Digging into: Humbug

Jump to: TL;DR

While I’ve already used Humbug a few time, a recent article made my realize that I didn’t really know how it worked.

That’s when I got the idea to dig into Humbug to learn how it works, and publish my findings here.

Before we start, let’s quickly give some context on Humbug.

What’s Humbug?

« Humbug is a Mutation Testing framework for PHP »

In a nutshell, you write unit tests to prevent regressions in your application and use code coverage as an indicator of how well your application is tested or protected.

Mutation testing comes after the redaction of these tests and helps you to judge how well your unit tests actually protect your code.

To do that, mutation testing tools inject small defects into your source code and then check if the unit tests noticed the changes. If they did, then the coverage is sufficient and we say that they killed the mutation. If they didn’t, then you have a problem because the mutation escaped.

According to Humbug’s README, mutations usually involve:

  • switching binary arithmetic operators: + becomes -, * becomes /, …
  • substituting logical operations: && becomes ||, true becomes false, …
  • inversing conditions: > becomes <, === becomes !===, …

The purpose of this article is to answer the following question: how does Humbug manipulate your code? How are mutations generated?

The analysis

What follows is a redacted transcription of my investigation in Humbug’s source code.

First step: composer.json

I’ll use the composer.json file as entry point to this analysis. The goal here is to see if there is a dependency that could be use to analyze and manipulate source code. I’m thinking about a library like nikic/PHP-Parser for instance.

 2 3    "require": {
 4        "php": ">=5.4.0",
 5        "phpunit/phpunit": "^4.5|^5.0",
 6        "symfony/console": "^2.6|^3.0",
 7        "symfony/finder": "^2.6|^3.0",
 8        "symfony/process": "^2.6|^3.0",
 9        "symfony/event-dispatcher": "^2.6|^3.0",
10        "sebastian/diff": "^1.1",
11        "padraic/phpunit-accelerator": "^1.0.2",
12        "padraic/phpunit-extensions": "^1.0.0",
13        "padraic/phar-updater": "^1.0.0"
14    },

After reading the dependencies list, I see nothing that would help to manipulate source code.

I do see interesting things though:

  • phpunit/phpunit is a hard dependency: meaning that Humbug won’t work with other test frameworks
  • symfony/finder and symfony/process: common components used to find files and launch processes
  • sebastian/diff: could be used to generate diffs between original source code and mutants

If there is nothing useful in the dependencies, it means that the source manipulation is in Humbug’s repository.

Time to read some code!

When I open the sources, the Mutator directory immediately catches my eye. So let’s see what’s inside.

 2├── Arithmetic
 3├── Boolean
 4├── ConditionalBoundary
 5├── ConditionalNegation
 6├── IfStatement
 7├── Increment
 8├── MutatorAbstract.php
 9├── Number
10└── ReturnValue

First thought: it reminds me of the mutations listed in the README. Let’s open a file to check it. I’ll begin with src/Mutator/Arithmetic/Addition.php.

The Addition class contains two methods: getMutation(array &$tokens, $index) and mutates(array &$tokens, $index). Both of them receive an array of tokens which indicates that Humbug operates at the token level, not at the AST level.

N.B: a list of tokens is usually the result of the lexical analysis, which is the process of converting a source code into a sequence of strings — or tokens — with a precise signification (identifier, literal, assignment operator, …).

The method getMutation() seems to be the one applying the mutation:

 2 * Replace plus sign (+) with minus sign (-)
 3 *
 4 * @param array $tokens
 5 * @param int $index
 6 * @return array
 7 */
 8public static function getMutation(array &$tokens, $index)
10    $tokens[$index] = '-';

The method mutates(), on the other hand, checks whether the current mutation can be applied on the given token. This method’s docblock gives a really meaningful example where it can’t: $var = ['foo' => true] + ['bar' => true]

N.B: I think that the documentation for these two methods is well redacted and useful.

By reading the implementation of the mutates() method, I noticed the usage of the T_ARRAY constant. This constant being provided by PHP itself, its usage leads me to believe that the tokenization of the source code isn’t done by hand by Humbug, but delegated to a function provided by PHP’s standard library: token_get_all().

That being said, the tokenization process itself isn’t done in the Addition class. Let’s inspect its parent: MutatorAbstract

In addition to the “utility” methods, this class has an interesting method that performs the mutation on the given tokens and returns the result as source code:

 2 * Perform a mutation against the given original source code tokens for
 3 * a mutable element
 4 *
 5 * @param array $tokens
 6 * @param int $index
 7 * @return string
 8 */
 9public static function mutate(array &$tokens, $index)
11    static::getMutation($tokens, $index);
12    return Tokenizer::reconstructFromTokens($tokens);

What’s interesting here is the usage of a Tokenizer class.

At this stage, I already know that Humbug manipulates tokens and rewrites them to generate mutants, but I still didn’t see the code responsible for the tokenization process itself or even the reconstruction of the code source from the modified tokens.

Logically, I expect to find all of this in the Tokenizer.

And bingo! The class contains the two expected methods: reconstructFromTokens() and getTokens().

The tokenization is indeed done using token_get_all(). According to the documentation, this function returns an array of token identifiers. Each individual token identifier is either a single character (i.e.: ;, ., >, !, etc…), or a three element array containing the token index in element 0, the string content of the original token in element 1 and the line number in element 2.

All of these information allow us to analyze the tokens and reconstruct the original source code. That’s exactly what Humbugs does: mystery solved!

Where are the mutations generated?

So far, we’ve seen how a single mutation manipulates the tokens generated by the Tokenizer. The last thing that I’d like to determine is, given a file, how are the mutations generated?

When going back to the src/ directory, I notice the Mutable class which could represent a mutable file/source code.

And indeed, when opening it I see that the constructor accepts a file name! I also notice that the $mutators attribute contains a list of all the possible mutations.

Farther in the file is the generate() method. It’s were the file is opened, tokenized (using the Tokenizer) and modified using the mutators.

One interesting thing: the mutators are applied only to method bodies!


In the end, the mutation generation by Humbug is pretty straightforward:

  1. it starts by splitting the code in tokens, using the standard token_get_all() function
  2. then it applies a list of mutators to the previously generated tokens. A mutator being responsible to alter tokens (ie: replacing a + with a -)
  3. once all the mutators are applied, it uses the tokens to rebuild the source code.
  4. code having changed, the unit tests should fail. So it launches them and expects failures.

Now I see why I didn’t find something like nikic/PHP-parser in the dependencies: it simply would have been overkill and too complex to work with an AST.