Читать книгу Reliable JavaScript - Lawrence Spencer - Страница 20
PART I
Laying a Solid Foundation
CHAPTER 1
Practicing Skillful Software Engineering
WRITING CODE THAT STARTS CORRECT
ОглавлениеWhat Johann Sebastian Bach said about playing a keyboard instrument applies equally to programming a computer: There’s nothing remarkable about it. All one has to do is hit the right keys at the right time and the instrument plays itself.
This section is about hitting the right keys at the right time. As you might guess, there’s more to it than the uninitiated might think.
But first, a story.
Mastering the Features of JavaScript
Have you ever seen someone get his head chopped off on a squash court? One of us nearly did. It was during an introductory college course in the sport, but the episode had a lot to teach about writing reliable JavaScript.
In case you’re not familiar with the game, it’s played in a court that is like a large room. Two players alternate hitting a ball with their rackets toward the back wall, which they both face. In the most basic scenario, you hit the ball at the wall; it bounces off and then bounces off the floor toward your opponent, who is standing next to you. Then he smashes it toward the back wall for you to try to hit.
Anyway, it was the first day of the course. The instructor was standing to the student’s left and a little behind him, and the rest of us were watching through the glass front wall. The instructor directed the student to hit the ball toward the back wall.
The student, who was a tennis player, hit a forehand as he would in tennis, stroking from low to high, with a high follow-through that wrapped around his body. That is how you hit with topspin in tennis. It’s also how you chop off the head of whoever happens to be standing to your left and a little behind you.
Fortunately, the instructor knew this would happen and had positioned his racket in front of his face to defend himself.
The student’s racket crashed against the instructor’s, making a lot of noise and causing the student some embarrassment, but no harm was done.
The instructor pointed out that in tennis, you generally hit with topspin so the ball dives down and bounces up with a kick toward your opponent. However, that same stroke in squash does the opposite. If you hit with topspin, the squash ball will kick up off the wall, making an easy, looping arc, and then bounce in a lazy manner off the floor, whence your opponent will crush it. In squash, you want to hit with backspin. The ball will then kick down off the wall, and kick off the floor toward your opponent with increased velocity.
The normal stroke in squash, then, is a chopping, downward motion to impart backspin – just the opposite of the typical stroke in tennis.
Even though the two sports have basic features in common (two players, rackets, and a ball) as well as common demands (good hand-eye coordination, good anticipation and movement on your feet), you won’t play squash well if you try to hit the ball as you would in tennis.
In the same way, JavaScript makes its particular demands on the programmer. If you come to large-scale JavaScript development with primary experience in another language, you will do well to attune yourself to the differences in technique.
The differences are at both the small scale of syntax and the large scale of architecture and engineering.
Throughout this book, you will encounter JavaScript’s unique syntactic delights. Many of them are summarized in Chapter 25. This chapter looks at the larger issues of how JavaScript’s peculiarities make certain engineering techniques possible.
By employing these techniques, you will write JavaScript with kick. Your game will improve. You will “win” more often because you will be working with the language instead of contrary to it.
Case Study: D3.js
Mike Bostock’s JavaScript masterpiece, D3.js
, is a perfect example.
D3 stands for Data-Driven Documents, so called because it lets you create beautiful SVG graphics from data. For example, Figure 1.1 is a D3 diagram that shows class dependencies in a software system (from http://bl.ocks.org/mbostock/4341134).
Figure 1.2 presents the same data in a radial layout (http://bl.ocks.org/mbostock/1044242). D3 is very flexible. It is also very concise; each diagram takes just a few dozen lines of pleasingly formatted JavaScript to create.
D3’s home page is http://d3js.org, with source code available at https://github.com/mbostock/d3. This is real JavaScript, not for the faint of heart and orders of magnitude more artful than the field-validators and button-handlers that are sprinkled through a typical website.
In fact, it’s so artful as to be overwhelming at first read, so we have simplified just one corner of it for discussion. Listing 1-1 is an abridged version of d3.svg.line
, a function that creates an SVG line generator. An explanation follows the listing.
