Fernflower

A decompiler from Java bytecode to Java.

About Fernflower

Fernflower is the first actually working analytical decompiler for Java and probably for a high-level programming language in general. Naturally, it is still under development. Please send your bug reports and improvement suggestions to the issue tracker (in subsystem

).

Naming

The correct name is Fernflower, not FernFlower.

Credits

Fernflower was originally written by Stiver.

Fernflower includes some patches from ForgeFlower. Sincere appreciation is extended to the maintainers of ForgeFlower for their valuable contributions and enhancements.

A mirror of this repository has been maintained for many years by Andrew McRae until JetBrains offered to take over. Sincere appreciation is extended to Andrew for his work maintaining the mirror.

License

Fernflower is licensed under the Apache License Version 2.0.

Usage

IntelliJ

The Fernflower IDE plugin is bundled in IntelliJ IDEA. Open any

file and you should see the decompiled Java source code: this is Fernflower in action.

The plugin is also open-source and can be found here.

Running from the command line

means zero or more times

means one or more times

: file or directory with files to be decompiled. Directories are recursively scanned. Allowed file extensions are class, zip and jar. Sources prefixed with -e= mean "library" files that won't be decompiled but taken into account when analyzing relationships between classes or methods. Especially renaming of identifiers (see the

ren

option) can benefit from information about external classes.

: destination directory to place the resulting Java source into

: a command-line option with the corresponding value (see "Command-line options" below).

Examples

Command-line options

Except for

and

urc

the value of 1 means the option is activated, 0 - deactivated. The default value, if any, is given between parentheses.

Typically, the following options will be changed by user, if any: hes, hdc, dgs, mpm, ren, urc The rest of options can be left as they are: they are aimed at professional reverse engineers.

• rbr
  (1): hide bridge methods
• rsy
  (0): hide synthetic class members
• din
  (1): decompile inner classes
• dc4
  (1): collapse 1.4 class references
• das
  (1): decompile assertions
• hes
  (1): hide empty super invocation
• hdc
  (1): hide empty default constructor
• dgs
  (0): decompile generic signatures
• ner
  (1): assume return not throwing exceptions
• den
  (1): decompile enumerations
• rgn
  (1): remove
  getClass()
  invocation, when it is part of a qualified new statement
• lit
  (0): output numeric literals "as-is"
• asc
  (0): encode non-ASCII characters in string and character literals as Unicode escapes
• bto
  (1): interpret int 1 as boolean true (workaround to a compiler bug)
• nns
  (0): allow for not set synthetic attribute (workaround to a compiler bug)
• uto
  (1): consider nameless types as
  java.lang.Object
  (workaround to a compiler architecture flaw)
• udv
  (1): reconstruct variable names from debug information, if present
• ump
  (1): reconstruct parameter names from corresponding attributes, if present
• rer
  (1): remove empty exception ranges
• fdi
  (1): de-inline finally structures
• mpm
  (0): maximum allowed processing time per decompiled method, in seconds. 0 means no upper limit
• ren
  (0): rename ambiguous (resp. obfuscated) classes and class elements
• urc
  (-): full name of a user-supplied class implementing
  IIdentifierRenamer
  interface. It is used to determine which class identifiers should be renamed and provides new identifier names (see "Renaming identifiers")
• inn
  (1): check for IntelliJ IDEA-specific @NotNull annotation and remove inserted code if found
• lac
  (0): decompile lambda expressions to anonymous classes
• nls
  (0): define a new line character to be used for output. 0 -
  '\r\n'
  (Windows), 1 -
  '\n'
  (Unix), default is OS-dependent
• ind
  : indentation string (default is 3 spaces)
• crp
  (0): use record patterns where it is possible
• cps
  (0): use switch with patterns where it is possible
• log
  (INFO): a logging level, possible values are TRACE, INFO, WARN, ERROR
• iec
  (0): include the entire classpath in context when decompiling
• isl
  (1): inline simple lambda expressions
• ucrc
  (1): hide unnecessary record constructor and getters
• cci
  (1): check if resource in try-with-resources actually implements
  AutoCloseable
  interface
• jvn
  (0): overwrite any local variable names with JAD style names
• jpr
  (0): include parameter names in JAD naming

Renaming identifiers

Some obfuscators give classes and their member elements short, meaningless and above all ambiguous names. Recompiling of such code leads to a great number of conflicts. Therefore, it is advisable to let the decompiler rename elements in its turn, ensuring uniqueness of each identifier.

Option

(i.e.

-ren=1

) activates renaming functionality. The default renaming strategy goes as follows:

• rename an element if its name is a reserved word or is shorter than 3 characters
• new names are built according to a simple pattern: (class|method|field)_<consecutive unique number>
  You can overwrite these rules by providing your own implementation of the 4 key methods invoked by the decompiler while renaming. Simply pass a class that implements
  org.jetbrains.java.decompiler.main.extern.IMemberIdentifierRenamer
  in the option
  urc
  (e.g. -urc=com.example.MyRenamer) to Fernflower. The class must be available on the application classpath.

The meaning of each method should be clear from naming: toBeRenamed determine whether the element will be renamed, while the other three provide new names for classes, methods and fields respectively.

Development

Build an executable start-up script:

The startup script is generated in

.