Introduction
It might happen so that there is a bug in your release build while in debug everything works fine. In many cases, it might be an issue with Proguard/R8 configuration. Of course, it is better to test your code thoroughly, properly configure Proguard if you, let’s say, load some classes only via reflection and so on. But reality sometimes strikes and bugs might go to production.
In this case, the first thing that is needed is to find a bug and fix it. And only then have some retrospective to mitigate such situations in the future.
When the bug is in production already every minute counts, therefore it is important to have some plan. Proguard configuration seems complex to someone who didn’t work with it, so I recommend to take a look at its main features.
Most of the time I’d say that issue is easy can be found and fixed by analyzing crash report logs. But sometimes the log is not that clear.
In this article I’d like to introduce the plan to find what part of “Proguard” causing an issue, so you can debug more effectively.
Proguard Basics
Just a quick overview of what we have with Proguard/R8 and what might cause issues in release builds.
The basic setup for release build is:
buildTypes {
release {
minifyEnabled true
shrinkResources true
proguardFiles getDefaultProguardFile('proguard-android-optimize.txt'), 'proguard-rules.pro'
}
}
Flag minifyEnabled
controls whether obfuscation and code optimization is enabled when shrinkResources
controls the optimization of resources (and can be set true only if minify enabled). Therefore there are at least three parts of “Proguard” that might cause issues when used incorrectly (or not correctly configured):
- code obfuscation - replacing names with shorter variants. Makes it more difficult to reverse engineer the app and makes bytecode smaller.
- code optimization - including but not limited to inlining code blocks, removing unused classes, methods, and so on
- resources optimization - removing unused resources, etc.
One of the most advanced ways of investigating issues in release builds is analyzing the content of the APK file. This can be done via Build > Analyze APK ...
.
There one can see classes.dex
containing all the bytecode, resources.arsc
which contains the mapping between original resources IDs and obfuscated and all resources under /res
folder.
This method is advanced because for example bytecode will be shown to you as:
.class public final Lcom/krossovochkin/proguardtest/MainActivity;
.super La/b/c/e;
.source ""
# direct methods
.method public constructor <init>()V
.registers 1
invoke-direct {p0}, La/b/c/e;-><init>()V
return-void
.end method
# virtual methods
.method public onCreate(Landroid/os/Bundle;)V
.registers 2
invoke-super {p0, p1}, La/b/c/e;->onCreate(Landroid/os/Bundle;)V
const p1, 0x7f0b001c
invoke-virtual {p0, p1}, La/b/c/e;->setContentView(I)V
return-void
.end method
Which while not being much readable still might provide you some hints.
To debug Proguard/R8 configuration one can assemble release build and navigate to app\build\outputs\mapping\release
. There you’ll see the following files:
configuration.txt
– merged file with all configurations – from your app, default Android, AAPT, all the libraries, etc. Here you can find what rules might cause an issue.mapping.txt
– file with mappings of original names to obfuscated ones. This might help you analyzing logs and the content of the APK file.seeds.txt
– file with kept files/classes/etc. You can verify here that some particular file not being removed by R8usage.txt
– opposite to seeds – what was removed. Here you can see whether some class you need was removed
Debugging algorithm
Here we’ll see steps of the algorithm to define what part of release optimizations is responsible for the issue, and where you should look deeply to find the root cause.
Step 1 - Ensure issue related to release optimizations
First of all, we need to be sure that the issue we have is because of some misconfiguration of the release build. To confirm that we just need to disable release optimizations and check whether we still have a bug or it is disappeared.
For this one need to disable minification and shrinker:
minifyEnabled false
shrinkResources false
If the issue still reproduces then the issue is not with proguard. Look for some usages of BuildConfig.DEBUG
in your codebase. Dive deep into logs.
Additionally, you can make your release build debuggable by placing in your release config:
debuggable true
Anyway, there is nothing to do with Proguard here. And that is partially good because you don’t have to look at proguard configuration.
If after disabling release optimizations you see that issue doesn’t reproduce anymore, then something is wrong with release optimization configurations. Go to the next step.
Step 2 - Check whether it is a shrinker issue
Next, we’ll check whether we have the issue because of shrinker configuration or not. To do that we need to enable minification while keeping the shrinker disabled:
minifyEnabled true
shrinkResources false
If the issue disappeared then indeed we have some issue with configuring shrinker. Probably it removed something we’ve relied on. Look at the resources.txt
file looking for the resources you access dynamically. Check whether these resources are added to keep.xml
.
If the issue still happens then go to the next step.
Step 3 - Check whether it is an obfuscation issue
To check whether it is an issue with obfuscation we need to disable it.
Inside your proguard-rules.pro
file add the line to direct Proguard to disable obfuscation:
-dontobfuscate
If the issue doesn’t reproduce anymore then there is some issue because of obfuscation.
Look for classes or methods you use via reflection and looking at mapping.txt
check that classes you access via reflection are not obfuscated. If they are obfuscated, then you need to add some keep
rules in the proguard-rules.pro
.