LISTING 1-1: A function to create an SVG line (code filename: rj3\rj3.js)
You would use this function to turn an array of data into an SVG path. SVG paths are just strings in the small language of SVG. Suppose you wanted to draw a line like the one in Figure 1.3.
The SVG <path>
element would be
In English, that says to pick up the pen and move it (“M”) to the (x, y) coordinate (10, 130), and then draw a line (“L”) to (100, 60), and then draw another line to (190, 160), and then finish with a line to (280, 10).
So how does the code in Listing 1-1 create a path like that? Consider Listing 1-2, which contains a sample call.
LISTING 1-2: Sample call to rj3.svg.line() (code filename: rj3\pathFromArrays.js)
On the highlighted line, what ends up in lineGenerator? Well, according to the last line of Listing 1-1, a call to rj3.svg.line()
will return something called line
. What is that? It is a function nested inside the outer function rj3.svg.line
!
NOTE In JavaScript, functions can nest inside other functions. This becomes an important way to control scope.
By the way, we have retained D3’s names for most properties and variables so you can study the full listing at https://github.com/mbostock/d3/blob/master/src/svg/line.js if you wish, and be as well-oriented to it as possible. In only a few cases have we attempted to clarify things by changing a variable’s name. If you find it confusing that both the outer and inner functions are named line
, well, this is very much in the spirit of all the D3 source code so you might as well learn to enjoy it.
Yes, the function returns a function. This is a confusing no-no in most languages, but in JavaScript it’s a very idiomatic yes-yes that broadens your architectural options. If you’re going to code industrial-strength JavaScript, get used to functions being first-class objects that are passed as arguments, sent back as return values and just about anything else you can imagine. As first-class citizens of JavaScript, they can even have properties and methods of their own.
NOTE In JavaScript, functions are objects that can have methods and properties. Your functions can have more flexibility and power than they might in other languages.
You can see an example of attaching a method to a function in this part of Listing 1-1:
It creates a function, x
, that is a member of the returned function, line
. Shortly, you will see how x
and its twin, y
, are used, and learn the very JavaScript-ey peculiarities of what’s inside them.
So the call rj3.svg.line()
returns a function. Continuing with Listing 1-2, the function is called with arrayData, which becomes the data argument to that inner line
function from Listing 1-1. From there, the while
loop fills the points array from the incoming data:
Each element of data, held in the variable d, is passed to the getX
and getY
functions, which extract the x and y coordinates. (The use of call
to invoke getX
and getY
will be covered at the end of this Case Study, as well as in Chapter 18. The + in front of getX
and getY
is a little trick to ensure that actual numbers, not numeric strings, go in the points array.) By default, those coordinates are the first and second elements of the 2-element array that comprises each element of arrayData. This occurs in the following snippet of Listing 1-1.
Next, the segment
function is called. This is a function at yet another level of nesting, private to the line
function. It fills the segments variable, putting the SVG "M"
command in the first element and the path in the second. From Listing 1-1 again:
The path is produced by the interpolate
function, which in the default implementation just joins the points (each implicitly converted to a string), putting an "L"
between them. (We’ll cover interpolate
in more detail later in this chapter.)
Thus, the array
becomes
As a final step, the two elements of segments ("M"
and the points-as-string) are joined in the return
statement to produce the SVG path
That’s the basic operation. Now for some complications that will illustrate additional ways that you can use JavaScript idiomatically.
Suppose that each point in your data were an object instead of an [x,y]
coordinate pair in array form. It might look something like this:
How could you use rj3.svg.line
to draw it? One way would be to transform the data on the way in, as in Listing 1-3.
LISTING 1-3: Transforming the data on the way in (code filename: rj3\pathFromTransformedObjects.js)
However, that would be wasteful, as it creates a second, complete copy of the data. It’s the sort of thing a C# programmer accustomed to the efficiencies of LINQ would do. (LINQ peels off just one element at a time from an array as requested, without making a second copy of the whole array.)
The strategy in Listing 1-3 would also limit your possibilities in the user interface. You probably want your line to change dynamically if the data change. Thanks to the design decision that you’re going to see in a moment, D3 does this for you with no effort – but only with the data it knows about. If you have called its functions with only a one-time copy of the real data, you don’t get this benefit.
