1552186 - Remove nsIRemoteTab::windowGlobalParents getter

Status: RESOLVED FIXED

(Core :: DOM: Content Processes, defect)

Description

[Mike Conley (:mconley) (:⚙️)]

Bug 1549753 introduced a new handy getter, windowGlobalParents, on nsIRemoteTab.

Right now, it returns all WindowGlobalParents for all subtrees.

This was originally in support of my work in bug 1533949 - however, upon testing, there's a slight alteration that we need in order for bug 1533949 to finish.

Specifically, I'm hoping windowGlobalParents can returns only the WindowGlobalParent's from the roots of process-contiguous subtrees.

Hopefully this will help illustrate what I mean by process-contiguous trees:

Suppose that a document has the following frame structure:

                  a
               /     \
              b        c
            /   \    /  \
           b    d   c    c
               /          \
              d            d

where each letter represents the process that each frame runs in, then we consider
process-contiguous trees of frames to be ones where all frames are directly connected
and running in the same process. Process-contiguous subtrees are denoted in the following
graph with shared numbers.

                  a1
               /     \
              b2        c3
            /    \    /   \
           b2    d4  c3   c3
                /          \
              d4            d5

Specifically, note that the d5 leaf node, while belonging to the same process as the
d4 nodes are not directly connected, so it's not process-contiguous.

Comment 1

[Mike Conley (:mconley) (:⚙️)]

To further illustrate, I've marked the roots of the process contiguous subtrees with a *:

                  a1*
               /     \
              b2*      c3*
            /    \    /   \
           b2    d4* c3   c3
                /          \
              d4            d5*

Comment 2

[Ryan Hunt [:rhunt]]

Thanks for the diagrams, those really helped!

Taking another attempt here, I think we can do this with two pieces.

1. <xul:browser>.frameloader.browsingContext.currentWindowGlobalParent (already exists)
   a. This should give you 'a' in that diagram
2. WindowGlobalParent.getNearestRemoteDescendants() (doesn't exist yet)
   a. This would give you the closest '' descendants of a given WindowGlobalParent
   b. e.g. For a1, you would get [b2*, c3*]. For b2*, you would get [d4*]. For c3*, you would get [d5*].
   b. With this you could traverse the contiguous browser tree roots by starting at currentWindowGlobalParent

Comment 3

[Ryan Hunt [:rhunt]]

Attachment: Bug 1552186 - Add `getNearestRemoteDescendants` method to WindowGlobalParent. r?nika

This commit adds getNearestRemoteDescendants as a way to get the contiguous
browser subtree roots that are descendant from a given WindowGlobalParent.

Comment 4

[Ryan Hunt [:rhunt]]

Attachment: Bug 1552186 - Remove `windowGlobalParents` getter from nsIRemoteTab. r?nika

This method wasn't correct for what mconley needed and is no longer needed.

Depends on D31505

Comment 5

[Ryan Hunt [:rhunt]]

Now that bug 1554280 has landed, is it possible to implement this method entirely in JS now?

Roughly something like:

function getNearestRemoteDescendants(frameLoader) {
  let result = [];
  function visitBC(bc) {
    if (!bc) {
      return;
    }
    let wgp = child.currentWindowGlobal;
    if (wgp && wgp.isProcessRoot) {
      result.push(wgp);
    }
    for (let child of bc.getChildren()) {
      visitBC(child);
    }
  }
  visitBC(frameLoader.browsingContext);
  return result;
}

I'm not sure if something additional is needed to filter out BFCache'd WGP's, or if only checking currentWindowGlobal's is enough.

Comment 6

[Nika Layzell [:nika] (ni? for response)]

(In reply to Ryan Hunt [:rhunt] from comment #5)

Now that bug 1554280 has landed, is it possible to implement this method entirely in JS now?

Yeah, it should be doable. Your method is missing a few return calls, (e.g. we shouldn't look at the children of a BC which is also a process root), and has some minor typos (like child instead of bc), but should be correct.

Comment 7

[Mike Conley (:mconley) (:⚙️)]

Cool, yeah, I can use something like that implementation in JS now for the PermitUnload case, so I guess we can close this out as essentially fixed by bug 1554280. Thanks!

Comment 8

[Ryan Hunt [:rhunt]]

I'm going to reuse this bug to remove the getter that was originally added to solve this problem.

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/b2620ccb1d84
Remove `windowGlobalParents` getter from nsIRemoteTab. r=nika

Comment 10

[Stefan Hindli [:stefan_hindli]]

https://hg.mozilla.org/mozilla-central/rev/b2620ccb1d84