/automotive/vehicle/2.0/default/tests/ .(Destination folder)
Extract GCNO files from GCNODIR (archive file generated at
out/overage/data/nativetest64/vendor/vehicle-manager-unit-test ) to
same folder with GCDA files
llvm-cov gcov -f -b *.gcda (https://llvm.org/docs/CommandGuide/llvm-cov.html )
lcov --directory . --base-directory . --gcov-tool /usr/bin/gcov-4.6 --capture -o cov.info (http://ltp.sourceforge.net/coverage/lcov.php)
genhtml cov.info -o output
Here's the script which wraps all these commands:
https://gist.github.com/pankajgangwar/f070b8b54e83543f8e3638dcd2cae1b8

here it is explaned how to generate coverage reports, which do require GTest:
these flags enable the generation of test coverage: -fprofile-arcs -ftest-coverage
then one has to use gcov: gcov main_test.cpp
which's output then can be passed on to lcov (for reference):
$ lcov --coverage --directory . --output-file main_coverage.info
from which one can generate an lcov coverage report in HTML format:
$ genhtml main_coverage.info --output-directory out
these .gcda files in .gcnodir are gcov data files. gcov also has an output option --json-format, which might come handy when wanting to consume the coverage data with a web-service.
one of the examples from the links you've provided can be used to generate it for a whole project:
Collect the code coverage results:
$ lcov --directory . \
--base-directory . \
--gcov-tool gcov.sh \
--capture -o cov.info
Generate HTML files:
$ genhtml cov.info -o output
where the only difference is, that the wrapper script would need to be adjusted to call gcov. probably one could even omit the wrapper passed with option --gcov-tool, since it should be directly called.
since one can only prepare the coverage report by adding the compiler flags, the gcov and lcov commands should be setup as post-build script, so that they would automatically generate the report.

subprocess.check_output fails to execute a command but the same works in windows

I am trying to connect two devices to my pc and run some commands on them using python and adb.
when I run the command from the command prompt, it goes through fine, but when i put those in the python script, they give me errors.
this is causing the errors all the time:
from subprocess import check_output, CalledProcessError
try:
adb_ouput = check_output(["adb","devices","-l","|", "grep", "\"model\""])
print adb_ouput
except CalledProcessError as e:
print e
The error message I get is this:
Usage: adb devices [-l]
Command '['adb', 'devices', '-l', '|', 'grep', '"model"']' returned non-zero exit status 1
When I try the same code without the grep command, it works
adb_ouput = check_output(["adb","devices","-l"])
It gives me the right output.
When I try the same in windows command prompt, It works fine (I am replacing grep with FINDSTR because i'm using it in windows, and I have tried doing the same in the python script as well, with 'shell = True' and also without.)
eg:
adb devices -l | FINDSTR "model"
This gives me an ouptut without any problems.
The output I get is
123ab6ef device product:xxxxxxxxx model:xxxxxxxxx device:xxxxxxxxx
bd00051a4 device product:yyyyyyyyyy model:yyyyyyyyyy device:yyyyyyyyy
I am trying to understand where I am going wrong here, but can't figure it out.
So far I have checked the docs: https://docs.python.org/3/library/subprocess.html
https://docs.python.org/3/library/subprocess.html#subprocess.CalledProcessError
These just give me the error codes.
I have also looked at these answers:
Python, adb and shell execution query
I took some error checking from here and added to my code.
Python subprocess.check_output(args) fails, while args executed via Windows command line work OK
python check_output fails with exit status 1 but Popen works for same command
I think i am close but just can't put my finger on it.
Any help would be appreciated.

First
adb_ouput = check_output(["adb","devices","-l","|", "grep", "\"model\""])
certainly requires shell=True, but even with that it's not equivalent to
adb devices -l | FINDSTR "model"
When using check_output, you're passing "model" as grep argument literally but you should pass just model. "model" is not in your output (with quotes) so grep fails to find it, and returns exitcode 1 which is not really an error for grep but makes check_output trigger an exception because it expects 0.
So I would to this as a quickfix:
adb_ouput = check_output(["adb","devices","-l","|", "grep", "model"],shell=True)
And as a longfix I'd perform the grep command directly with python.
adb_output = check_output(["adb","devices","-l"])
for l in adb_output.splitlines():
if "model" in l:
print(l)

