HMR runtime references global by wrong name in Worker

[bwrrp]

🐛 bug report

I've created a very simple project with a web worker. It looks like the script reference in the Worker constructor call is picked up automatically by Parcel and compiled into a separate bundle, which is great. However, the generated bundle by default includes the hot module reloading runtime, which contains code referencing window.location. In a Worker scope, the global is called self, and the bundle code throws a ReferenceError before the actual worker script can run.

🎛 Configuration / code sample

Create the following files and run parcel serve index.html

index.js

var w = new Worker('./worker.js');
w.onmessage = e => console.log(e.data);
setInterval(() => w.postMessage('test'), 1000);

worker.js

self.onmessage = e => self.postMessage(e.data + ' from worker');

index.html

<script src="./index.js"></script>

🤔 Expected Behavior

Expected code to load and work correctly.

😯 Current Behavior

Uncaught ReferenceError: window is not defined

The code in worker.js does not run and no "test" messages are logged to the console.

💁 Possible Solution

The error is thrown in what looks to be the following code, included in the Worker bundle:

parcel/src/builtins/hmr-runtime.js

Line 18 in 06d5ffc

var ws = new WebSocket('ws://' + window.location.hostname + ':{{HMR_PORT}}/');

The HRM runtime should either detect it is running in a Worker scope and use the self global, or be disabled for Worker bundles if there are other issues that prevent it from working in that context.

🌍 Your Environment

Software	Version(s)
Parcel	1.4.1
Node	8.4.0
npm/Yarn	npm 5.6.0
Operating System	Windows 10

[DeMoorJasper]

related to #476

[bwrrp]

Oops, I think I forgot to search for PRs. This is indeed the same issue, sorry for the duplicate.

[habemus-papadum]

Hello, I'm currently getting an HMR error in my WebWorkers that seems to be related to a use of the window global in the hmr runtime here. It wasn't clear to me if the fix was as simple as changing window.location.reload() to location.reload() and if there had been an unintentional regression introduced by #2676 so for now I'm just leaving this note -- any comments welcome. thanks, nehal