How to Extend the Web Test Framework

The Web Test Framework that Blink uses is a regression testing tool that is multi-platform and it has a large amount of tools that help test varying types of regression, such as pixel diffs, text diffs, etc. The framework is mainly used by Blink, however it was made to be extensible so that other projects can use it test different parts of chrome (such as Print Preview). This is a guide to help people who want to actually the framework to test whatever they want.


Before you can start actually extending the framework, you should be familiar with how to use it. See the web tests documentation.

How to Extend the Framework

There are two parts to actually extending framework to test a piece of software. The first part is extending certain files in: /third_party/blink/tools/blinkpy/web_tests/ The code in blinkpy/web_tests is the web test framework itself

The second part is creating a driver (program) to actually communicate the web test framework. This part is significantly more tricky and dependent on what exactly exactly is being tested.

Part 1

This part isn’t too difficult. There are basically two classes that need to be extended (ideally, just inherited from). These classes are:

  • Driver. Located in web_tests/port/ Each instance of this is the class that will actually an instance of the program that produces the test data (program in Part 2).
  • Port. Located in web_tests/port/ This class is responsible creating drivers with the correct settings, giving access to certain OS functionality to access expected files, etc.

Extending Driver

As said, Driver launches the program from Part 2. Said program will communicate with the driver class to receive instructions and send back data. All of the work for driver gets done in Driver.run_test. Everything else is a helper or initialization function.

run_test() steps:

  1. On the very first call of this function, it will actually run the test program. On every subsequent call to this function, at the beginning it will verify that the process doesn’t need to be restarted, and if it does, it will create a new instance of the test program.
  2. It will then create a command to send the program
    • This command generally consists of an html file path for the test program to navigate to.
    • After creating it, the command is sent
  3. After the command has been sent, it will then wait for data from the program.
    • It will actually wait for 2 blocks of data.
      • The first part being text or audio data. This part is required (the program will always send something, even an empty string)
      • The second block is optional and is image data and an image hash (md5) this block of data is used for pixel tests
  4. After it has received all the data, it will proceed to check if the program has timed out or crashed, and if so fail this instance of the test (it can be retried later if need be).

Luckily, run_test() most likely doesn’t need to be overridden unless extra blocks of data need to be sent to/read from the test program. However, you do need to know how it works because it will influence what functions you need to override. Here are the ones you’re probably going to need to override


This function creates a set of command line arguments to run the test program, so the function will almost certainly need to be overridden.

It creates the command line to run the program. Driver uses subprocess.popen to create the process, which takes the name of the test program and any options it might need.

The first item in the list of arguments should be the path to test program using this function:


This is an absolute path to the test program. This is the bare minimum you need to get the driver to launch the test program, however if you have options you need to append, just append them to the list.


If your program has any special startup needs, then this will be the place to put it.

That’s mostly it. The Driver class has almost all the functionality you could want, so there isn’t much to override here. If extra data needs to be read or sent, extra data members should be added to ContentBlock.

Extending Port

This class is responsible for providing functionality such as where to look for tests, where to store test results, what driver to run, what timeout to use, what kind of files can be run, etc. It provides a lot of functionality, however it isn’t really sufficient because it doesn’t account of platform specific problems, therefore port itself shouldn’t be extend. Instead LinuxPort, WinPort, and MacPort (and maybe the android port class) should be extended as they provide platform specific overrides/extensions that implement most of the important functionality. While there are many functions in Port, overriding one function will affect most of the other ones to get the desired behavior. For example, if web_tests_dir() is overridden, not only will the code look for tests in that directory, but it will find the correct TestExpectations file, the platform specific expected files, etc.

Here are some of the functions that most likely need to be overridden.

  • driver_class
    • This should be overridden to allow the testing program to actually run. By default the code will run content_shell, which might or might not be what you want.
    • It should be overridden to return the driver extension class created earlier. This function doesn’t return an instance on the driver, just the class itself.
  • driver_name
    • This should return the name of the program test p. By default it returns ‘content_shell’, but you want to have it return the program you want to run, such as chrome or browser_tests.
  • web_tests_dir
    • This tells the port where to look for all the and everything associated with them such as resources files.
    • By default it returns the absolute path to the web tests directory.
    • If you are planning on running something in the chromium src/ directory, there are helper functions to allow you to return a path relative to the base of the chromium src directory.

