Skip to content

Latest commit

 

History

History
263 lines (173 loc) · 15.2 KB

README.md

File metadata and controls

263 lines (173 loc) · 15.2 KB
Raw

HaxeContracts - Unit's Bane

A Design by contract library for Haxe.

Heavily inspired by Microsoft Code Contracts, with a few code convention changes like camelCase and better use of Haxe's type inference.

And of course everything contract-related in software started with Eiffel by Bertrand Meyer. Thank you!

Download and Install

Install via haxelib: haxelib install HaxeContracts (Note the UpperCased characters)

Then add -lib HaxeContracts in your hxml.

Usage & Tutorial

You use contracts while designing your classes, so let's design a Rational class as a tutorial how to use the library.

A rational number is quite simple: It is expressed by a quotient: numerator/denominator. The denominator cannot be zero. Lets model a class based on that, and we'll add the two imports required for HaxeContracts:

package;
import haxecontracts.*;

class Rational implements HaxeContracts {
    public var numerator(default, default) : Int;
    public var denominator(default, default) : Int;

    public function new(numerator : Int, denominator : Int) {
        this.numerator = numerator;
        this.denominator = denominator;
    }
    
    public function toFloat() {
        return numerator / denominator;
    }
}

There we have the basics, a class implementing haxecontracts.HaxeContracts. Now let's consider the important rule: The denominator cannot be zero. Using Contracts, we can enforce this.

When adding a Contract condition, which is simply a Boolean statement, the first step is usually to consider a few things:

  • Is this an invariant condition? I.e, must it always be true for the entire lifetime of the object?
  • Must this condition be true before we enter a method (precondition), or when we return (postcondition)?

Preconditions

Starting with the constructor, the denominator parameter cannot be zero, which is a precondition. You add that with the Contract.requires method:

public function new(numerator : Int, denominator : Int) {
    Contract.requires(denominator != 0);
    
    this.numerator = numerator;
    this.denominator = denominator;
}

Postconditions

That was quite simple, and not so much difference from an if-statement that throws an exception. Postconditions however gives more insight to the power of Contracts. As you may have noticed the class properties are public, so someone can set the denominator to zero after instantiation. Setting aside how "correct" that is, our job is again to enforce the rule. There is no way of adding Contract conditions to default properties however, so we must use accessors, a set_denominator method:

class Rational implements HaxeContracts {
    public var numerator(default, default) : Int;
    public var denominator(default, set) : Int;

    public function new(numerator : Int, denominator : Int) {
        Contract.requires(denominator != 0);
        
        this.numerator = numerator;
        this.denominator = denominator;
    }
    
    private function set_denominator(v : Int) {
        Contract.ensures(Contract.result != 0);
        
        return denominator = v;
    }    
}

This is more interesting. We're using Contract.ensures to define the postcondition, and also Contract.result to test the actual return value. The condition goes on top of the method, as all conditions are, but in reality Contract.ensures will be called right before the method returns, checking this time that nobody is breaking the denominator rule.

Invariants

You may have realized it already, but the denominator rule is actually an invariant. No matter what happens to our Rational objects, the denominator cannot be zero. So we can plan for the future and save code at the same time by making an invariant method:

@invariants function invariants() {
    Contract.invariant(denominator != 0, "Denominator cannot be zero.");
}

All invariant conditions will be tested as postconditions to every public method, and public properties with get/set accessor methods.

Two things to remember for the invariant method:

  1. Mark it with @invariants (the method name can be anything)
  2. Call Contract.invariant in the same way as pre/postconditions. (Contract.result cannot be used in invariants, naturally.)

As a bonus we added a text message after the condition. All Contract methods have this feature, so you can describe the rules straight away.

(Note: For technical reasons the toString method is excluded from invariants.)

Old

Contract.old is used to test if an expression conforms to some condition before and after method execution. It can only be used within Contract.ensures. A common usage is to test if some counter has been increased, for example:

public function increaseCounter(i : Int) : Int {
    Contract.ensures(Contract.old(i) == Contract.result - 1);
    i++;
    return i;
}

The expression within Contract.old(...) denotes the value the expression had on method entry.

The finished class

A final touch: It's possible to skip the static Contract class name, keeping only the method calls.

This essentially reserves the words requires, ensures, invariant, old and result in a class implementing HaxeContracts. It's quite convenient when you've memorized the API, but if you don't like "magic methods", or this creates a problem with existing method names or variables, you can disable it. See the "Compilation flags" section further below for instructions how to do that.

Here's our completed Rational class:

import haxecontracts.*;

class Rational implements HaxeContracts {
    public var numerator(default, default) : Int;   
    public var denominator(default, set) : Int;

    public function new(numerator : Int, denominator : Int) {
        requires(denominator != 0);
        
        this.numerator = numerator;
        this.denominator = denominator;
    }
    
    public function toFloat() : Float {
        return numerator / denominator;
    }
    
    private function set_denominator(d : Int) {
        ensures(result != 0);
        
        return denominator = d;
    }

    @invariants function invariants() {
        invariant(denominator != 0, "Denominator cannot be zero");
    }
}

(The call to ensures is redundant because of the invariant in this simple example, but keeping it to show the syntax.)

Contract violations

When a condition fails, a haxecontracts.ContractException object is created and thrown. It has many useful properties:

  • message - The error message from the Contract call.
  • object - A reference to the object where the condition failed.
  • arguments - An array containing the arguments of the failing method of the object.
  • callStack - A stack trace.
  • innerException - If an execption occured in the condition itself, this is the exception caught.
  • pos - Where the failure occured in the code.