The design decision is exemplified by the little functions, line.x
and line.y
. Listing 1-4 shows how to use them.
LISTING 1-4: Using line.x and line.y (code filename: rj3\pathFromObjects.js)
The call
replaces the default value of Listing 1-1’s getX variable with your new function. Now, when the while
loop calls
the getX.call
will invoke your function, which returns the x
property of your objects – the original, authoritative objects, and not copies of them.
There’s something else worth noting about those call
s. Without stealing all the thunder from Chapter 18, we’ll state that whatever function is installed to get the x
coordinate is actually called with two arguments, even though your function(d){return d.x;}
only took one. The second argument, i, is the index of the datum, d, in the array. You didn’t use i, but you could have. This is how the object-oriented concept of function overloading works in JavaScript.
Another example of JavaScript’s function overloading is in the line.x
function itself. Did you notice the if
test of arguments?
In JavaScript, arguments is an array-like object that is available inside every function, containing the arguments the function was called with. Here, the test inspects the length of that pseudo-array. Zero is a “falsy” value in JavaScript (see “Values May Be Truthy or Falsy” in Chapter 25) so if there are no arguments, the function just returns the current value of getX.
To recap, if line.x
is called with no arguments, it returns the current accessor for x-coordinates. If it is called with an argument, it sets the x-coordinate accessor to it and returns something else entirely, namely the line
function-object. This, and the possibility of the extra argument, i
, exemplify function overloading in JavaScript.
NOTE In JavaScript, the object-oriented concept of function overloading is done by inspecting the function’s arguments and adjusting accordingly.
Now why would a function that sets the x-accessor return the line
? You probably know the answer: It allows you to chain the calls as you saw in Listing 1-4:
The design possibilities of call-chaining are explored at length in Chapter 15.
Now here’s a question for you. What do you suppose would happen if you were to add a z-coordinate to each data point?
If you guessed that the program would happily produce exactly the same result, you are right. In JavaScript, an object with x
, y
, and z
properties can also function as an object with x
and y
properties.
You could also produce the objects with a constructor function, which looks completely different but has the same result:
This is called duck typing, after the saying, “If it looks like a duck, walks like a duck and quacks like a duck, it is a duck.” In JavaScript, ducks are some of your best friends. It is possible to distinguish the cases thus:
However, there is almost never a reason to do so. A C# or Java programmer might attempt to learn whether an object is up to snuff through such inspections, but the JavaScript way is to simply check for the existence of the properties:
or
Duck typing is not sloppiness. It is an important way to give a component more reach.
NOTE Embrace duck typing. It allows a little code to accommodate a wide range of objects.
If you read Listing 1-1 with unusual attention, you might have wondered how the inner line
function manages to access the private variables of the outer rj3.svg.line
after the outer function has returned. Programmers from other languages might expect the variables getX, getY, and interpolate to pop off the stack once control exits the function that declared them. And so they would, except for one thing: JavaScript’s concept of closures.
We said earlier that when you call rj3.svg.line()
, it returns the inner line
function. There's more to it than that. It actually returns a closure, which you can think of as an object that from the outside looks like the function (inner line
), but on the inside also remembers the environment that prevailed when the function was created (the variables getX, getY and interpolate). You call inner line
’s functions as you normally would, but they are aware of line
’s original environment.
NOTE Closures are a very powerful design element in JavaScript. Every function is a closure.
Consider once more the call
statements in the while
loop:
What does getX.call(this,d,i)
really do? In English, it calls the getX
function, pretending that it is a member of the object this (more on that in a moment) and passing the arguments d and i. The special variable this is, loosely speaking, the “object before the dot” when you call the function in which this appears.
Why all this fuss and bother? Why not just say getX(d,i)
and be done with it? In JavaScript, the ability to specify this is an important design opportunity.
NOTE In JavaScript, “
this
” offers a design opportunity. Use it!
Listing 1-5 shows the power of this language feature. Here, the data are just an array of years. The function line.x
computes the desired x coordinate based on the index, i (now we’re using i!), but what’s going on with line.y
? It appears to be calling a function, getValue
, that is nowhere in scope.
