Reviewed by John.

[WebKit-https.git] / WebCore / ChangeLog-2005-08-23

2005-04-22  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - a small editing-related code cleanup

        * khtml/rendering/render_text.h: Added positionForOffset and made offsetForPosition const.
        * khtml/rendering/render_text.cpp:
        (InlineTextBox::offsetForPosition): Made const.
        (InlineTextBox::positionForOffset): Added. Moved code here from caretRect.
        (RenderText::caretRect): Call positionForOffset instead of doing the work here.

2005-04-21  David Hyatt  <hyatt@apple.com>

        Fix for 4095839, wrong background image used on flechtwerk.de.  Make sure that the global mapped
        attribute cache hashed background attributes into per-document buckets.

        * khtml/html/html_baseimpl.cpp:
        (HTMLBodyElementImpl::mapToEntry):
        * khtml/html/html_elementimpl.h:
        (DOM::):
        * khtml/html/html_tableimpl.cpp:
        (HTMLTableElementImpl::mapToEntry):
        (HTMLTablePartElementImpl::mapToEntry):
        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::DocumentImpl):
        * khtml/xml/dom_docimpl.h:
        (DOM::DocumentImpl::docID):

2005-04-21  Vicki Murley  <vicki@apple.com>

        - layout test for 4065447, outerHTML on images

        * layout-tests/fast/dynamic/outerHTML-img-expected.txt: Added.
        * layout-tests/fast/dynamic/outerHTML-img.html: Added.

2005-04-20  Vicki Murley  <vicki@apple.com>

        Reviewed by hyatt.

        - fixed <rdar://problem/4065447> support outerHTML on IMG elements

        * khtml/html/html_elementimpl.cpp:
        (HTMLElementImpl::setOuterHTML):

2005-04-18  David Hyatt  <hyatt@apple.com>

        Fix min-height so that when it resolves to auto it does not use the box's intrinsic height.
        
        * khtml/rendering/render_box.cpp:
        (RenderBox::calcHeight):
        (RenderBox::calcHeightUsing):

2005-04-18  David Hyatt  <hyatt@apple.com>

        Back out fix for 4032346, since it is causing garbled image content on many sites.

        The bug tracking the fix is 4069093.
        
        (khtml::RenderBlock::matchedEndLine):

2005-04-18  David Hyatt  <hyatt@apple.com>

        Fix the smile in the Acid2 test.  Floats should not grow to contain other floats unless height is auto.  Otherwise
        the float should use the specified height.

        Also fix row 14 of the Acid2 test.  Although ambiguous, just modify the table cell baseline alignment code to align
        to the bottom of the cell's content height if no suitable baseline could be found.
        
        * khtml/rendering/render_block.cpp:
        (khtml::RenderBlock::layoutBlock):
        * khtml/rendering/render_block.h:
        (khtml::RenderBlock::firstRootBox):
        (khtml::RenderBlock::lastRootBox):
        * khtml/rendering/render_table.cpp:
        (RenderTableSection::calcRowHeight):
        (RenderTableCell::baselinePosition):
        
2005-04-15  David Hyatt  <hyatt@apple.com>

        Make sure empty tables honor CSS-specified heights when they have no rows or sections.  This is done only
        in strict mode, since it is not compatible with WinIE.

        * khtml/rendering/render_table.cpp:
        (RenderTable::layout):

2005-04-15  David Hyatt  <hyatt@apple.com>

        Fix for row 13 of the Acid2 test.  Change HTML comment parsing in strict mode to do proper SGML parsing,
        checking for pairs of -- and only being willing to close the comment if every -- is paired up.
        
        * khtml/html/htmltokenizer.cpp:
        (khtml::HTMLTokenizer::parseComment):

2005-04-12  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Richard.

        - use custom single-threaded malloc for all non-GC JavaScriptCore
        allocations, for a 9.1% speedup on JavaScript iBench
         
        * khtml/ecma/kjs_binding.cpp:
        (UString::UString):
        * khtml/ecma/kjs_proxy.cpp:
        (KJSProxyImpl::evaluate):

2005-04-15  David Hyatt  <hyatt@apple.com>

        Fix the six pixel gap between rows nine and ten of the Acid2 test.  Make sure that percentage heights that
        resolve to auto are properly treated as though they have auto height by the self-collapsing block check (as per
        section 8.3.1, paragraph 7 of the CSS2.1 spec).
        
        * khtml/rendering/render_block.cpp:
        (khtml::RenderBlock::isSelfCollapsingBlock):

2005-04-15  David Hyatt  <hyatt@apple.com>

        The Acid2 test and the reference rendering both make use of overflow:hidden on the <html> element.  Turns out
        the CSS2.1 wording for this behavior has been revised (based off WinIE/Mozilla behavior).  Change our behavior
        to match and make <html> overflow apply to the viewport.
        
