Technique:Recognizing code obfuscation/Java and Android obfuscators
Recognition notes for Java and Android obfuscation: ProGuard/R8 renaming and its mapping file, plus what is and is not documented for Allatori and DexGuard.
ProGuard and R8
Both shrink, optimize and rename. Typical output: classes in the default package or packages like a.a.a, members named a, b, c, many methods with identical names but different signatures (overloading allowed once the descriptor differs). Line-number tables can be kept and then mapped back with the build's mapping file: see Format:ProGuard and R8 mapping file. R8 output lives in Format:DEX files inside an APK (Platform:Android). Names such as R8$$SyntheticClass and -$$Lambda$... are compiler-generated synthetics (Unverified: from memory, not read in a source).
DexGuard and Allatori
These are commercial. This wiki has not read their sources, and no license-compatible documentation was used. What an analyst may expect at the recognition level (all Unverified): string encryption with a static decrypt method called with a constant, class and method renaming beyond a-b-c style, reflection-based calls, and in DexGuard encrypted assets and classes decrypted at start. Check the apps' own strings and the loader code; a decryptor method called from almost every class is the common indicator. Add facts here only when an open source rule or document states them.
Practical checks
- Does the app ship a mapping file (release builds uploaded to crash services)? That removes renaming.
- Are names single letters with deep package nesting? ProGuard/R8 family.
- Are strings absent from the string_ids of the DEX and replaced by decrypt calls? String encryption.
See also Technique:Recognizing code obfuscation.