If the issue doesn’t reproduce, then it is most likely related to code optimizations. Probably you access some classes via reflection only and that class was removed during release optimizations. For that check usage.txt
to ensure. If class or method indeed was removed, then again you probably need to add some keep
rules to your proguard configuration.
Examples
Let’s look at the examples of various issues that might happen.
App issue
Suppose we have in our project some code like:
class MainActivity : AppCompatActivity() {
override fun onCreate(savedInstanceState: Bundle?) {
super.onCreate(savedInstanceState)
setContentView(R.layout.activity_main)
if (!BuildConfig.DEBUG) {
throw RuntimeException()
}
}
}
Running it in debug has no issue, while in release app crashes:
Caused by: java.lang.RuntimeException
at com.krossovochkin.proguardtest.MainActivity.onCreate(:1)
Going with our algorithm:
- disable release optimizations - the issue still reproduces in the release build, therefore it doesn’t relate to proguard configuration and we need to check our logs and debug the app itself.
We quickly see that we have incorrect logic for the release build, fix and we’re done.
Shrinker issue
It is a bit tricky. Shrinker is doing its best to not remove resources that even might be accessed dynamically. So if we have in our code something like:
private fun shrinker() {
val id = resources.getIdentifier("test", "layout", packageName)
setContentView(id)
}
or even:
// ...
shrinker('t')
// ...
private fun shrinker(char: Char) {
val id = resources.getIdentifier("${char}est", "layout", packageName)
setContentView(id)
}
Shrinker won’t remove the test.xml
layout resource.
Usually, an issue might happen when we receive ids from outside of the APK (say from the server). For our test example we can do some weird calculation which won’t be optimized:
shrinker(Class.forName("com.krossovochkin.proguardtest.MainActivity").simpleName[6])
If we launch the release build app will crash. In resources.xml
we’ll see the message:
Skipped unused resource res/layout/test.xml: 880 bytes (replaced with small dummy file of size 104 bytes)
And if we look inside the APK we’ll see that our resource will look like:
<?xml version="1.0" encoding="utf-8"?>
<x />
So, going with our algorithm it will look like this:
- disable release optimizations - the issue still here
- enable minification while keeping shrinker enabled - no crash
Therefore we know that the issue is with shrinker removing unused resource. To fix that we need to add our resource to /res/raw/keep.xml
like this:
<?xml version="1.0" encoding="utf-8"?>
<resources xmlns:tools="http://schemas.android.com/tools"
tools:keep="@layout/test" />
We enable release optimizations, check once again and it works fine. Cool!
Obfuscation issue
Obfuscation issues usually happen when we rely on some class/methods/etc names. For example:
private fun obfuscate() {
if (TestClass::class.java.name != "com.krossovochkin.proguardtest.TestClass") {
throw RuntimeException("obfuscate")
}
}
When running the app in the release, it crashes.
Going with our algorithm:
- disable release optimizations - app crashes
- enable minification while keeping shrinker disabled - app crashes
- disable obfuscation in
proguard-rules.pro
- no crash.
We check mapping.txt
and see that our class was obfuscated (which we don’t want to happen). So we need to make an exception in our proguard rules. For example, add to proguard-rules.pro
:
-keepnames class com.krossovochkin.proguardtest.TestClass
And the issue is gone.
Optimization issue
Optimization issues usually happen when we rely on some class via reflection only and it was removed during code optimization. For example, if we have in our app:
private fun optimizer() {
Class.forName("com.krossovochkin.proguardtest.TestClass").newInstance()
}
class TestClass {
var value: Int = 5
}
Then everything is fine. Because code optimization is smart and it doesn’t remove as much as it could. It can determine that class is used by analyzing strings we have in our app. So, let’s emulate the issue by hiding the exact class we need:
private fun optimizer(char: Char) {
Class.forName("com.krossovochkin.proguardtest.${char}estClass").newInstance()
}
App crashes in release because there is no such a class in the release build.
We even can see that by looking at logs:
Caused by: java.lang.ClassNotFoundException: com.krossovochkin.proguardtest.TestClass
But sometimes it might be not so obvious, so we can stick to our algorithm:
- disable release optimizations - app crashes
- enable minification keeping shrinker enabled - app crashes
- disable obfuscation - app crashes
Here we understand that issue is because of code optimizations. We look at usage.txt
and see that our class was removed as a part of code optimization. So, we need to keep our class explicitly by adding to our proguard-rules.pro
:
-keep class com.krossovochkin.proguardtest.TestClass
And the issue is fixed.
Conclusion
Proguard configuration issues might be annoying and not easily spotted. Therefore it is good to be prepared and know what to do when you face the issue. The algorithm provided in the article not only allows you to concentrate quickly on the part of release optimizations that most likely cause the issue but also gives you a clearer inside on what release optimizations exist and how they can impact your release build.
Also, I recommend to look at Proguard configuration options to know what can be done and what various configurations mean. And take a look at Android documentation for release optimizations configuration.
Happy coding