The rest of the functions can definitely be overridden for your projects specific needs, however these are the bare minimum needed to get it running. There are also functions you can override to make certain actions that aren’t on by default always take place. For example, the web test framework always checks for system dependencies unless you pass in a switch. If you want them disabled for your project, just override check_sys_deps to always return OK. This way you don’t need to pass in so many switches.

As said earlier, you should override LinuxPort, MacPort, and/or WinPort. You should create a class that implements the platform independent overrides (such as driver_class) and then create a separate class for each platform specific port of your program that inherits from the class with the independent overrides and the platform port you want. For example, you might want to have a different timeout for your project, but on Windows the timeout needs to be vastly different than the others. In this case you can just create a default override that every class uses except your Windows port. In that port you can just override the function again to provide the specific timeout you need. This way you don’t need to maintain the same function on each platform if they all do the same thing.

For Driver and Port that’s basically it unless you need to make many odd modifications. Lots of functionality is already there so you shouldn’t really need to do much.

Part 2

This is the part where you create the program that your driver class launches. This part is very application dependent, so it will not be a guide on how implement certain features, just what should be implemented and the order in which events should occur and some guidelines about what to do/not do. For a good example of how to implement your test program, look at MockDRT in mock_drt.pyin the same directory as and It goes through all the steps described below and is very clear and concise. It is written in python, but your driver can be anything that can be run by subprocess.popen and has stdout, stdin, stderr.


Your goal for this part of the project is to create a program (or extend a program) to interface with the web test framework. The web test framework will communicate with this program to tell it what to do and it will accept data from this program to perform the regression testing or create new base line files.


This is how your code should be laid out.

  1. Initialization
    • The creation of any directories or the launching of any programs should be done here and should be done once.
    • After the program is initialized, “#READY\n” should be sent to progress the run_test() in the driver.
  2. Infinite Loop (!)
    • After initialization, your program needs to actually wait for input, then process that input to carry out the test. In the context of web testing, the content_shell needs to wait for an html file to navigate to, render it, then convert that rendering to a PNG. It does this constantly, until a signal/message is sent to indicate that no more tests should be processed
    • Details:
      • The first thing you need is your test file path and any other additional information about the test that is required (this is sent during the write() step in run_tests() is This information will be passed through stdin and is just one large string, with each part of the command being split with apostrophes (ex: “/path’foo” is path to the test file, then foo is some setting that your program might need).
      • After that, your program should act on this input, how it does this is dependent on your program, however in content_shell, this would be the part where it navigates to the test file, then renders it. After the program acts on the input, it needs to send some text to the driver code to indicate that it has acted on the input. This text will indicate something that you want to test. For example, if you want to make sure you program always prints “foo” you should send it to the driver. If the program every prints “bar” (or anything else), that would indicate a failure and the test will fail.
      • Then you need to send any image data in the same manner as you did for step 2.
      • Cleanup everything related to processing the input from step i, then go back to step 1.
        • This is where the ‘infinite’ loop part comes in, your program should constantly accept input from the driver until the driver indicates that there are no more tests to run. The driver does this by closing stdin, which will cause std::cin to go into a bad state. However, you can also modify the driver to send a special string such as ‘QUIT’ to exit the while loop.

That’s basically what the skeleton of your program should be.


This is information about how to do some specific things, such as sending data to the web test framework.

  • Content Blocks
    • The web test framework accepts output from your program in blocks of data through stdout. Therefore, printing to stdout is really sending data to the web test framework.
    • Structure of block
      • “Header: Data\n”
        • Header indicates what type of data will be sent through. A list of valid headers is listed in
        • Data is the data that you actually want to send. For pixel tests, you want to send the actual PNG data here.
        • The newline is needed to indicate the end of a header.
      • End of a content block
        • To indicate the end of a a content block and cause the driver to progress, you need to write “#EOF\n” to stdout (mandatory) and to stderr for certain types of content, such as image data.
      • Multiple headers per block
        • Some blocks require different sets of data. For PNGs, not only is the PNG needed, but so is a hash of the bitmap used to create the PNG.
        • In this case this is how your output should look.
          • “Content-type: image/png\n”
          • “ActualHash: hashData\n”
          • “Content-Length: lengthOfPng\n”
          • “pngdata”
            • This part doesn’t need a header specifying that you are sending png data, just send it
          • “#EOF\n” on both stdout and stderr
        • To see the structure of the data required, look at the read_block functions in