Re-bind placeholder nodes after they reconnect to the document

[bmomberger-bitovi]

Sometimes users will move component markup around in the DOM. A reduced example can be seen here: https://codepen.io/bmomberger-bitovi/pen/bGNYvBg?editors=1010

The underlying reason for this is that can-view-live tears down live binding on placeholders when those placeholders are removed from the document. But this also breaks live binding on nodes which are just getting shuffled around by other code. This can happen with jQuery plugins and other mutating libraries that happen to work alongside CanJS.

This PR addresses this possibility by letting removed placeholder nodes re-bind if they are later added to the document.

[bmomberger-bitovi]

The additional text test is currently failing on IE11. After removing the text node, no updates happen even after reconnection. This will have to be considered a WIP until that failure is addressed, but should still be reviewed.

[bmomberger-bitovi]

This part of the patch set is boyscouting from the queue changes I made recently. Since updateRange is never queued, it is not necessary to remove it from the dom queue.

[justinbmeyer]

Does the ability to rebind create a memory leak? I couldn’t think of a way to make this work without creating a memory leak.

Does can-view-live support move anymore? I might have made a way to ignore DOM nodes moving.

[bmomberger-bitovi]

I don't think this is in danger of memory leaks now that we only use WeakMaps in can-dom-mutate. I tried profiling the window from the QUnit tests to see for sure, and what I found was that the only "leaks" came from the no can.setElement symbol on observable warnings and one test that set up a mutation listener on the document body.

[justinbmeyer]

I thought can-dom-mutate also held things in normal arrays and such so it could loop through handlers

[bmomberger-bitovi]

I think you're referring to https://github.com/canjs/can-dom-mutate/blob/master/can-dom-mutate.js#L84-L87 ?

The arrays are keyed on the elements (and documents) in the WeakMaps, so when elements are garbage collected the whole array is also GC'ed.

Mutation records are held in normal arrays while flushing and queueing, but those arrays are emptied regularly.

[bmomberger-bitovi]

If you're still concerned about leaks, I adapted the benchmark test from can-dom-mutate, and can show that after you create, insert, and remove a bunch of elements (both text and comment placeholders), once garbage is collected, the weakmap is empty of entries:

[bmomberger-bitovi]

For reference this is the function being run on click:

	function run() {
		var doc = document.implementation.createHTMLDocument();
		var div = doc.createElement("div");
		doc.body.appendChild(div);
		// create range of comments
		div.appendChild(doc.createComment(""));
		div.appendChild(doc.createComment("can-end-placeholder"));
		live.html(div.firstChild, new Observation(() => "<span>foo</span>"));
		var text = doc.createTextNode("bar");
		div.appendChild(text);
		live.text(text, new Observation(() => "baz"));

		DOCUMENT(doc);
		doc.body.removeChild(div);
		DOCUMENT(document);
	}