harmony 鸿蒙ArkTS Migration Background

2023-10-30
浏览 (344)

ArkTS Migration Background

This chapter explains why it makes sense to migrate from the standard TypeScript to ArkTS. In general, there are two reasons for doing this:

Program stability. Dynamically typed languages like JavaScript are very good at allowing programs to write code fast. At the same time, these languages are notorious for unexpected runtime errors. For example, a developer may forget to check some value for undefined, and as a result of this, the program may crash, which causes inconvenience to the users. Detecting such issues during development time would be much more beneficial. TypeScript helps greatly here: It allows to annotate the code with types, and many errors will be detected by the compiler, prior to deployment and usage of the code. However, even TypeScript has limitations and sometimes permits to annotate the code with types “loosely”, which still leaves a gap for runtime errors. ArkTS tries to overcome this drawback: It enforces static typing for even stricter type checking and less runtime errors.
Program performance. To ensure correctness of the program, dynamically languages have to check actual types of objects when the program actually runs. Back to our example, JavaScript does not allow to read a property from undefined. But the only way to check if some value is undefined is to perform a runtime check, that all JavaScript engines do: if the value is not undefined, the property is read, otherwise an exception is thrown. Modern engines can optimize such checks greatly, but these checks cannot be eliminated completely, which leads to code slowdown. Since the standard TypeScript compiles to JavaScript, the code written in TypeScript has exactly the same issues as described above. ArkTS addresses this problem. Since static typing is enforced, ArkTS compiles the program not to JavaScript, but to some special execution format called bytecode, which is faster to execute and easier to optimize even further.

Below there are a couple of examples that try to explain how ArkTS can help improve program stability and performance.

Explicit Initialization of Fields for Better Stability

ArkTS requires that all fields are explicitly initialized with some values either when the field is declared or in the constructor. This is similar to strictPropertyInitialization mode of the standard TypeScript.

Let’s take a look at the following TypeScript code:

class Person {
    name: string // Automatically is set to undefined

    setName(n:string): void {
        this.name = n
    }

    getName(): string {
    // Return type "string" hides from the developers the fact
    // that name can be undefined. The most correct would be
    // to write the return type as "string|undefined". By doing so
    // we tell the users of our API about all possible return values.
        return this.name
    }
}

let buddy = new Person()
// Let's assume that the developer forgets to call setName:
// buddy.setName("John")
console.log(buddy.getName().length); // runtime exception: name is undefined

Since ArkTS requires explicit initialization, the code looks like this:

class Person {
    name: string = "" // The field always is defined

    setName(n:string): void {
        this.name = n
    }

    // The type is string in all cases, null and undefined are impossible.
    getName(): string {
        return this.name
    }
}

let buddy = new Person()
// Let's assume that the developer forgets to call setName:
// buddys.setName("John")
console.log(buddy.getName().length); // 0, no runtime error

If name can be undefined, this is also should be specified explicitly:

class Person {
    name ?: string // The field may be undefined

    setName(n:string): void {
        this.name = n
    }

    // Compile-time error:
    // name can be "undefined", so we cannot say to those who use this API
    // that it returns only strings:
    getNameWrong(): string {
        return this.name
    }

    getName(): string|undefined { // Return type matches the type of name
        return this.name
    }
}

let buddy = new Person()
// Let's assume that the developer forgets to call setName:
// buddy.setName("John")

// Compile-time(!) error: Compiler suspects that we
// may possibly access something undefined and won't build the code:
console.log(buddy.getName().length); // The code won't build and run

console.log(buddy.getName()?.length); // Builds ok, no runtime error

This case demonstrates how we can improve stability and correctness of our code by enforcing stricter type checking in ArkTS.

Null Safety for Better Performance

Let’s take a look at the following code:

function notify(who: string, what: string) {
    console.log(`Dear ${who}, a message for you: ${what}`)
}

notify("Jack", "You look great today")

In most cases, the notify function will take two string variables as an input and produces a new string. However, what if we pass some “special” values to the function, for example notify(null, undefined)? The program will continue to work, the output will be as expected (Dear undefined, a message for you: null), so from the first glance everything is fine. But please note that the engine that runs our code should always check for such special cases to ensure correct behavior. In pseudocode, something like this happens:

function __internal_tostring(s: any): string {
    if (typeof s === "string")
        return s
    if (s === undefined)
        return "undefined"
    if (s === null)
        return "null"
    // ...
}

Now imagine that our notify function is a part of some complex heavy-loaded system which sends real notifications instead of just writing to the log. In this scenario, executing all these checks from our __internal_tostring function may turn into a performance problem.

But what if we could somehow guarantee to our exectuion engine that the only values that are passed to the notify function are “real” strings, but not some “special” values like null or undefined? In this case, checks like __internal_tostring become redundant because when we execute the program we are 100% sure that there will be no corner cases. For this particular case this mechanism would be called “null-safety”, i.e. guarantee that null is not a valid value of the string type. If we had such feature, the code would not simply build:

function notify(who: string, what: string) {
    console.log(`Dear ${who}, a message for you: ${what}`)
}

notify("Jack", "You look great today")
notify(null, undefined) // Compile-time error

In TypeScript such behavior can be turned on by a special compiler flag called strictNullChecks. But since the standard TypeScript is compiled to JavaScript, which does not have such feature, “strict null checks” work only in compile-time, for better type checking. However, ArkTS considers null-safety a very important feature from both stability and performance points of view. That’s why it is enforced in the language and the example above always produces compile-time errors. In exchange, we give our running engine much more information and guarantees about possible type values, which helps better optimize performance.