Cross compiling "Hello World" on Mac for Android

I'm trying to build a standard "Hello, World!" command-line executable for Android. The executable is to be run via adb shell.
0. The Go (Golang) Source
package main
import (
"fmt"
)
func main() {
fmt.Println("Hello, world!")
}
1A. The Build Command
$ CGO_ENABLED=0 GOOS=android GOARCH=arm GOARM=7 go build .
1B. The Output (Line Breaks Rearranged to Prevent Scrollbars)
# github.com/asukakenji/cross
warning: unable to find runtime/cgo.a
/usr/local/go/pkg/tool/darwin_amd64/link: running clang failed: exit status 1
ld: warning: ignoring file
/var/folders/dd/6k6vkzbd6d5803xj9zkjdhmh0000gn/T/go-link-150305609/go.o,
file was built for unsupported file format
( 0x7F 0x45 0x4C 0x46 0x01 0x01 0x01 0x00
0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 )
which is not the architecture being linked (x86_64):
/var/folders/dd/6k6vkzbd6d5803xj9zkjdhmh0000gn/T/go-link-150305609/go.o
Undefined symbols for architecture x86_64:
"_main", referenced from:
implicit entry/start for main executable
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
1C. The Build Command, Again
The following command gives the same result:
$ env CGO_ENABLED=0 GOOS=android GOARCH=arm GOARM=7 go build .
2. The Build Command (Verbose)
I've tried using "-v" as mentioned like this:
$ CGO_ENABLED=0 GOOS=android GOARCH=arm GOARM=7 go build \
-x -ldflags "-extldflags -v" .
It gives me more than 100 lines of messages, so I don't post it here unless it's necessary. The go build command seems to try compiling the source with the clang bundled with Xcode.
3A. The Build Command (Successful, but...)
Given the hint that the wrong compiler is found, I tried to set $CC like this:
$ CGO_ENABLED=0 GOOS=android GOARCH=arm GOARM=7 \
CC=/path/to/arm-linux-androideabi/bin/clang go build .
arm-linux-androideabi is the output from make_standalone_toolchain.py (or make-standalone-toolchain.sh).
3B. The Output
The executable (named cross) is successfully built, with the following messages:
# github.com/asukakenji/cross
warning: unable to find runtime/cgo.a
I tried adb push it and run it with adb shell on Android, it worked fine.
My Questions
Why does it need a C compiler? Doesn't Go cross-compile out-of-the-box?
When building for Linux (instead of Android), the compilation works fine:
$ CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build .
Why?
The go build command keeps looking for runtime/cgo.a, even when I didn't use CGO in the source code, and even when I set CGO_ENABLED=0. How can I get rid of the warning? How is it harmful not having one?

if you run that code you find the android as an official target platform
listed as GOOS/GOARCH
$go tool dist list

Android isn't official target platform for cross-compilation. If all you need are command-line executables then you can set GOOS=linux because android is a linux under the hood, else take a look at https://github.com/golang/go/wiki/Mobile

The cgo requirement might be because go requires libc for DNS lookups on Android: https://github.com/golang/go/issues/8877

