REserve — Tests & Quality

After drafting the first working version of REserve, behavior validation and quality assessment were some of the remaining challenges. This article will detail the different strategies adopted for testing as well as the tooling used to ensure the quality of the project.

Basic testing

There are many ways to deal with web application testing. On one hand, you may use some selenium-based tools (or any alternative) to automate a browser, run a given set of scenarios and assess the expected results. On the other hand, you may leverage a simpler http client (for instance curl or Node.js’ helpers such as request, got or superagent) and analyze the responses.

With regards to this project, the most important thing is to validate the way the server behaves according to its configuration. Since it is very flexible, every possible combination must be tested and the error management finely controlled.

When the project started, an initial — not to say basic — test was created by building a small website illustrating most of the features. First, a mix of different handlers and mappings was consolidated inside a mappings.json file. Then, two extension files exposed this definition through http and https.

Initial mappings used to prototype REserve

http.json file used to expose mappings through http

Finally, a web page was designed to leverage the different mappings and the result was validated with a list of assertions (thanks to the gpf-js http helper).

Basic tests executed in a browser

However, this can hardly be automated (or the lazy me did not want to go with selenium).

This is the reason why the Node.js command line all.js was introduced.

It runs the different configuration files (http & https) by creating a child process with child_process.fork. It relies on REserve’s parent process notification feature to wait for the server startup. Once synchronized, it executes all the assertions and then stops the server (by killing the process).

In-depth testing

The previous approach considers the whole project as a standalone — monolithic — component and tests it by leveraging a client. This is also known as end to end testing. If you understand the pyramid of test, this category of test is situated almost on the top of it… Meaning that this should be the place where the least effort is being put.

A simplified version of the pyramid of tests

Another angle for testing is to isolate each class and service by mocking or stubbing their dependencies and test them individually. This is known as unit testing. It begins the pyramid of tests, meaning this is where most effort should be put.

mocha

If you are familiar with the mocha framework, you know that it is simple to implement, widely used and it supports asynchronous features such as promises.

Mocha framework logo

For each source file of the project a corresponding test file is created, for instance: handlers/custom.js is tested by tests/mocha/handlers/custom.test.js. The directory structure is also recreated under the mocha tests folder.

This pattern makes the configuration of mocha easier since you just need to setup the spec file pattern to *.test.js.

When running, it produces a report that helps you identify which test case failed.

Mocha execution excerpt

Isolation in a nutshell

As explained before, the project didn’t start with the tests first. As a result, no encapsulation was done and the code is heavily relying on Node.js native APIs. In particular :

It is possible to substitute any Node.js modules (including the native ones) using mock-require. This simple API allows you to predefine a module with a mocked version. Consequently, whenever the API is used, the mocked object is returned.

Mocking the file system

Mocking the file system is not really mandatory: using a dedicated directory structure in the project could have been enough.

However, the project aims to run on any platform. And, actually, the development environment (Windows) is different from the continuous integration platform one (Linux).

The file system differences between operating systems have a significant impact on REserve. Indeed, a web server running on a UNIX-like operating system would be case sensitive with the URLs. On windows, it might not.

Since REserve uses only a subset of the fs APIs, a custom mocked version was build to redefine only the APIs that are really used.

The whole file system is virtualized with a dictionary where members are object representing either files (when they contain a content property) or folders (when no content is found).

A specific API controls whether the file system is case sensitive or not.

The main function of the virtual file system

Mocking of http, requests & responses

According to REserve, there are not much differences between the http and the https modules. Actually, only two methods are used :

Since these modules are widely used, there is no value in testing them again. The same way, we want to simplify the tests and avoid going through the http(s) layer.

As a result :

And to enable individual handlers testing, REserve provides two classes to simulate the request and response objects. They both implement streams.

Implementations can be found in:

Those classes are part of the NPM module together with the mock helper: it provides a simple test framework to develop new custom handlers.

Quality tools

How do you make sure that the code is maintainable ? How many tests are necessary to ensure that the code works as expected ?

