## Description
This PR upgrades Prettier to v2 + enforces TypeScript’s [`import
type`](https://www.typescriptlang.org/docs/handbook/release-notes/typescript-3-8.html#type-only-imports-and-export)
syntax where applicable. It’s submitted as a separate PR so we can merge
it easily.
As a part of this PR, we reformat the codebase heavily:
- add `import type` everywhere where it’s required, and
- re-format the code to account for Prettier 2’s breaking changes:
https://prettier.io/blog/2020/03/21/2.0.0.html#breaking-changes
This PR is submitted against `release` to make sure all new code by team
members will adhere to new formatting standards, and we’ll have fewer
conflicts when merging `bundle-optimizations` into `release`. (I’ll
merge `release` back into `bundle-optimizations` once this PR is
merged.)
### Why is this needed?
This PR is needed because, for the Lodash optimization from
7cbb12af88,
we need to use `import type`. Otherwise, `babel-plugin-lodash` complains
that `LoDashStatic` is not a lodash function.
However, just using `import type` in the current codebase will give you
this:
<img width="962" alt="Screenshot 2023-03-08 at 17 45 59"
src="https://user-images.githubusercontent.com/2953267/223775744-407afa0c-e8b9-44a1-90f9-b879348da57f.png">
That’s because Prettier 1 can’t parse `import type` at all. To parse it,
we need to upgrade to Prettier 2.
### Why enforce `import type`?
Apart from just enabling `import type` support, this PR enforces
specifying `import type` everywhere it’s needed. (Developers will get
immediate TypeScript and ESLint errors when they forget to do so.)
I’m doing this because I believe `import type` improves DX and makes
refactorings easier.
Let’s say you had a few imports like below. Can you tell which of these
imports will increase the bundle size? (Tip: it’s not all of them!)
```ts
// app/client/src/workers/Linting/utils.ts
import { Position } from "codemirror";
import { LintError as JSHintError, LintOptions } from "jshint";
import { get, isEmpty, isNumber, keys, last, set } from "lodash";
```
It’s pretty hard, right?
What about now?
```ts
// app/client/src/workers/Linting/utils.ts
import type { Position } from "codemirror";
import type { LintError as JSHintError, LintOptions } from "jshint";
import { get, isEmpty, isNumber, keys, last, set } from "lodash";
```
Now, it’s clear that only `lodash` will be bundled.
This helps developers to see which imports are problematic, but it
_also_ helps with refactorings. Now, if you want to see where
`codemirror` is bundled, you can just grep for `import \{.*\} from
"codemirror"` – and you won’t get any type-only imports.
This also helps (some) bundlers. Upon transpiling, TypeScript erases
type-only imports completely. In some environment (not ours), this makes
the bundle smaller, as the bundler doesn’t need to bundle type-only
imports anymore.
## Type of change
- Chore (housekeeping or task changes that don't impact user perception)
## How Has This Been Tested?
This was tested to not break the build.
### Test Plan
> Add Testsmith test cases links that relate to this PR
### Issues raised during DP testing
> Link issues raised during DP testing for better visiblity and tracking
(copy link from comments dropped on this PR)
## Checklist:
### Dev activity
- [x] My code follows the style guidelines of this project
- [ ] I have performed a self-review of my own code
- [ ] I have commented my code, particularly in hard-to-understand areas
- [ ] I have made corresponding changes to the documentation
- [x] My changes generate no new warnings
- [ ] I have added tests that prove my fix is effective or that my
feature works
- [ ] New and existing unit tests pass locally with my changes
- [ ] PR is being merged under a feature flag
### QA activity:
- [ ] Test plan has been approved by relevant developers
- [ ] Test plan has been peer reviewed by QA
- [ ] Cypress test cases have been added and approved by either SDET or
manual QA
- [ ] Organized project review call with relevant stakeholders after
Round 1/2 of QA
- [ ] Added Test Plan Approved label after reveiwing all Cypress test
---------
Co-authored-by: Satish Gandham <hello@satishgandham.com>
Co-authored-by: Satish Gandham <satish.iitg@gmail.com>
220 lines
6.9 KiB
TypeScript
220 lines
6.9 KiB
TypeScript
import { cancelled, delay, put, take } from "redux-saga/effects";
|
|
import type { Channel } from "redux-saga";
|
|
import { channel, buffers } from "redux-saga";
|
|
import { uniqueId } from "lodash";
|
|
import log from "loglevel";
|
|
import type { TMessage } from "./MessageUtil";
|
|
import { MessageType, sendMessage } from "./MessageUtil";
|
|
|
|
/**
|
|
* Wrap a webworker to provide a synchronous request-response semantic.