LISTING 1-5: Extending the line generator to get values from an outer object (code filename rj3\pathFromFunction.js)
So where does getValue
come from? In the second part of the listing, a yearlyPriceGrapher object is instantiated that combines a line generator with a function, getValue
, that returns the value for a given year. In the call
the yearlyPriceGrapher is “dotted with” lineGenerator. That means that yearlyPriceGrapher becomes this in the y-accessor, which causes its getValue
to be invoked properly. The result is in Figure 1.4.
It is natural to think that this refers to the function in which it appears, or maybe the object enclosing the function. Not so. It refers to the object on which the function is called.
JavaScript Is Single-Threaded
Just one more thing to close out this section about language features: JavaScript is single-threaded. That doesn’t mean it uses a blocking model – far from it. It just means that you do asynchronous programming differently.
Where a multi-threaded language would allow you to start a task that runs in parallel to the code that spawned it, in JavaScript you merely enqueue a function to execute as soon after a certain event as possible. The triggering event may be the passage of a certain amount of time (in the case of setTimeout
), the arrival of data from a website (in the case of XMLHttpRequest.send
), or the click of a mouse, among many possibilities. JavaScript has an event loop that consumes the functions thus enqueued one at a time.
From a design point of view, this makes your life easier than it would be in a true multi-threaded environment. You never have to worry about getting interrupted, or about other objects accessing your variables when you think you have control.
It also means you shouldn’t hog the processor!
In Chapter 6, you will see how JavaScript Promise
s let you write code that does not block, yet is not a confusing scatter of event-handlers awkwardly connected by variables.
Avoiding JavaScript’s Pitfalls in Larger Systems
Why is a system that contains 50 classes (or objects, in JavaScript) more than ten times as challenging to write and maintain as a system that contains five? With five objects, even if each one draws on the services of all the others, there are at most 20 channels of communication (each of 5 objects calling 4 others – allowing ourselves to count A calling B as well as B calling A). With 50, there are 2450 (50 times 49) – more than 100 times as many.
With the advent of Single-Page Applications, node.js
, and other ways of making JavaScript shoulder the burdens of larger and larger systems on both client and server, the best JavaScript developers get serious about trimming those channels of communication to a bare minimum.
Where an object must interface with others to do its job, the connections are managed assiduously to ensure that they function properly in all circumstances.
This section will suggest ways to meet these goals.
Scripts Are Not Modules
Just last week, we were on the website of a company that makes a certain specialized device for user input. They had helpfully provided sample JavaScript code for using their device.
Argh! Their JavaScript library, suggested for all programmers to use, was over 1900 lines of one global variable or function after another – over 200 global functions in all. Most of the global functions were at least named so that collisions with other libraries were unlikely, but some, such as makeUri
or toHex
, were not.
This is the “scripting” heritage of JavaScript at work. In the old days, when your script was probably the only one on the page, there was little harm in adding to the global namespace. With today’s JavaScript applications, that is never the case.
Your script is in no way isolated because it is in its own .js
file. If your file starts with
as this one did (the names have been changed to protect the guilty), then myVariable
is visible to all the other scripts in your application, and the makeValue
function evidently is, too.
JavaScript presents an unusually diverse menu of choices for creating modules that properly encapsulate their data and functions. Script files are not one of them! (You will read more about data encapsulation in Chapter 3.)
Nested Functions Control Scope
In C# or Java, one class can contain another. However, this practice is not widespread. Microsoft even cautions against it. Code Analysis warning 1034 is “Nested types should not be visible” and their rationale is “Nested types include the notion of member accessibility, which some programmers do not understand clearly” (https://msdn.microsoft.com/en-us/library/ms182162.aspx).
JavaScript does not have classes, but nested functions serve the same purpose of organizing the code hierarchically. Crucially, the hierarchy not only helps the programmer find what he’s looking for; it also helps the program minimize the scope of its variables and functions. That’s key to keeping a large system under control, and it is the warp and woof of the best JavaScript code.
