Bug 43974 – Geolocation requests in progress when the frame is disconnected should invoke the error callback

• Status: RESOLVED
• Resolution: FIXED
• Component: WebCore Misc.

[Steve Block]

Currently, Geolocation::disconnectFrame() calls stopTimers() which prevents any future callbacks. This was added in http://trac.webkit.org/changeset/59859 for Bug 39388 to prevent callbacks after the script context has gone away.

Instead, we should attempt to invoke the error callback for all ongoing requests. This is safe because we now make sure that the script context is still good before we make each callback, see Bug 40162.

[Steve Block]

Created attachment 64349 [details]
Patch

[Steve Block]

Created attachment 64481 [details]
Patch

[Alexey Proskuryakov]

Comment on attachment 64481 [details]
Patch

> Instead, we should attempt to invoke the error callback for all ongoing requests.

I think that this is a good change because it matches XMLHttpRequest (which dispatches an abort event when window is closed). But for posterity, it would be best to document why else we want this. Does any spec say so?

+    if (m_fatalError)
+        return;

It might be helpful to add a comment explaining why the first error wins.

r=me. The split js-test makes me sad.

[Steve Block]

Committed r65416: <http://trac.webkit.org/changeset/65416>

[Steve Block]

> I think that this is a good change because it matches XMLHttpRequest (which
> dispatches an abort event when window is closed). But for posterity, it would
> be best to document why else we want this. Does any spec say so?
No, the spec isn't specific to JavaScript or browsers so doesn't mention anything about Frames or windows.

> It might be helpful to add a comment explaining why the first error wins.
Done

[WebKit Review Bot]

http://trac.webkit.org/changeset/65416 might have broken SnowLeopard Intel Release (Tests)

[Steve Block]

Problems with fast/dom/Geolocation/disconnected-frame-permission-denied.html are being tracked in Bug 44059