|
|
*
|
|
* Usage on main thread:
|
|
* w = GracefulWorkerService(Worker);
|
|
* yield w.start(); // Start the worker
|
|
* const workerResponse = yield w.request("my_action", { hello: "world" }); // Send a request, wait for response
|
|
*
|
|
* Worker will receive:
|
|
* {
|
|
* method: "my_action",
|
|
* requestId: "<unique request id>",
|
|
* requestData: { hello: "world" },
|
|
* }
|
|
*
|
|
* Worker is expected to respond with an object with exactly the `requestId`, `timeTaken` and `responseData` keys:
|
|
* {
|
|
* requestId: "<the id it received>",
|
|
* responseData: 42,
|
|
* timeTaken: 23.33,
|
|
* }
|
|
* All other keys will be ignored.
|
|
* We make no assumptions about data type of `requestData` or `responseData`.
|
|
*
|
|
* Note: The worker will hold ALL requests, even in case of restarts.
|
|
* If we do not want that behaviour, we should create a new GracefulWorkerService.
|
|
*/
|
|
// TODO: Extract the worker wrapper into a library to be useful to anyone with WebWorkers + redux-saga.
|
|
// TODO: Add support for timeouts on requests and shutdown.
|
|
// TODO: Add a readiness + liveness probes.
|
|
export class GracefulWorkerService {
|
|
// We keep track of all in-flight requests with these channels.
|
|
private readonly _channels: Map<string, Channel<any>>;
|
|
// The actual WebWorker
|
|
private _Worker: Worker | undefined;
|
|
|
|
// Channels in redux-saga are NOT like signals.
|
|
// They operate in `pulse` mode of a signal. But `readiness` is more like a continuous signal.
|
|
// This variable provides the equivalent of the `hold` state signal.
|
|
// If isReady is false, wait on `this._readyChan` to get the pulse signal.
|
|
private _isReady: boolean;
|
|
// Channel to signal all waiters that we're ready. Always use it with `this._isReady`.
|
|
private readonly _readyChan: Channel<any>;
|
|
|
|
private readonly _workerClass: Worker;
|
|
|
|
private listenerChannel: Channel<TMessage<any>>;
|
|
|
|
constructor(workerClass: Worker) {
|
|
this.shutdown = this.shutdown.bind(this);
|
|
this.start = this.start.bind(this);
|
|
this._broker = this._broker.bind(this);
|
|
this.request = this.request.bind(this);
|
|
this.respond = this.respond.bind(this);
|
|
this.ping = this.ping.bind(this);
|
|
|
|
// Do not buffer messages on this channel
|
|
this._readyChan = channel(buffers.none());
|
|
this._isReady = false;
|
|
this._channels = new Map<string, Channel<any>>();
|
|
this._workerClass = workerClass;
|
|
this.listenerChannel = channel();
|
|
}
|
|
|
|
/**
|
|
* Start a new worker and registers our broker.
|
|
* Note: If the worker is already running, this is a no-op
|
|
*/
|
|
*start() {
|
|
if (this._isReady || this._Worker) return;
|
|
this._Worker = this._workerClass;
|
|
this._Worker.addEventListener("message", this._broker);
|
|
// Inform all pending requests that we're good to go!
|
|
this._isReady = true;
|
|
yield put(this._readyChan, true);
|
|
return this.listenerChannel;
|
|
}
|
|
|
|
/**
|
|
* Gracefully shutdown the worker.