Since it's an exception it can be caught, but be aware: Don't catch the ContractException for anything but logging purposes! Jon Skeet explains it very well, but in short, contract violations puts the system in an invalid state, which can propagate to other parts of the system unless it shuts down quickly. Catch it high up in the stack, log it somewhere, then rethrow or exit as gracefully as possible.

API reference

Contract.requires(condition : Bool, ?message : String)
Specifies a requirement (precondition). Executed at the beginning of the method.

Contract.ensures(condition : Bool, ?message : String)
Ensures a final condition (postcondition). Executed right before the method returns.

Contract.invariant(condition : Bool, ?message : String)
A condition that must hold throughout the object's lifetime. Executed right before every public method returns, including public properties with accessor methods.

Contract.result
Refers to the return value of the method. Can only be used within Contract.ensures.

Contract.old(arg : Dynamic)
Refers to the original value of the current method argument arg. Can only be used within Contract.ensures.

Contract.assert(condition : Bool, ?message : String, ?objectRef : Dynamic)
A general assertion that can be placed anywhere in the code.

Imports and implements

Any class calling haxecontracts.Contract must implement haxecontracts.HaxeContracts, except when using only Contract.assert.

All API methods except Contract.assert can be used without the static Contract class. If this creates a problem with existing method names or variables, see below for how to disable it.

If you want to use assert in the same way, just import it normally: import haxecontracts.Contract.assert;

Compilation flags

Flag (-D) Effect
contracts-disabled Disables the whole Contract code generation
contracts-preconditions-only Disables Contract code generation, except for preconditions (on method entry)
contracts-no-imports If the Contract method names conflicts with existing fields or variables, this flag disables it, and you must use the static Contract class explicitly.

Please note that disabling contracts as above doesn't affect the Contract.assert method. It's a general assertion, not a contract bound to an object or method.

Why "Unit's Bane?"

Because the downsides of TDD and unit testing are getting more and more obvious. There are two very well-written articles by James Coplien that deserves better spreading:

Design by Contract combined with a system architecture like DCI could be a much better way forward than the test-driven reign. The massive, almost paranoid testing focus we see today is mostly a consequence of fundamental limitations in the software architectural model.

In testing terms, we have the unit level, which quickly becomes a "throw as much input as possible into this method", as stated in the article we're commenting on. Could be fun for a discrete math-loving nerd, but let's not be navel-gazing. We code mainly for others. Also, since the tests frequently only concerns single methods we're not far from stepping back from OO to plain old procedural thinking (Pascal, Fortran).

Machines can handle this level much better. If we know the boundaries of the public class interface, in another word the Contract of the class, we can let a program figure out the input variations and test it automatically. Contracts gives us a way. Pex is an unfortunately dead Microsoft project that made some promising work in this area.

A much more interesting testing level is system behavior, since the interesting stuff (for users and stakeholders) happens between objects. Unfortunately in the current software "object" model, behavior is spread out through classes, making it very hard to grasp the polymorphic, abstract mess that "OO code" usually evolves into. System architecture today is actually class-oriented rather than object-oriented, since we only see the class structure; there is no easy way to reason about object behavior and collaboration. And that's usually where the bugs are... (Yet another design pattern won't simplify either, sorry.)

This elephant in the room has forced programmers to create bloated testing harnesses, often with a deteriorating codebase the same size as the application itself. Unit testing is a cumbersome, semi-static contract checking that slowly drags the project down.

In other words, the time has come for computer engineers to realize the underlying problem, instead of getting excited over the next slick testing tool. The rest of the world demands it, and unless you program alone in your spare time, the rest of the world probably pays you for doing a good job, in good time.

Are there any options?

Now that Contracts have helped us create reliable objects and data, we still have to test that things work, right? Quoting the user-focused legend Jef Raskin:

"As far as the customer is concerned, the interface is the product."

Testing the product/interface has always been done in good old QA through exploratory testing, and nowadays we can reach some level of automation through browser testing frameworks like Playwright and cypress.io. But there are still a few things to be aware of:

  • Tests should be written by someone else than the programmer
  • Make sure the tests aren't TDD in disguise.

The first point is needed because tests usually becomes a self-fulfilling prophecy. If the programmer writes the tests, he/she will make them pass. Secondly, it's important to ignore the code and focus on behavior. Domain knowledge is more important than code knowledge here, which is the main reason a domain expert or stakeholder should write the tests. But it's hard to map domain knowledge to a browser test, so this can be difficult to achieve. This means that if you are alone on the project or other constraints puts you in the role of "tester", you have to step out of the programmer's shoes for a while.

The second point is more obvious. Don't use for example a browser testing framework to write unit tests, bashing the interface instead of functions. It's also more subtle; it's a good thing that we're taking tests to a higher level. Specifications can be detailed and low-level though, so the framework may not be expressive/convenient enough to cover the whole complexity of the system. This is where DCI (Data, Context and Interaction) makes its entry, as a real solution to the above described problems that we see in many system architectures. We finally get to reason about system behavior in a specific Context. No polymorphism or layers of abstractions, just object collaboration as seen at runtime. (And those collaborating objects are now protected by their Contract specifications!)

Hopefully I made a good enough case for you to consider Contracts as a viable alternative to TDD and most unit testing, and a small but useful introduction to DCI. Writing and manipulating an ever-growing series of tests forced me to look for alternatives, maybe it's the same for you? I've been researching DCI for 10 years and will answer any questions you'll have about it.

In software development, reliability should be built-in, not an afterthought. -- Eiffel Software

More Design by Contract information

Build Status