Bug 220478

Summary: Relax assertion in Element::dispatchFocusOutEvent() for non-web process case
Product: WebKit Reporter: Julian Gonzalez <julian_a_gonzalez>
Component: DOMAssignee: Nobody <webkit-unassigned>
Status: RESOLVED FIXED    
Severity: Normal CC: cdumez, cmarcelo, darin, esprehn+autocc, ews-watchlist, kangil.han, rniwa
Priority: P2 Keywords: InRadar
Version: WebKit Nightly Build   
Hardware: Unspecified   
OS: Unspecified   
Attachments:
Description Flags
Patch none

Julian Gonzalez
Reported 2021-01-08 13:01:26 PST
This assertion (along with the identical one in Element::dispatchFocusInEvent()) can be erroneously hit in DumpRenderTree. frame #0: JavaScriptCore`WTFCrashWithSecurityImplication+9 frame #1: WebCore`WebCore::Element::dispatchFocusOutEvent(WTF::AtomString const&, WTF::RefPtr<WebCore::Element, WTF::RawPtrTraits<WebCore::Element>, WTF::DefaultRefDerefTraits<WebCore::Element> >&&)+730 frame #2: WebCore`WebCore::Document::setFocusedElement(WebCore::Element*, WebCore::FocusDirection, WebCore::Document::FocusRemovalEventsMode)+1279 frame #3: WebCore`WebCore::FocusController::setFocusedElement(WebCore::Element*, WebCore::Frame&, WebCore::FocusDirection)+1498 frame #4: WebCore`WebCore::Element::focus(WebCore::SelectionRestorationMode, WebCore::FocusDirection)+1315 frame #5: WebCore`WebCore::HTMLFormControlElement::didAttachRenderers()::$_1::operator()() const+79 <rdar://problem/72670556>
Attachments
Patch (4.61 KB, patch)
2021-01-08 13:20 PST, Julian Gonzalez
no flags
Julian Gonzalez
Comment 1 2021-01-08 13:20:28 PST
Darin Adler
Comment 2 2021-01-08 13:57:49 PST
Comment on attachment 417298 [details] Patch I don’t get it. this turns off the assertion entirely for legacy WebKit. Why is that correct?
Julian Gonzalez
Comment 3 2021-01-08 14:12:48 PST
(In reply to Darin Adler from comment #2) > Comment on attachment 417298 [details] > Patch > > I don’t get it. this turns off the assertion entirely for legacy WebKit. Why > is that correct? Ryosuke mentioned to me (and on the radar) that: "We probably just need to disable the assertion here. There isn’t much we can do to mitigate this since application that embeds WebKit can do anything they like."
Ryosuke Niwa
Comment 4 2021-01-08 14:40:17 PST
(In reply to Darin Adler from comment #2) > Comment on attachment 417298 [details] > Patch > > I don’t get it. this turns off the assertion entirely for legacy WebKit. Why > is that correct? In the following call stack, frame #22-23 callbacks into WebView delegates. There, client application can do anything. So long as these delegates exist, we can't really guarantee the safety / consistency of WebCore states. In WebKit2, we made editor state update async so this issue of updating the state doesn't really exist. We could, in theory, make _updateFontPanel async but that doesn't resolve the issue that WebEditorClient still issues WebViewDidChangeSelectionNotification which could trigger arbitrary code. We've already had to disable similar kinds of assertions in WebKit1 elsewhere in ScriptController::canExecuteScripts and isSafeToUpdateStyleOrLayout in Document.cpp. frame #0: JavaScriptCore`WTFCrashWithSecurityImplication+9 frame #1: WebCore`WebCore::Element::dispatchFocusOutEvent(WTF::AtomString const&, WTF::RefPtr<WebCore::Element, WTF::RawPtrTraits<WebCore::Element>, WTF::DefaultRefDerefTraits<WebCore::Element> >&&)+730 frame #2: WebCore`WebCore::Document::setFocusedElement(WebCore::Element*, WebCore::FocusDirection, WebCore::Document::FocusRemovalEventsMode)+1279 frame #3: WebCore`WebCore::FocusController::setFocusedElement(WebCore::Element*, WebCore::Frame&, WebCore::FocusDirection)+1498 frame #4: WebCore`WebCore::Element::focus(WebCore::SelectionRestorationMode, WebCore::FocusDirection)+1315 frame #5: WebCore`WebCore::HTMLFormControlElement::didAttachRenderers()::$_1::operator()() const+79 frame #6: WebCore`WTF::Detail::CallableWrapper<WebCore::HTMLFormControlElement::didAttachRenderers()::$_1, void>::call()+13 frame #7: WebCore`WTF::Function<void ()>::operator()() const+63 frame #8: WebCore`WebCore::Style::PostResolutionCallbackDisabler::~PostResolutionCallbackDisabler()+115 