Annotation of `this` in function declarations
jussi-kalliokoski opened this issue · comments
Currently it seems that you can only annotate the this
magic reference by putting the function as a method in a class declaration.
However, in the light of the new bind syntax proposal, it would make sense to be able to annotate functions that take in this
as data, e.g.
function head (count) {
return this.slice(0, count);
}
[1,2,3]::head(2) // [1,2]
could be annotated something like:
function head <T> Array<T> -> (count : number) : Array<T> {
...
}
Flow is (mostly) smart about this already without annotations. I'm not clear what this new bind syntax is (do you have a link to a spec?), but I imagine flow will support it the same way that flow currently supports bind
, call
, and apply
.
/* @flow */
function head(count) {
return this.slice(0, count);
}
head.bind([]) // OK
head.bind(1) // bad
I'm not clear what this new bind syntax is (do you have a link to a spec?)
Ah, sorry, forgot the link, fixed.
Flow is (mostly) smart about this already without annotations
Yes, but it would be nice to be able to annotate this instead, for documentational purposes (easier to understand at a glance and also allows for better error messages, as well as documentation generation), but also for standalone definitions.
Having thought about this a bit more, a simpler syntax would actually be to allow specifying this
as a parameter, as so:
function add (this : number, b : number) : number {
return this + b;
}
function add (/*: this : number, */ b /*: number */) /*: number */ {
return this + b;
}
/*: (this : number, b: number) => number */
function add (b) {
return this + b;
}
Agree that we should support this
to be annotated as a parameter. This requires some parser work too, btw, since this
is not a valid identifier everywhere.
We need this whenever methods are exported, by the way. It's true Flow is good in inferring this
types, but it sucks that this information is lost across modules.
Related: constructor functions can be annotated as of 618186a
Hi guys! what about the following examples: microsoft/TypeScript#1985 (comment)
Are they too complex/ugly?
+1
Pulling in some material from #968: an example, and a link to how Google Closure does this.
There's an error (marked with <---
) in this snippet which Flow could potentially catch:
class RemoteFile {
url: string;
...
promiseXHR(xhr: XMLHttpRequest): Q.Promise<[any, Event]> {
var url = this.url;
var deferred = Q.defer();
xhr.addEventListener('load', function(e) {
if (this.status >= 400) {
deferred.reject(this.status + ' ' + this.statusText);
} else {
deferred.resolve([this.response, e]);
}
});
xhr.addEventListener('error', function(e) {
deferred.reject(`Request for ${this.url} failed: ${this.status}`); // <---
});
this.numNetworkRequests++;
xhr.send();
return deferred.promise;
}
}
The problem is that this
inside the addEventListener
callback refers to the XMLHttpRequest
object (I believe) and not the RemoteFile
class. The latter has a url
property but not a status
property, whereas the former has status
but not url
. Clearly this code isn't right!
Flow doesn't complain, though. It thinks this
has type any
in the event listener.
Where possible, the DOM declarations should indicate the type of this
in callbacks to prevent this sort of mistake.
For example, Google's Closure Compiler provides an @this
annotation. You can see an example of this in action in the declaration for Array.prototype.find
.
In case you land here before this +1 functionality is added, and you need types just for the fn body, just do something like this:
type Ctx = { foo: string, bar: number }
myMethod() {
const myCtx: Ctx = this;
// do stuff with myCtx instead of this
}
Just for reference, TS 2.0 landed with this feature last month. It special-cases the first parameter if its name is this
. I like that, reminds me of Python.
function f(this: void) {
// make sure `this` is unusable in this standalone function
}
interface UIElement {
addClickListener(onclick: (this: void, e: Event) => void): void;
}
An interesting note from the ticket when they were speccing it:
Yeah, the callback literally does "receive" this object as a parameter. That's how it works for all functions. It just happens that there are different rules for how you pass the value of this, it doesn't appear in the arguments array, and there's a different default value in non-strict mode. Otherwise it's basically just another argument.
