Since r165841, the connection is opened after the pending messages are sent, because connectionWillOpen might send messages that we want to happen after the ones already pending. The problem is that Connection::canSendOutgoingMessages() returns false when connection hasn't been opened (m_isConnected). So, I think we should not send any message before Connection::open(). This broke a lot of GTK+ unit tests since it's very common in unit tests to have messages sent before the web process has been launched. I think we can change connectionWillOpen, and use ConnectionDidOpen instead, called after Connection::open.
Created attachment 229366 [details]
I think we should ensure no messages are sent before the connection has been opened. This patch moves the visited link provider process initialization after the connection has been opened.
This patch solved many crashes of layout tests/unit tests on the EFL port.
Could someone review this, please? Otherwise I would suggest to rollout r165841.
This patch also fixes bug 127352. See: https://bugs.webkit.org/show_bug.cgi?id=127352#c14
I think it is a duplicate of Bug131037. Zsolt, could you check if this
patch solves the problem you ran into with EFL perf bot in Bug131037.
As far as I know r165841 is reverted locally on the EFL perf bot.
I have double-checked, this patch solves the problem that was mentioned in https://bugs.webkit.org/show_bug.cgi?id=131037 Thanks.
*** Bug 131037 has been marked as a duplicate of this bug. ***
In my opinion, if more than a month wasn't enough for WK2 owners to review
the fix for the regression they caused, we should simply revert r165841.
(In reply to comment #9)
> In my opinion, if more than a month wasn't enough for WK2 owners to review
> the fix for the regression they caused, we should simply revert r165841.
Perhaps we should just send a polite request to the mailing list asking for a review. It seems like this patch is quite important, as it's causing our bots to rot.
(In reply to comment #4)
> Could someone review this, please?
The patch not longer applies on trunk. It needs to be rebased.
(In reply to comment #11)
> (In reply to comment #4)
> > Could someone review this, please?
> The patch not longer applies on trunk. It needs to be rebased.
Yes, now we have the same problem with the user content controller.
Created attachment 231972 [details]
Comment on attachment 231972 [details]
Patch looks good, please land it. If it turns out to be a problem for Mac we can roll it out and figure out a better way to solve it!
Committed r169306: <http://trac.webkit.org/changeset/169306>