Different tools are available to answer those questions.

A linter can analyze the codebase statically and report on common errors. It usually provides settings to define what must be verified and how the findings must be treated in terms of severity. Sometimes, the linter may even automatically fix some of the findings.

For instance :

if (parameter = '--config') {
return false
}

An example of faulty code

> standard --fix
standard: Use JavaScript Standard Style (https://standardjs.com)
.\reserve\index.js:13:9: Expected a conditional expression and instead saw an assignment.
.\reserve\index.js:13:9: Unexpected constant condition.

The linter output

A linter can also check the formatting being used through the sources, such as the choice between tabs and spaces, how strings should be delimited… This guarantees consistency when reading the code and improves maintainability.

REserve project uses standardJS, a linter based on ESLint with a predefined configuration.

Code coverage with Istanbul

One way to check if the tests are relevant is to quantify how many lines of the codebase are executed during the tests. This operation is called the code coverage measurement.

The 100% paradox

Before getting into tools, I would like to illustrate one common misunderstanding about the code coverage.

Let’s consider the following function :

function divide (a, b) {
return a / b;
}

An example of function to be tested

The test below executes the function and, as a result, it is enough to cover it completely.

assert.strictEqual(divide(4,2), 2)

An example of test

Since we reached 100% of coverage, does it mean that the function is fully tested ?

Actually, no. It really depends on the specification of the function.

For instance, the following questions are not answered by the test :

So, to put it in a nutshell, 100% of coverage does not guarantee that everything is tested but, on the contrary, if it does not reach 100% of coverage it means that some parts are not tested.

Tools

There are many tools to measure the code coverage and most of them require instrumenting the code base first. Luckily this step is usually transparent.

After the tests execution, the tool produces a report that contains :

And depending on the granularity :

This report is important to easily identify which part of the code needs more tests.

These metrics are consolidated and expressed in percentage of coverage for each category. Thresholds can be defined to fail the process when the percentage does not reach the expectations.

REserve uses which, itself, relies / is an evolution of istanbul. The whole process is easy since nyc supports the execution of an external command and it takes care of everything transparently. Consequently, the code coverage is triggered with nyc mocha.

--------------------------|---------|---------|----------|---------|
File | % Stmts | % Branch | % Funcs | % Lines |
--------------------------|---------|---------|----------|---------|
All files | 100 | 100 | 100 | 100 |
reserve | 100 | 100 | 100 | 100 |
body.js | 100 | 100 | 100 | 100 |
checkMethod.js | 100 | 100 | 100 | 100 |
configuration.js | 100 | 100 | 100 | 100 |
dispatcher.js | 100 | 100 | 100 | 100 |
iconfiguration.js | 100 | 100 | 100 | 100 |
index.js | 100 | 100 | 100 | 100 |
interpolate.js | 100 | 100 | 100 | 100 |
mapping.js | 100 | 100 | 100 | 100 |
mock.js | 100 | 100 | 100 | 100 |
schema.js | 100 | 100 | 100 | 100 |
serve.js | 100 | 100 | 100 | 100 |

Excerpt of the coverage report

Continuous integration

Tests, static checks and code coverage are good ways to assess the quality of the code but they work only if executed. A continuous integration automated pipeline ensures that whenever the code is pushed to the code repository, these tools are executed.

REserve leverages the Travis CI platform and all these tools are triggered during the build.

Also, the code coverage results are uploaded to the Coveralls platform.

Code smells

Last but not least, the project is also registered on Code Climate.

This online platform performs code analysis and detects maintainability issues (usual code smells such as duplicated code, oversized function, code complexity…).

The result is quantified with a badge as well as an estimated cleaning time.

Conclusion

To conclude, the project is fully tested and its quality is constantly controlled and measured. But, still, it can be improved. For instance, in terms of project management, there are no stories or acceptance criteria documented. Also, the tests quality can be assessed using mutation testing.

Passionate developer, always keen on finding new ways to do things, welcome to my thoughts.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store