Recall this snippet from Listing 1-1:
The inner line
function has a member function, line.x
. Although x
is a member of line
, it cannot see line
’s local variables, such as segments. Both line
and line.x
can see the getX
variable in the enclosing function. Combine this sort of artfulness with closures, and you have some very powerful tools for keeping large JavaScript systems under control.
Coding by Contract
There is no better way to make a large system more manageable than to make it smaller. JavaScript, with the extraordinary flexibility of pervasive duck-typing, lets you write a little code that can do a lot. (Recall the variety of inputs handled in the D3 case study earlier in the chapter.)
The flip side is that you never know what someone is going to throw at your software.
If your function expects its arguments to meet certain requirements, consider validating them. In Chapters 16 through 21, you will see one way to do this as unobtrusively as possible: the ContractRegistry
.
In a nutshell, the registry allows you to verify anything you wish about an argument or return value, without adding any code to your function. It does this through the magic of aspect-oriented programming (covered in Chapter 2) and in such a way that the overhead of validation can be eliminated in the shipped version.
Applying the Principles of Software Engineering
Have you ever been to a concert by a virtuoso musician? Maybe you play the same instrument, and you’ve marveled that the performer makes it look so easy. The truth is, he makes it look easy because it is easy – for him. And the reason it’s easy for him is that he has trained his fingers to move efficiently, trained his body to relax and breathe, trained his mind to listen to the music rather than be distracted by anxiety.
He probably learned the piece by playing it very, very slowly at first. Only when he had mastered it completely at that pace did he take the metronome up one notch. Thus, he did not practice-in any mistakes. One of us, a classical guitarist, went to a masterclass taught by one of the world’s best. The teacher boasted, “I bet I can play this piece slower than any of you.” He has learned that the quickest way to learn to play a piece flawlessly is to play it slowly.
When you have mastered the principles in this section, you will write flawless software more quickly and with less effort. Your fellow developers will look at your code and say, “He makes it look so easy!”
The SOLID Principles
The acronym SOLID was coined by Michael Feathers as a way to remember the five principles of object-oriented design that Robert Martin set forth in the late 1990s (summarized at http://www .objectmentor.com/resources/articles/Principles_and_Patterns.pdf). They are:
● The Single Responsibility Principle
● The Open/Closed Principle
● The Liskov Substitution Principle
● The Interface Segregation Principle
● The Dependency Inversion Principle
The Single Responsibility Principle
Stated in its most extreme form, the Single Responsibility Principle is that a class (or function, in JavaScript) should have only one reason to change.
That is a very tall order. Surely every line of code represents something that could change. Must every function consist of just one line of code?
No, but don’t give up on this principle too quickly. Consider once more the rj3.svg.line
function in Listing 1-1. As you saw, it is able to generate SVG line paths from just about any data source you can imagine, but what about it might change? It takes almost every cue from outside, even down to how it obtains the x and y
coordinates for each data point.
By the way, one of the casualties of our abridgement of D3’s code was the interpolate
function. In the full version, D3 lets you specify this just as you can specify the functions that obtain x
and y
. And what does interpolate
do? It connects the points in an SVG path. The default is to connect the points with straight line segments, but you could plug in an interpolator that constructs graceful curves instead, and D3 supplies several such interpolators.
Thus, when it comes right down to it, rj3.svg.line
really doesn’t “know” much. All it does is return a function (the inner line
) that can create an SVG path out of an array of data points – somehow.
What reasons could there be for rj3.svg.line
to change? Its one responsibility is to produce an SVG path from an array. Everything about how it carries out that responsibility is external to the function and therefore not a reason for it to change!
All together now: “Mike Bostock, you make it look so easy!”
The Open/Closed Principle
This principle states that “Software entities should be open for extension, but closed for modification” (http://www.objectmentor.com/resources/articles/ocp.pdf).
In other words, you should never change working code. Instead, reuse it somehow (for example, by inheritance) and extend it.
This, too, is a tall order. Robert Martin even admits in the same article, “In general, no matter how ‘closed’ a module is, there will always be some kind of change against which it is not closed. Since closure cannot be complete, it must be strategic. That is, the designer must choose the kinds of changes against which to close his design. This takes a certain amount of prescience derived from experience.”