Makes sense to me. If babel-plugin-flow-strip-types
stripped the entire first parameter when its name is this
, it could work. Will likely require special-casing in ESLint etc.
This already works in Flow:
function head (count: number) {
(this: Array<number>);
return this.slice(0, count);
}
head.call([1, 2, 3], 10);
head.call(['123'], 20); // error
Do we really need new syntax for this?
However there are some bugs with the approach as this totally bugs out:
function head<T>(count: number): Array<T> {
(this: Array<T>);
return this.slice(0, count);
}
head.call([1, 2, 3], 10);
head.call(['123'], 20);
@nmn Yes we do need.
It's for library declaration and use site completion. This is quite common for context injection.
function addListener<T: HTMLElement>(elem: T, func: (this: T) => void) {
// implementation
}
addListener(htmlInputElement , function() {
this.value // should have this completion
})
Another usage is in library like Vue/backbone.
Vue.extend({
methods: {
log() {
this.instanceProperty // this is injected as Vue instance
}
}
})
Without this annotation, it is impossible for library author to provide a good type declaration for automatic completion.
I hope this feature is provided 🙏
+1 for this feature
Has there been no progress on this yet?
@calebmer was looking for people to help implement it a while ago. I'm not sure if anyone started on it. I'd love to do it if I had more time.
Any updates on this? :)
@danvk It seems that you're using flow type version of Q.Promise
, is it? If so do you mind to share a link to flow typed Q.Pomise
library?
@akoppela https://github.com/hammerlab/pileup.js/blob/0182b225205e52e1489727004d69e84442eafb88/lib/q.js but take note that these are ~3 years old and there may be better ones available elsewhere.
Hey! I tried to implement it at #7807
Note that this
constraints in classes can be emulated via $Call
declare class O<T> {
m(): $Call<(string => string), T>;
m(): $Call<(number => number), T>;
}
declare var o1: O<number>
const a1: number = o1.m() // ok
const b1: string = o1.m() // error
declare var o2: O<string>
const a2: number = o2.m() // error
const b2: string = o2.m() // ok
declare var o3: O<{}>
const a3 = o3.m() // error
const b3 = o3.m() // error
^ flow try
class M<+State> {
+s: State
constructor(s: State) { this.s = s }
to2: $Call<(1 => () => M<2>) & () => {...}, State> = () => {
return new M(2) // ok
}
to3: $Call<(1 => () => M<3>) & () => {...}, State> = () => {
return new M(2) // error
}
}
const m1 = new M(1)
const m2 = m1.to2() // ok
const m3 = m2.to2() // error
^ flow try
declare class O<+T> {
constructor(T): void;
m(): $Call<<U>(Array<U>) => U, T>
}
const oarr = new O([1,2,3])
const n: number = oarr.m() // ok
const s: string = oarr.m() // error
declare class Ap<+T> {
ap<A, B>(Ap<A>): $Call<(A => B) => Ap<B>, T>
}
const a = new Ap<number => string>
const m: Ap<number> = a.ap(new Ap<3>) // error
const d: Ap<string> = a.ap(new Ap<3>) // ok
^ flow try
declare class A {
m(): $Call<B => 1, this>;
n(): $Call<<U>(C<U>) => U, this>;
}
declare class B extends A {}
declare class C<T> extends A {}
const a = new A
a.m() // error (not in this place tho)
const b = new B
b.m() // ok
const c = new C<string>()
const n0: string = c.n() // ok
const n1: number = c.n() // error
^ flow try
Wow, this is gold! Thank you @Bannerets!
Fixing this
was mentioned in the latest roadmap.
The this
annotation was released in 0.139.
To use, add this to flowconfig:
[options]
experimental.this_annot`= true
To use:
function foo(this: {toString: (number) => string}, a: number): string {
return this.toString(a);
}
It is also usable at https://flow.org/try.