[Webkit-unassigned] [Bug 174633] Web Automation: implicit navigations leak frame objects and don't cause browsing context switch

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=174633

--- Comment #1 from Brian Burg <bburg at apple.com> ---
(In reply to Carlos Garcia Campos from comment #0)
> I've noticed this debugging
> testShouldFocusOnTheReplacementWhenAFrameFollowsALinkToA_TopTargetedPage.
> The situation is something like this:
> 
> 1. a frame set is loaded
> 2. switch frame to any frame in the set
> 3. get a link from the subframe and click it
> 4. a new page is loaded in the main frame
> 5. find an element that is in the new page
> 
> from 3 to 4 there's actually a browsing context switch, but web driver
> doesn't know about it. The subrame from which the navigation happened is
> still the current browsing context, and it's still referenced in
> WebAutomationSession (frames are never removed from the maps) and in
> WebAutomationSessionProxy (the script object is still protected, and pending
> callbacks if any are nos dispatched, it seems we are leaking the frame in
> the web process). So, step 5 is still trying to find the element in the
> previous frame, and since the frame is still alive, the
> evaluateJavaScriptFunction is called, but the element is not found and
> returns a no such element error. This would be even more problematic if the
> previous page had an element matching the find command.

Great investigation!

> I'm not sure how to
> solve this, if the frame was not leaked in the web process, at least we
> could get a no such frame error, and handle it as special case, switching to
> top level and trying again. Something like that is done in chromium:
> 
> if (attempt == 2) {
>    // Switch to main frame and retry command if subframe no longer exists.  
> 
>    session->SwitchToTopFrame();
> }
> 
> I don't like this approach that much, but if there's no other way, at least
> we should do it only when error returned is no such frame, (and there's an
> active browsing context, of course).
> 
> Any other ideas? Did you solve that in safaridriver somehow?

No we haven't solved this. This is probably the cause of many nonspecific bugs we get with unexpected "element not found" errors.

What about changing the context in WebAutomationSession::navigationOccurredForPage() if needed, and sending a notification out to the driver that the current top level context changed?

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20170718/9f785036/attachment.html>