This repository contains tools for detecting/classifying flaky tests.
More details about iDFlakies can be found in its paper and website.
Linux based systems only (in Windows system there are some problems).
iDFlakies may work on Maven-based or Gradle-based projects. Instructions are here for Maven-based projects and here for Gradle-based projects.
One may use iDFlakies on a specific Maven project by getting it from Maven Central or building it themselves.
Run the following command to automatically setup the pom.xml for iDFlakies.
bash pom-modify/modify-project.sh path_to_maven_project
By default, modify-project.sh will use the version of iDFlakies from Maven Central. If you wish to use the version of iDFlakies built locally, you can run the following instead.
bash pom-modify/modify-project.sh path_to_maven_project idflakies-maven-plugin 2.0.1-SNAPSHOT
Copy the following plugin into the Maven project's pom.xml. You do not need to perform this step if you have already completed the instructions in Automatically setting up the pom.xml for iDFlakies.
<build>
...
<plugins>
...
<plugin>
<groupId>edu.illinois.cs</groupId>
<artifactId>idflakies-maven-plugin</artifactId>
<version>2.0.1-SNAPSHOT</version>
</plugin>
</plugins>
</build>
Once iDFlakies is added to a Maven project, one can then run iDFlakies on the project with the following command.
mvn idflakies:detect -Ddetector.detector_type=random-class-method -Ddt.randomize.rounds=10 -Ddt.detector.original_order.all_must_pass=false
iDFlakies configuration options:
detector.detector_type
- Configurations of iDFlakies as described on pages 3 and 4 of our paper. Default israndom
(random-class-method).dt.randomize.rounds
- Number of times to run the test suite. Default is20
.dt.detector.original_order.all_must_pass
- Controls whether iDFlakies must use an original order of tests where all of them pass or not. Default istrue
(i.e., iDFlakies will exit if within three runs of the test suite, it does observe all tests to pass in one of the runs).dt.original.order
- Enables one to specify to iDFlakies the exact list of tests that should be run. Test names should be fully-qualified, use only.
to separate different parts of the test name (e.g.,com.github.kevinsawicki.http.EncodeTest.encode
), and test names are separated from each other by line breaks. This option is unlikely to be useful when running multiple modules at once and is best suited for running the tests of a specific module. Absolute paths should be used or the file path should be relative to the module that contains the tests.
You can also run iDFlakies on a Maven project without modifying the project's pom.xml file. You would specify the full group ID and artifact ID as part of the command. For example, if you want to do the same command as above, you would instead run.
mvn edu.illinois.cs:idflakies-maven-plugin:2.0.0:detect -Ddetector.detector_type=random-class-method -Ddt.randomize.rounds=10 -Ddt.detector.original_order.all_must_pass=false
Note how the command is still running the detect
goal, but it needs to specify the group ID to be edu.illinois.cs
and the artifact ID to be idflakies-maven-plugin
.
The version the command uses is 2.0.0
, meaning it can download this version from Maven Central, but you can change it to a locally-built version, e.g., 2.0.1-SNAPSHOT
.
The main script is located in scripts/docker/create_and_run_dockers.sh
.
To use the script, you will need Docker and Python (e.g., Python version 2.7.17) installed and a csv
file containing the GitHub URL and SHA that you wish to run iDFlakies on with one subject per line.
For example:
https://github.com/kevinsawicki/http-request,2d62a3e9da726942a93cf16b6e91c0187e6c0136
Once you have created the csv file, you can simply run:
bash create_and_run_dockers.sh <path to csv> <round num> <timeout (seconds)>
The script creates a new Docker image for each project/SHA, if one does not already exist.
Otherwise, the script will reuse the existing Docker image for the project.
When the script starts up the Docker container, it will checkout this repository (or update it to the latest version if it is already checked out).
To make changes to the scripts and tools and have them be used within the container, one may need to change scripts/docker/update.sh
.
cd scripts/docker
bash create_and_run_dockers.sh icst-dataset/comprehensive-individual-split/kevinsawicki.http-request.csv 10 1000000
This example takes about 22 minutes to run.
Running the above will output scripts/docker/all-output/kevinsawicki.http-request_output
as soon as iDFlakies is running.
An example of the log and output directory generated can be found in the here. Note that due to the inherent nondeterminism of the tool (e.g., use of random test orderings) and flaky tests, the log and output directory generated from another run may not exactly match the example log and output directory.
To see all of the flaky tests detected by iDFlakies one can refer simply to the scripts/docker/all-output/kevinsawicki.http-request_output/all_flaky_tests_list.csv
file.
The file lists all the flaky tests detected for each round of each configuration (e.g., com.github.kevinsawicki.http.HttpRequestTest.basicProxyAuthentication,./lib/detection-results/random/round0.json
).
Note that multiple rounds may detect the same test.
To see more information about a specific test (e.g., a passing and failing ordering of the test), please see the JSON file(s) that detected the test.
The output of iDFlakies is explained in depth in scripts/README.md
and the csv files used to evaluate iDFlakies are in scripts/docker/icst-dataset
.
iDFlakies now includes iFixFlakies. iFixFlakies automatically generates patches for order-dependent tests by finding and utilizing helper test methods in the test suite.
More details about iFixFlakies can be found in its paper and website.
First, include the iDFlakies plugin into your project or by specifying the full group ID and artifact ID as part of the running command (see instructions above).
iFixFlakies provides two new goals: minimize
and fix
.
Minimize aims to determine the minimal set of tests that can make the order-dependent test pass/fail. It also searches for cleaners in the case of polluter/victim. Assuming the iDFlakies plugin is included in the Maven project, use command:
mvn idflakies:minimize
You should only run this command after using idflakies:detect
to have detected some order-dependent tests.
This command generates, under .dtfixingtools/minimized/
, a JSON file per detected order-dependent test.
The JSON file contains information pertaining to the polluters, state-setters, and cleaners found for the order-dependent test.
Fix aims to generate a patch to be added to the order-dependent test so it can pass in different orders. Use command:
mvn idflakies:fix
You should only run this command after using idflakies:minimize
to have reported polluter/state-setters/cleaners for detected order-dependent tests.
This command generates, under .dtfixingtools/fixer/
, a JSON file per order-dependent test representing information pertaining to how it fixed the test.
The directory also contains for each order-dependent test some .patch
files that contain the possible patches.
If you use iDFlakies or iFixFlakies, please cite our corresponding ICST paper or ESEC/FSE paper:
@inproceedings{LamETAL2019ICST,
author = {Lam, Wing and Oei, Reed and Shi, August and Marinov, Darko and Xie, Tao},
title = {{iDFlakies}: {A} Framework for Detecting and Partially Classifying Flaky Tests},
booktitle = {12th {IEEE} International Conference on Software Testing, Verification and Validation},
pages = {312--322},
year = {2019},
}
@inproceedings{ShiETAL2019FSE,
author = {Shi, August and Lam, Wing and Oei, Reed and Xie, Tao and Marinov, Darko},
title = {{iFixFlakies}: {A} Framework for Automatically Fixing Order-Dependent Flaky Tests},
booktitle = {ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering},
pages = {545--555},
year = {2019},
}