frame #9: WebCore`WebCore::Style::PostResolutionCallbackDisabler::~PostResolutionCallbackDisabler()+9 frame #10: WebCore`WebCore::Document::resolveStyle(WebCore::Document::ResolveStyleType)+1998 frame #11: WebCore`WebCore::Document::updateStyleIfNeeded()+572 frame #12: WebCore`WebCore::Document::updateLayout()+404 frame #13: WebCore`WebCore::Document::updateLayoutIgnorePendingStylesheets(WebCore::Document::RunPostLayoutTasks)+147 frame #14: WebCore`WebCore::VisiblePosition::canonicalPosition(WebCore::Position const&)+283 frame #15: WebCore`WebCore::VisiblePosition::VisiblePosition(WebCore::Position const&, WebCore::Affinity)+184 frame #16: WebCore`WebCore::VisiblePosition::VisiblePosition(WebCore::Position const&, WebCore::Affinity)+9 frame #17: WebCore`WebCore::VisibleSelection::visibleStart() const+58 frame #18: WebCore`WebCore::adjustedSelectionStartForStyleComputation(WebCore::VisibleSelection const&)+228 frame #19: WebCore`WebCore::Editor::styleForSelectionStart(WebCore::Frame*, WebCore::Node*&)+355 frame #20: WebCore`WebCore::Editor::fontForSelection(bool&) const+898 frame #21: WebKitLegacy`-[WebHTMLView(WebInternal) _updateFontPanel]+561 frame #22: WebKitLegacy`-[WebHTMLView(WebInternal) _selectionChanged]+95 frame #23: WebKitLegacy`WebEditorClient::respondToChangedSelection(WebCore::Frame*)+210 frame #24: WebCore`WebCore::Editor::respondToChangedSelection(WebCore::VisibleSelection const&, WTF::OptionSet<WebCore::FrameSelection::SetSelectionOption>)+325 frame #25: WebCore`WebCore::FrameSelection::setSelectionWithoutUpdatingAppearance(WebCore::VisibleSelection const&, WTF::OptionSet<WebCore::FrameSelection::SetSelectionOption>, WebCore::FrameSelection::CursorAlignOnScroll, WebCore::TextGranularity)+1745 frame #26: WebCore`WebCore::FrameSelection::setSelection(WebCore::VisibleSelection const&, WTF::OptionSet<WebCore::FrameSelection::SetSelectionOption>, WebCore::AXTextStateChangeIntent, WebCore::FrameSelection::CursorAlignOnScroll, WebCore::TextGranularity)+347 frame #27: WebCore`WebCore::FrameSelection::moveWithoutValidationTo(WebCore::Position const&, WebCore::Position const&, bool, bool, WebCore::SelectionRevealMode, WebCore::AXTextStateChangeIntent const&)+787 frame #28: WebCore`WebCore::HTMLTextFormControlElement::setSelectionRange(int, int, WebCore::TextFieldSelectionDirection, WebCore::SelectionRevealMode, WebCore::AXTextStateChangeIntent const&)+1984 frame #29: WebCore`WebCore::HTMLTextAreaElement::setValueCommon(WTF::String const&)+863 frame #30: WebCore`WebCore::HTMLTextAreaElement::setNonDirtyValue(WTF::String const&)+14 frame #31: WebCore`WebCore::HTMLTextAreaElement::childrenChanged(WebCore::ContainerNode::ChildChange const&)+418 frame #32: WebCore`WebCore::ContainerNode::removeChildren()+1283 frame #33: WebCore`WebCore::ContainerNode::replaceAllChildren(std::nullptr_t)+188
Darin Adler
Comment 5 2021-01-08 14:54:11 PST
Yes, the fix has to be about the timing of calling *out* of legacy WebKit to clients. Not about what the client does.
Ryosuke Niwa
Comment 6 2021-01-08 16:28:43 PST
(In reply to Darin Adler from comment #5) > Yes, the fix has to be about the timing of calling *out* of legacy WebKit to > clients. Not about what the client does. I don't think that would be practical given the number of sync delegate calls that happen when it's not safe to run arbitrary scripts.
Darin Adler
Comment 7 2021-01-08 17:09:13 PST
There are lots of other assertions we’ll have to remove. This is only the first of *many*.
Ryosuke Niwa
Comment 8 2021-01-08 17:19:47 PST
(In reply to Darin Adler from comment #7) > There are lots of other assertions we’ll have to remove. This is only the > first of *many*. That's probably true. But removing assertions won't introduce new app compatibility regressions whilst changing the timing of delegate callbacks will.
EWS
Comment 9 2021-01-11 14:57:10 PST
Committed r271382: <https://trac.webkit.org/changeset/271382> All reviewed patches have been landed. Closing bug and clearing flags on attachment 417298 [details].
Note You need to log in before you can comment on or make changes to this bug.