CSC/ECE 517 Fall 2017/M1754 Mutation Testing on Servo
Servo uses the Web Platform Test (WPT) suite for testing, but does not perform an evaluation of the breadth of the tests. The goal of this project is to use techniques from mutation testing to evaluate the performance of the WPT suite when bugs are deliberately introduced into the code base.
The implementation of this project was done by writing python scripts that would modify the source code to generate mutants and run tests on them to check for failures. The scripts would temporarily modify the source codes, call the corresponding tests and revert back to the original code by reversing the changes that were made earlier. This process was repeated for multiple iterations by modifying various parts of the source code.
Introduction
Servo
Servo is a modern, high-performance browser engine designed for both application and embedded use. Servo is a web browser layout engine written in Rustand is currently being developed by Mozilla. The aim of the project is not to create a full browser but is rather to create a highly parallel environment that allows for many components be handled by fine-grained, isolated tasks. [1]
Rust
Rust is an open source, systems programming language sponsored by Mozilla Research. Rust performs the majority of its safety checks and memory management decisions at compile time, so that program’s runtime performance is not impacted. Making it useful in programs with predictable space and time requirements, embedding in other languages, and writing low-level code, like device drivers and operating systems.
Web-platform-tests
The web-platform-tests Project is a W3C-coordinated attempt to build a cross-browser test suite for the Web-platform stack. Writing tests in a way that allows them to be run in all browsers gives browser projects confidence that they are shipping software that is compatible with other implementations, and that later implementations will be compatible with their implementations. This in turn, gives Web authors/developers confidence that they can actually rely on the Web platform to deliver on the promise of working across browsers and devices without needing extra layers of abstraction to paper over the gaps left by specification editors and implementors.
Mutation Testing
Mutation Testing is a type of software testing where we mutate (change) certain statements in the source code and check if the test cases are able to find the errors.The goal of Mutation Testing is to assess the quality of the test cases which should be robust enough to fail mutant code. This method is also called as a Fault-based testing strategy as it involves creating faults in the program.Faults are introduced into the source code of the program by creating many versions called mutants. Each mutant should contain a single fault, and the goal is to cause the mutant version to fail which demonstrates the effectiveness of the test cases.[2]
Project description
Build process
- The steps to setup the environment were followed as mentioned in readme file link. The servo code was built in release mode on a Mac environment using the below command:
./mach build --release
The whole set of WPT was run on the servo code base using the below command:
./mach test -wpt --release
Implemented steps
The approach adopted was based on the requirements mentioned here. The below steps implement the initial steps mentioned in the project description.
- Step 1: A python script was written to mutate one source file and the corresponding WPT was run on it to check if the mutant was killed.
- Step 2: A test framework was defined to identify the source files that require mutation testing along with their corresponding WPTs. This framework is implemented in /servo/components/script/dom.
- Step 3: The script was expanded to include the test framework and automate the process of generating mutations for multiple source files and running their corresponding WPTs based on the test_mapping.json. The script also traverses through sub folders of the parsed path to check for the .json file. The script also logs on the terminal any mutants that the WPT failed to kill.
- Step 4: Integrated the script so that it can be invoked from the CI tool.
Execution
- The steps to setup the environment were followed as mentioned in readme file link.
Example on how to setup the environment and run the mutation testing scripts.
Step A-Setting up the environment for Mac OS: Install python version 3 and run the following for installation of virtualenv:
sudo port install python27 py27-virtualenv cmake yasm
Step B-Build Process
- Clone the repository from link
- From the Servo directory invoke the following script :
python python/servo/mutation/init.py components/script/dom
Mutation test can be ran using the following command:
python python/servo/mutation/init.py components/script/dom
Output
The log for success to kill a mutant is as shown below:
The log for failure to kill a mutant is as shown below:
Testing
The project is about writing a python script to change the source code and run tests. It does not add any functionality to Servo. So there is no scope for testing in this project.
Subsequent steps
The following subsequent steps will be followed to meet the project requirements as per this [3].
- implement mutations like replacing if statements by if true/if false, duplicating statements, reordering statements, changing arithmetic & atomic string constant.
- improving the performance of the testing, for example randomizing the test order, fast-failing, running tests with faster builds (e.g. ./mach build -d).
- find heuristics for identifying false positives, that is mutations which are expected to have no effect, for example removing logging.
- find search heuristics for identifying mutations that cause no test failures.
Challenges
1. Setting up environment in local machine:
- The amount of time taken to build and test the WPT in local machine was pretty long due to machine memory. Additionally faced problems on intermittent WPT test failures. It is expected that any reviewer will face a similar challenge.
We overcame this challenge by discussing possibilities with Mozilla team who suggested us to use web application such as Janitor Technology. Using a web based platform for our testing made the project agnostic of the environment on the local machine and any hassle associated with same.
2. Defining the test mapping framework:
- Servo consists of multiple components of which there are many wpt tests written for each component. The test cases are organized according to the functionality they test, whereas the servo source code is organized according to the specific components. Due to a large number of source files and WPT tests for the entire project, it was difficult for one individual to have information on the mapping of components to their corresponding WPTs.
Hence, we had to have a mapping to run the corresponding tests for a source file. The mapping framework enabled us to run tests on a few components and provide a base for other members from Mozilla team to add and replicate the same framework as per their requirement.
3. Understanding the extent of mutation testing to be performed.
There are standard techniques that define what a mutant is. For example, replacing '&&' with '||'. However, there is no specification for the number of replacements that must be made within a single source code, nor is there any specification on the permutations of possible mutations for each technique. This is an ambiguity in the project and our implementation replaces all instances of '&&' to '||' to generate a single type of mutant for the each file of the source code.
Pull Request
Here is our pull request link.
References
<references/>
1. https://en.wikipedia.org/wiki/Servo_(layout_engine)
2. https://www.guru99.com/mutation-testing.html
3. https://github.com/servo/servo
4. https://github.com/servo/servo/wiki/Mutation-testing-project