universalmutator

This is a tool based on source-based rewrite of code lines for mutation generation, including multi-language rules aided by special rules for languages or even projects. Originally, the approach used only regular expressions, treating code as text. However, there is also a mode that can use the Comby tool for more sophisticated mutation that produces fewer invalid mutants. Regular-expression based mutation works well, in our experience; comby-aided mutation works even better. The key advantage of either approach is that the tool can probably mutate approximately any interesting source code you have, and language changes don't force rewriting of the mutation tool. To use the comby mode, just make sure comby is installed and add --comby when you run mutate.

More information on this project can be found in a 2024 FSE paper, and in the original 2018 ICSE Tool Paper.

A guest blog post for Trail of Bits shows how to use the universalmutator to help improve a C/C++ API fuzzing effort using DeepState and libFuzzer.

The universalmutator has support for extracting coverage information to guide mutation from the TSTL testing tool for Python.

HOW TO USE IT

To use this, you should really just do:

pip install universalmutator

then

mutate --help

SIMPLE EXAMPLE USAGE

mutate foo.py

or

mutate foo.swift

should, if you have the appropriate compilers installed, generate a bunch of valid, non-trivially redundant, mutants.

A MORE COMPLEX EXAMPLE

Sometimes the mutated code needs to be built with a more complicated command than a simple compiler call, and of course you want help discovering which mutants are killed and not killed. For example, to mutate and test mutants for the mandelbrot plotting example included in the PROGRAMMING RUST book (http://shop.oreilly.com/product/0636920040385.do), just do this:

git clone https://github.com/ProgrammingRust/mandelbrot
cd mandelbrot
cargo build
target/debug/mandelbrot origmandel.png 1000x750 -1.20,0.35 -1,0.20
mkdir mutants
mutate src/main.rs --mutantDir mutants --noCheck
analyze_mutants src/main.rs "cargo clean; cargo build; rm mandel.png; target/debug/mandelbrot mandel.png 1000x750 -1.20,0.35 -1,0.20; diff mandel.png origmandel.png" --mutantDir mutants

(It will go faster if you edit main.rs to lower the maximum number of threads used to something like 8, not 90.) At the moment, this won't use any Trivial Compiler Equivalence, but still kills about 60% of the 1000+ mutants. The killed mutant filenames will be in killed.txt and the non-killed ones in not-killed.txt.

Working with something like maven is very similar, except you can probably edit the complicated build/clean stuff to just a 'mvn test' or similar.

CURRENTLY SUPPORTED LANGUAGES

The tool will likely mutate other things, if you tell it they are "c" or something, but there is auto-detection based on file ending and specific rule support for:

C
C++
Java
JavaScript
Python
Swift
R
Rust
Go
Lisp
Fortran
Solidity
Vyper
Fe

(the last three are smart contract languages for the Ethereum blockchain).

All but C, C++, JavaScript, and Go will try, by default, to compile the mutated file and use TCE to detect redundancy. Of course, build dependencies may frustrate this process, in which case --noCheck will turn off TCE and just dump all the mutants in the directory, for pruning using a real build process. In the long run, we plan to integrate with standard build systems to avoid this problem, and with automated test generation systems such as TSTL (https://github.com/agroce/tstl) for Python or Echidna for Solidity (https://github.com/trailofbits/echidna). Even now, however, with analyze_mutants it is fairly easy to set up automatic evaluation of your automated test generator.

MUTATING SOLIDITY CODE

The universalmutator has been most frequently applied to smart contracts written in the Solidity language. It supports a few special features that are particularly useful in this context.

First, Solidity libraries are often written with only internal functions --- and the compiler will not emit code for such functions if you compile a library by itself, resulting in no non-redundant mutants. In order to handle this case, mutate can take a --compile option that specifies another file (a contract using the library, or the tests in question) that is used to check whether mutants are redundant.

Second, swapping adjacent lines of code is a seldom-used mutation operator that is unusually attractive in a Solidity context because swapping a state-changing operation and a requirement may reveal that testing is incapable of detecting some re-entrancy vulnerabilities. The testing may notice the absence of the check, but not a mis-ordering, and these mutants may reveal that. To add code swaps to your mutations, just add --swap to the mutate call. Note that swaps work in any language; they are just particularly appealing for smart contracts.

MORE INFORMATON

For much more information, again see https://agroce.github.io/icse18t.pdf -- demo/tool paper at ICSE 18 and especially our full FSE 2024 paper -- https://agroce.github.io/fse24.pdf -- the latter discusses the latest version of the tool/approach, and includes a comparison with many other mutation testing tools.

The aim of this project is partly to see how quickly mutation can be applied to new languages, partly how much the work of a tool can be offloaded to the compiler / test analysis tools.

Authors: Alex Groce, Josie Holmes, Darko Marinov, August Shi, Lingming Zhang, Kush Jain, Rijnard van Tonder, Sourav Deb