You need to use Android NDK to compile for android, you can download it from the link, or from Android Studio:
Then you can find the compilerlink in the route as below:
Last login: Fri Sep 4 09:25:16 on console
The default interactive shell is now zsh.
To update your account to use zsh, please run `chsh -s /bin/zsh`.
For more details, please visit https://support.apple.com/kb/HT208050.
Hasans-Air:~ hajsf$ pwd
/Users/hajsf
Hasans-Air:~ hajsf$ cd Library
Hasans-Air:Library hajsf$ cd android
Hasans-Air:android hajsf$ cd sdk
Hasans-Air:sdk hajsf$ cd ndk
Hasans-Air:ndk hajsf$ ls
21.3.6528147
Hasans-Air:ndk hajsf$ cd 21.3.6528147
Hasans-Air:21.3.6528147 hajsf$ cd toolchains
Hasans-Air:toolchains hajsf$ cd llvm
Hasans-Air:llvm hajsf$ cd prebuilt
Hasans-Air:prebuilt hajsf$ ls
darwin-x86_64
Hasans-Air:prebuilt hajsf$ cd darwin-x86_64
Hasans-Air:darwin-x86_64 hajsf$ cd bin
Hasans-Air:bin hajsf$ pwd
/Users/hajsf/Library/android/sdk/ndk/21.3.6528147/toolchains/llvm/prebuilt/darwin-x86_64/bin
Then you can use the one reflectiong:
Archeticture you want to build your android app for, like aarch64
Android API you want to compile for, like Android 30
You can see full list of availbe options there, and you can seect the one you want like aarch64-linux-android30-clang
If you are at Windows 10 you'll find it at:
"C:\Users\${user}\AppData\Local\Android\Sdk\ndk\${NKD_version}\toolchains\llvm\prebuilt\windows-x86_64\bin\aarch64-linux-android30-clang"
There are 4 avaiailble linkers which are:
//CC_FOR_TARGET=/Users/hajsf/Library/android/sdk/ndk/21.3.6528147/toolchains/llvm/prebuilt/darwin-x86_64/bin/aarch64-linux-android30-clang
//CC_FOR_TARGET=/Users/hajsf/Library/android/sdk/ndk/21.3.6528147/toolchains/llvm/prebuilt/darwin-x86_64/bin/i686-linux-android30-clang
//CC_FOR_TARGET=/Users/hajsf/Library/android/sdk/ndk/21.3.6528147/toolchains/llvm/prebuilt/darwin-x86_64/bin/x86_64-linux-android30-clang
//CC_FOR_TARGET=/Users/hajsf/Library/android/sdk/ndk/21.3.6528147/toolchains/llvm/prebuilt/darwin-x86_64/bin/armv7a-linux-androideabi30-clang
After that only you can do cross copiling, as:
$ CGO_ENABLED=1
$ GOOS=android
$ GOARCH=arm64
$ CC_FOR_TARGET=/Users/hajsf/Library/android/sdk/ndk/21.3.6528147/toolchains/llvm/prebuilt/darwin-x86_64/bin/aarch64-linux-android30-clang
$ go build -buildmode=c-shared -o lib-aarch64-android30.so lib.go
And a simple cgo file lib.go could be:
package main
import "C"
import "fmt"
//export HelloWorld
func HelloWorld() {
fmt.Printf("hello world from GO\n")
}
//export GetKey
func GetKey() *C.char {
theKey := "123-456-789"
return C.CString(theKey)
}
func main() {}
As shown copilation completed successfuly, and both lib-aarch64-android30.so and ib-aarch64-android30.h had been generated without any error.
Quick note, not to go far from the scope of the question, if you return a string, then the return value from this function must be explicitly freed in the C code if if you call it from C code, but as you call it from garbage collector environment, Java/Kotlin you do not want to worry about it.
If freeing the allocated buffer isn't convenient, its common to fill a buffer provided by the caller:
func GetKey(buff *C.char, n int) int
If you can allocate the memory but don't want to handle C strings, you can insert the buffer into a pointer and return the size.
func GetKey(buff **C.char) int

Trouble calling bash from within python

I am trying to use python to help do some automation around an incremental build function in the Android build system. Generally, from a given directory, I would execute the following command to build whatever is in that directory and subdirectories:
mm -j8
This is analogous to a "make" command, only it is incremental build and is defined as a function in a bash file called envsetup.sh. What is does it not important, just know that it's a function defined in a bash script somewhere in the file system. To execute this, I can also do:
bash -c ". /path/to/envsetup.sh; mm -j8"
This method of calling it will be important in calling the function from python. I have followed the solution here which shows how to call a function within a bash script from python. I have used this method in a simple script that, in theory, should just spit out the STDOUT and STDERR from executing the command:
import subprocess
command = ['bash', '-c', '. /path/to/envsetup.sh; mm -j8']
(stdout, stderr) = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=True).communicate()
print 'stdout: ' + stdout
print 'stderr: ' + stderr
The call to Popen, however, never returns. What am I doing wrong that would allow bash to execute the command properly, but Python hangs when executing the command?

