Apparently, scripts from topix.net are causing the page to scroll down opening them. Examples: http://www.dispatch.com/live/content/local_news/stories/2010/10/18/motorist-killed-when-car-police-cruiser-collide.html?sid=101 http://www.dispatch.com/live/content/local_news/stories/2010/10/18/18-bus-school-close.html?sid=101 http://www.dispatch.com/live/content/local_news/stories/2010/10/18/18-dump-dem.html?sid=101 It also appears to happen on this link: http://newsbusters.org/blogs/tim-graham/2010/10/18/wapo-promotes-gospel-according-jimmy-they-find-carters-preachy-know-it-a I cannot tell if they use topix scripts or just scripts similar to those on topix. If you open any of these in Minefield/Trunk, the pages will open and then scroll down. Pages should only open. (I did not see this on IE9.) I have tested in both Win7 and WinXP using the 32-bit builds in both safe mode and using a fairly new, no add-ons installed profile. Others have confirmed this bug, too. I think this started with the compartments landing. The nightly build of 13-Oct does not show the problem. The nightly of 14-Oct does.
FYI-May need to be logged in to the newsbusters.org web site to see this behavior.
Regression window: Works: http://hg.mozilla.org/mozilla-central/rev/b829cf5debce Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b8pre) Gecko/20101012 Firefox/4.0b8pre ID:20101013074132 Fails: http://hg.mozilla.org/mozilla-central/rev/178f26e21cfc Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b8pre) Gecko/20101011 Firefox/4.0b8pre ID:20101013081930 Pushlog: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=b829cf5debce&tochange=178f26e21cfc
Probably bug 597331
Happens on Linux too. http://hg.mozilla.org/mozilla-central/rev/cfd18201f49b Mozilla/5.0 (X11; Linux i686; rv:2.0b8pre) Gecko/20101018 Firefox/4.0b8pre ID:20101018030619 In local build, Build from 178f26e21cfc : fails Build from 18d0545c5f0e : works So, Landing of the following cset causes the problem. changeset 178f26e21cfc Ehsan Akhgari — Bug 597331 - Reframing a textarea sets the caret position to the end of its contents; r=bzbarsky a=blocking-final+
Created attachment 484305 [details] reduced sample CSS file should be external CSS file. Strange to say, If CSS is written in script tag, the scroll does not happen. And If removed SCRIPT tag, the scroll does not happen.
Created attachment 484698 [details] Reduced testcase Here's the most reduced testcase for this bug.
Created attachment 484858 [details] [diff] [review] Patch (v1) Please note that this requires an API change (so that we can pass SCROLL_FIRST_ANCESTOR_ONLY to nsIPresShell::ScrollFrameRectIntoView if needed. Can this land post b7, or on the b7 relbranch as well?
Instead of one (or multiple) PRBool parameters, can we use a flags word please?
Created attachment 485113 [details] [diff] [review] Patch (v2) (In reply to comment #9) > Instead of one (or multiple) PRBool parameters, can we use a flags word please? Of course.
You're still adding a boolean parameter to nsSelection::ScrollIntoView, which now has three boolean parameters. Can we flagize them too please?
Created attachment 485222 [details] [diff] [review] Part 2: flagize nsTypedSelection::ScrollIntoView (In reply to comment #11) > You're still adding a boolean parameter to nsSelection::ScrollIntoView, which > now has three boolean parameters. Can we flagize them too please? Sure.
Created attachment 485383 [details] [diff] [review] Part 2: flagize nsTypedSelection::ScrollIntoView I had made a mistake in part 2 of this patch. In nsTypedSelection::ScrollSelectionIntoViewEvent::Run, I had practically removed the aDoFlush and aIsSynchronous parameters and wasn't setting the corresponding flags. This version of the patch fixes that problem.
This bug happens for me in this page: http://es.paperblog.com/25-anos-de-back-to-the-future-primera-parte-el-de-lorean-206615/ It's working fine on Chrome, IE9 and Opera.
I pushed a follow-up to fix a Thunderbird build bustage: http://hg.mozilla.org/mozilla-central/rev/142c7b20fd8c
These API changes do not impact any currently-documented interfaces, so I'm removing the doc needed keyword. These changes will be picked up when these interfaces get documented in the future.