When Mike Bostock designed his d3.svg.line
function, he anticipated changes in the way coordinates might be plucked from data and how the points might be joined (interpolated), and he wisely abstracted those features out of his function.
What he did not think would change (at least not in a backward-incompatible way) was the SVG path specification. He dared to hard-code that a path could always start with "M"
and continue with the points in order, as a text string.
Short of a breaking change to the SVG spec, it is hard to imagine how d3.svg.line
would ever have to change.
The Liskov Substitution Principle
“The what??” you ask!
Coined by Barbara Liskov in a formal way in Data Abstraction and Hierarchy (SIGPLAN Notices 23, 5 [May, 1988]), this principle might be stated more colloquially for a JavaScript context as follows:
Code written to use an object of a certain type should not have to change if provided with an object of a derived type.
Another way of saying this is that when you derive one object from another, the base-level semantics should not change.
If you find yourself writing branching logic so that your function does one thing if provided with a base class, but something else for a derived class, you have violated this principle.
This does not apply to types that do not derive from each other. For example, it is a common and good practice in JavaScript for a function to branch one way if an argument is a Number, another way if it’s a String, and a third way if it’s not there at all and therefore of the Undefined type. As discussed previously, that’s how JavaScript fulfills the object-oriented idea of function overloading.
Incidentally, the use of duck-typing, while not the same as derivation, is very much in the spirit of this principle!
The Interface Segregation Principle
This principle arose in a milieu of interface-based languages such as C++ and Java. In those languages, an interface is a piece of code that describes the functions in a class (names, parameters, and return types) without implementing those functions.
The idea is that an interface with many functions should be broken up into smaller, cohesive parts. Consumers should rely on only one of the mini-interfaces, not on the “fat” whole.
Of course, this is in the service of minimizing the width of the connections between modules. As stated previously, trimming the channels of communication is critical to making large JavaScript systems manageable.
But wait a minute! In JavaScript, there are neither classes nor interfaces. Does that mean JavaScript programmers cannot experience the benefits of following this principle?
Not at all. In fact, we will devote all of Chapter 16 to how to implement this principle in JavaScript. In the meantime, here’s a preview: To follow the spirit of the Interface Segregation Principle, a function can make clear what it expects of its arguments, and those expectations should be minimized. As stated earlier, duck typing is your friend here. Rather than expecting an argument of a certain type, just expect it to have the few properties of that type that you actually need. The ContractRegistry
that will be developed in Chapters 16 through 21 provides a formal way to make the expectations clear and to enforce them. If the ContractRegistry
is not to your taste, you can always write argument-validation code or even write comments!
The Dependency Inversion Principle
This principle, too, was developed with interfaces in mind. Robert Martin states it thus: “High-level modules should not depend upon low-level modules. Both should depend upon abstractions” (http://www.objectmentor.com/resources/articles/dip.pdf).
In an interface-based language, this principle usually finds its expression in the related idea of dependency injection. If class A needs the services of B, it does not construct B. Instead, one parameter to A’s constructor is an interface that describes B. A no longer depends on B, but on its interface. When A is constructed, a concrete B is passed in. B, too, depends on its interface.
The benefit is that a derived version of B, which also fulfills the interface, can be supplied instead thanks to the Liskov Substitution Principle. Furthermore, if B does need to change (in spite of the Open/Closed Principle), the interface concisely describes how it must continue to behave in order to be backward-compatible.
Once again, in JavaScript there are no abstractions, but JavaScript programmers can still program in the spirit of this principle and enjoy its benefits.
The full version of D3’s d3.svg.line
function starts like this (from https://github.com/mbostock/d3/blob/master/src/svg/line.js):
The projection parameter of d3_svg_line
is used to possibly project the data points to another coordinate space. By default, projection is d3_identity
, which makes no changes to the points at all. However, other projections are possible. For example, d3.svg.line.radial
uses polar coordinates (an angle and distance from the origin) by injecting the d3_svg_lineRadial
projection (https://github.com/mbostock/d3/blob/master/src/svg/line-radial.js):
Конец ознакомительного фрагмента. Купить книгу