tl; dr:
Your issue is the use of shell=True. Set it to shell=False and it'll work.
With this option set, python will just run the first element of the command array, i.e. bash as a shell script. So currently, python is launching a shell of its own, in order to run your command (bash). It'll run bash with no arguments, and bash will then wait for input, blocking your python script.
The shell=True setting is for use cases where you are passing a shell script in as a single string. When you're explicitly specifying a shell and its parameters as the process to invoke, as you are doing above, you should set shell=False.
>>> import subprocess
>>> subprocess.Popen(['bash', 'whatever'], stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=True).communicate()
Here's what the proces tree looks like when I run the above:
\_ python
\_ /bin/sh -c bash whatever
\_ bash
The whatever is actually passed in, but it's a parameter to the sh, not a parameter to the inner bash, so the command being run is effectively ['/bin/sh', '-c', 'bash', 'whatever'], which is quite different from ['/bin/sh', '-c', 'bash whatever']

Debugging Android NDK native apps

I'm trying to debug and step through an Android application that segfaults. I've tried ndk-gdb, but with little luck. I've also referred to Android NDK Debugging without being able to debug my app.
When I try ndk-gdb --start, and I get:
$ ndk-gdb --start --verbose
Android NDK installation path: /opt/android-ndk-r7
Using default adb command: /opt/android-sdk-linux/platform-tools/adb
ADB version found: Android Debug Bridge version 1.0.29
Using final ADB command: '/opt/android-sdk-linux/platform-tools/adb'
Using auto-detected project path: .
Found package name: com.example.native_plasma
ABIs targetted by application: armeabi armeabi-v7a
Device API Level: 10
Device CPU ABIs: armeabi-v7a armeabi
Compatible device ABI: armeabi-v7a
Found debuggable flag: true
Found device gdbserver: /data/data/com.example.native_plasma/lib/gdbserver
Using gdb setup init: ./libs/armeabi-v7a/gdb.setup
Using toolchain prefix: /opt/android-ndk-r7/toolchains/arm-linux-androideabi-4.4.3/prebuilt/linux-x86/bin/arm-linux-androideabi-
Using app out directory: ./obj/local/armeabi-v7a
Found data directory: '/data/data/com.example.native_plasma'
Found first launchable activity: android.app.NativeActivity
Launching activity: com.example.native_plasma/android.app.NativeActivity
## COMMAND: /opt/android-sdk-linux/platform-tools/adb shell am start -n com.example.native_plasma/android.app.NativeActivity
Starting: Intent { cmp=com.example.native_plasma/android.app.NativeActivity }
## COMMAND: /opt/android-sdk-linux/platform-tools/adb shell sleep 2
Found running PID: 0
ERROR: Could not extract PID of application on device/emulator.
Weird, this probably means one of these:
- The installed package does not match your current manifest.
- The application process was terminated.
Try using the --verbose option and look at its output for details.
This indicates that the application segfaulted more less, but I don't know how to set a breakpoint here since gdb never actually gives a prompt.
I also tried this command:
$ ../../toolchains/arm-linux-androideabi-4.4.3/prebuilt/linux-x86/bin/arm-linux-androideabi-addr2line -f -e libs/armeabi/libnative-plasma.so
bedb2330
??
??:0
I have debug symbols I believe.
ndk-build -B V=1 APP_OPTIM=debug
Android.mk in jni/ has LOCAL_CFLAGS := -g
ant debug
I've also ndk-build NDK_DEBUG=1 but I still get where it looks like I don't have debug symbols.
Here's an image of the stack trace. It doesn't get any more informative:

Well NDK_DEBUG=1 and debuggable flag in manifest set to true are required. When you build the app,in your project/libs/armeabi, there should be a gdb.setup file. There is symbol search path there, check whether it is valid. And did you try this:
ndk-gdb --start --verbose --force
And looks like you are getting a null pointer exception.

In latest versions of NDK and Eclipse plug-in you can right click on package and choose Debug as -> Android Native Application

Make sure that you load your native library in either a launchable activity or in your Application class. Otherwise it wouldn't work and you'll get the following error No symbol table is loaded. Use the "file" command..
For example in Application class:
import android.app.Application;
public class MyApp extends Application {
static {
System.loadLibrary("Name");
}
public static native int doSomething();
}
Name is the name of your library (.so file) without the lib part.

I resolved putting --nowait option to the shell command:
ndk-gdb --start --verbose --nowait