|
|
* Note: If the worker is already stopped / shutting down, this is a no-op
|
|
*/
|
|
*shutdown() {
|
|
if (!this._isReady) return;
|
|
// stop accepting new requests
|
|
this._isReady = false;
|
|
// wait for current responses to drain, check every 10 milliseconds
|
|
while (this._channels.size > 0) {
|
|
yield delay(10);
|
|
}
|
|
// close the worker
|
|
if (!this._Worker) return;
|
|
this._Worker.removeEventListener("message", this._broker);
|
|
this._Worker.terminate();
|
|
this._Worker = undefined;
|
|
this.listenerChannel.close();
|
|
}
|
|
|
|
/**
|
|
* Check if the worker is ready, optionally block on it.
|
|
*/
|
|
*ready(block = false) {
|
|
if (this._isReady && this._Worker) return true;
|
|
if (block) {
|
|
yield take(this._readyChan);
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
*respond(messageId = "", data = {}): any {
|
|
if (!messageId) return;
|
|
yield this.ready(true);
|
|
if (!this._Worker) return;
|
|
const messageType = MessageType.RESPONSE;
|
|
sendMessage.call(this._Worker, {
|
|
body: {
|
|
data,
|
|
},
|
|
messageId,
|
|
messageType,
|
|
});
|
|
}
|
|
|
|
*ping(data = {}, messageId?: string): any {
|
|
yield this.ready(true);
|
|
if (!this._Worker) return;
|
|
const messageType = MessageType.DEFAULT;
|
|
sendMessage.call(this._Worker, {
|
|
body: data,
|
|
messageId,
|
|
messageType,
|
|
});
|
|
}
|
|
|
|
/**
|
|
* Send a request to the worker for processing.
|
|
* If the worker isn't ready, we wait for it to become ready.
|
|
*
|
|
* @param method identifier for a rpc method
|
|
* @param requestData data that we want to send over to the worker
|
|
*
|
|
* @returns response from the worker
|
|
*/
|
|
*request(method: string, data = {}): any {
|
|
yield this.ready(true);
|
|
// Impossible case, but helps avoid `?` later in code and makes it clearer.
|
|
if (!this._Worker) return;
|
|
|
|
/**
|
|
* We create a unique channel to wait for a response of this specific request.
|
|
*/
|
|
const messageId = `${method}__${uniqueId()}`;
|
|
const ch = channel();
|
|
this._channels.set(messageId, ch);
|
|
const mainThreadStartTime = performance.now();
|
|
let timeTaken;
|
|
|
|
try {
|
|
sendMessage.call(this._Worker, {
|
|
messageType: MessageType.REQUEST,
|
|
body: {
|
|
method,
|
|
data,
|
|
},
|
|
messageId,
|
|
});
|
|
// The `this._broker` method is listening to events and will pass response to us over this channel.
|
|
const response = yield take(ch);
|
|
timeTaken = response.timeTaken;
|
|
const { data: responseData } = response;
|
|
return responseData;
|
|
} finally {
|
|
// Log perf of main thread and worker
|
|
const mainThreadEndTime = performance.now();
|
|
const timeTakenOnMainThread = mainThreadEndTime - mainThreadStartTime;
|
|
if (yield cancelled()) {
|
|
log.debug(
|
|
`Main ${method} cancelled in ${timeTakenOnMainThread.toFixed(2)}ms`,
|
|
);
|
|
} else {
|
|
log.debug(`Main ${method} took ${timeTakenOnMainThread.toFixed(2)}ms`);
|
|
}
|
|
|
|
if (timeTaken) {
|
|
const transferTime = timeTakenOnMainThread - timeTaken;
|
|
log.debug(` Worker ${method} took ${timeTaken}ms`);
|
|
log.debug(` Transfer ${method} took ${transferTime.toFixed(2)}ms`);
|
|
}
|
|
// Cleanup
|
|
ch.close();
|
|
this._channels.delete(messageId);
|
|
}
|
|
}
|
|
|
|
private _broker(event: MessageEvent<TMessage<any>>) {
|
|
if (!event || !event.data) return;
|
|
const { body, messageType } = event.data;
|
|
if (messageType === MessageType.RESPONSE) {
|
|
const { messageId } = event.data;
|
|
if (!messageId) return;
|
|
const ch = this._channels.get(messageId);
|
|
if (ch) {
|
|
ch.put(body);
|
|
this._channels.delete(messageId);
|
|
}
|
|
} else {
|
|
this.listenerChannel.put(event.data);
|
|
}
|
|
}
|
|
}
|