This repository is designed to serve as a sandbox for exploring a handful of test smells that are common in real-world test suites.
The examples in this repo go hand-in-hand with an interactive training workshop for developers, testers, & teams who want to improve their testing game. If you're interested in bringing this training to your team, please contact us at Test Double!
This repo should work with Node.js 4.x & up. We're using Yarn to ensure everyone has a deterministic dependency tree.
$ npm install -g yarnpkg
$ yarn
$ yarn test
# Run a single test
$ yarn test:one [Path to test]
# Example:
$ yarn test:one smells/insufficient/invisible-assertions/invisible-assertions.js
$ npm install
$ npm test
# Run a single test
$ npm run test:one [Path to test]
# Example:
$ npm run test:one smells/insufficient/invisible-assertions/invisible-assertions.js
Most tests should pass. (If a few tests fail, that's okay. Some of the tests are designed to fail erratically.)
This repo supports Ruby 1.9 & up. Install and the tests with:
$ bundle
$ bundle exec rake
Most tests should pass. (If a few tests fail, that's okay. Some of the tests are designed to fail erratically.)
This repo was created to facilitate workshops conducted by Test Double, a software agency dedicated to improving how the world makes software. It is shared here, because it may also be valuable as a workbook of sorts, for you to peruse at your own direction and pace. If you're interested in hearing more about what the workshop format for this project is like, please drop us a line!
Each of our odorous tests are organized under the smells/
directory and broken
down into these categories:
- Insufficient - tests that fail to fully specify the behavior of their subject under test
- Unclear - tests whose organization and design can mislead future maintainers, incidentally increasing their carrying cost
- Unnecessary - tests that do more than they need to or are otherwise inessential
- Unrealistic - tests that undermine their intended value by inappropriately replacing real things with fake things
- Unreliable - tests that behave erratically based on the time, machine, configuration, or environment
In each category's directory is a listing of several examples, each codified by a test.
If you're exploring this repository without having sniffed a specific smell, feel free to explore the examples by themes.
Each smell's file listing is structured the same way, starting with:
Each test resides in a directory which contains a README.md file with the following structure:
- The name of the test smell & a brief description
- The "odor" emanating from the example test subject and test
- An enumeration of the problems the smell might (or might not!) indicate; for
each potential problem:
- A description of the nature of the problem
- General tips on how to "deodorize" the smell
- Details about the example
- A root cause analysis of the smell
- A challenge to the reader to improve the design of the test and/or test subject
- Language-specific notes (common examples, likelihood of occurence, etc.)
- Ruby
- JavaScript
- Additional resources about the smell
Remember, not every smell you detect in the wild indicates an actual problem! Smells are simply surface indications of common problems and not necessarily problematic in-and-of themselves.
In each test will be one to several functions meant to be the "production" source code. Typically these would be broken out into a separate file, but to keep everything straightforward, each smell is kept to a single file listing.
For the purpose of keeping the examples easy-to-understand, the subject code is typically minimal and trivial, unless greater complexity is called for by the test smell itself.
Once you've read the description of the test smell, try to sniff it out among the file listing's tests. Usually, only one will exhibit the smell, but use your own judgment to determine which tests should be reworked.
Once you've detected a smell, attempt to identify which root cause provided in the description is causing the smell and attempt to implement its prescription for reworking the test. (In a few cases, a test's improvements will depend on refactoring the subject code, as well.)
Remember, the tests themselves are untested, so be sure that your new-and-improved test still works! Consider forcing the test to break, verifying that a message indicates the test is still doing its job before you commit your changes. As we like to say, "never trust a test you haven't seen fail."
There are several ways you can help us make this repo more useful to other developers trying to improve their tests. Here are some ideas of pull requests we'd really appreciate for this repo:
- Additional test smells. I'm sure we didn't catch them all!
- URLs pointing to examples of smells. The examples in this repo are necessarily minimal, and as a result there is a risk that they won't sufficiently remind people of their real-world tests. Additional examples can help people recognize the smells in their own code
- Feedback about your experience. If you attended a workshop that used this repo or if you just took a stab at working through it yourself, please open an issue to tell us about anything else you felt could be improved
If this repo helped you out and you just want to give us a high-five, please say hi on twitter or by e-mail.