Bug 247372

See https://github.com/GoogleChromeLabs/comlink/issues/600.

The basic pattern comlink uses creates a MessageChannel, keeps one MessagePort for itself and transfers the other to the web worker.  It subscribes to the port it keeps for itself using AddEventListener, but otherwise does not retain a reference to that port, apart from the callback to AddEventListener which should keep the port alive...except, under certain conditions (memory pressure being a key factor), the port it keeps stops receiving message events from the paired web worker.  Once it stops receiving messages, it appears to stop forever -- it's not the case it drops occasional messages, but rather it just appears to permanently break.

A work around that has 100% mitigated the problem in our application is to patch the MessagePort add/remove event listener methods to add an explicit reference to the MessagePort.  E.g., this isn't what we did exactly, but it illustrates the idea:


        const retainedPorts = new Set();
        const pAddEventListener = MessagePort.prototype.addEventListener;
        MessagePort.prototype.addEventListener = function (
            type: string,
            listener: EventListenerOrEventListenerObject,
            options?: boolean | AddEventListenerOptions
        ) {
            if (type === 'message') {
                retainedPorts.add(this);
            }

            pAddEventListener.call(this, type, listener, options);
        }


With that change, the MessagePort keeps receiving messages.  Since a main effect of adding that reference within the Set is to keep the MessagePort from being garbage collected, my best guess is that Safari is inappropriately garbage collecting the port under certain conditions.  It is far from a 100% repro, however.

Our app uses postmessage to web workers in core infrastructure, so this issue completely breaks us, which is why I marked this as major. I see a couple other issues reported in this space, but these have been active for years without being addressed and are not reflective, I think, of the severity.

https://bugs.webkit.org/show_bug.cgi?id=193184
https://bugs.webkit.org/show_bug.cgi?id=184502