        * khtml/khtmlview.cpp:
        (KHTMLView::applyOverflowToViewport):
        (KHTMLView::layout):
        * khtml/khtmlview.h:
        * khtml/rendering/render_box.cpp:
        (RenderBox::setStyle):

2005-04-14  David Hyatt  <hyatt@apple.com>

        3258403 and 3258402 can now be fixed.  min/max-width/height support is now complete.  This patch makes them
        work for positioned elements and enables us to pass row one of the Acid2 test.
        
        * khtml/rendering/render_box.cpp:
        (RenderBox::calcAbsoluteHorizontal):
        (RenderBox::calcAbsoluteHorizontalValues):
        (RenderBox::calcAbsoluteVertical):
        (RenderBox::calcAbsoluteVerticalValues):
        * khtml/rendering/render_box.h:

2005-04-12  David Hyatt  <hyatt@apple.com>

        Working on the Acid2 test, Row 1.

        Improve checkChild for the DTD so that it knows what mode a document is in.  This allows it to adhere more
        strictly to the actual DTD in strict mode and almost strict mode.

        Change the <table>-inside-<p> check so that <table> is disallowed inside <p> in 
        strict mode and almost strict mode.  This matches Firefox behavior, which allows <table>
        inside <p> only in quirks mode.

        * khtml/html/dtd.cpp:
        (DOM::checkChild):
        * khtml/html/dtd.h:
        * khtml/html/htmlparser.cpp:
        (KHTMLParser::insertNode):
        * khtml/html/htmltokenizer.cpp:
        (khtml::HTMLTokenizer::parseTag):
        * khtml/xml/dom_elementimpl.cpp:
        (ElementImpl::childAllowed):

2005-04-12  Vicki Murley  <vicki@apple.com>

        Reviewed by Maciej. 

        - fixed <rdar://problem/3760895> Request for including an implementation of the elementFromPoint function
 
        * khtml/dom/dom_doc.cpp:
        (DOM::Document::elementFromPoint):
        * khtml/dom/dom_doc.h:
        * khtml/ecma/kjs_dom.cpp:
        (DOMDocumentProtoFunc::tryCall):
        * khtml/ecma/kjs_dom.h:
        (KJS::DOMDocument::):
        * khtml/ecma/kjs_dom.lut.h:
        (KJS::):
        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::elementFromPoint):
        * khtml/xml/dom_docimpl.h:

2005-04-12  David Hyatt  <hyatt@apple.com>

        Beginning of work to support the Acid2 CSS test put forward by the Web Standards Project.  Fix
        our handling of the rel attribute on <link> elements to do a proper tokenization so that stylesheets
        can be recognized even when other keywords are present in the rel attribute.

        * khtml/html/html_headimpl.cpp:
        (HTMLLinkElementImpl::HTMLLinkElementImpl):
        (HTMLLinkElementImpl::parseHTMLAttribute):
        (HTMLLinkElementImpl::tokenizeRelAttribute):
        (HTMLLinkElementImpl::process):
        * khtml/html/html_headimpl.h:

2005-04-12  John Sullivan  <sullivan@apple.com>

        - fixed these two bugs (I also fixed these on the experimental-ui-branch)
        <rdar://problem/3154293> Find Next should not scroll page if the next target is already visible
        <rdar://problem/3121828> scrollToVisible on find cuts off the left part of the view due to needless horiz. scroll

        Reviewed by Chris.

        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::jumpToSelection):
        Trey had written code to address these issues, but left it commented out due to other
        problems. The other problems no longer occur, so I uncommented Trey's code, and then
        discovered that I could make it behave more like TextEdit with many fewer lines of
        code.

2005-04-08  David Harrison  <harrison@apple.com>

        Reviewed by Dave Hyatt.

        <rdar://problem/4084106> Remove NSAccessibilityForegroundColorTextAttributeWrapper

        * kwq/KWQAccObject.mm:
        (AXAttributeStringSetStyle):
        Use NSAccessibilityForegroundColorTextAttribute directly.

2005-04-05  David Hyatt  <hyatt@apple.com>

        Fix for 4077106, make sure that mouse wheeling in overflow sections uses 40 rather than 10 as the base line
        step.
        
        Reviewed by darin

        * khtml/rendering/render_layer.cpp:
        (RenderLayer::positionScrollbars):
        * kwq/KWQScrollBar.mm:
        (QScrollBar::scroll):

2005-04-04  Vicki Murley  <vicki@apple.com>

        Reviewed by Maciej.

        - fixed <rdar://problem/3871669> no focus or blur methods on HTML button elements

        * khtml/dom/html_form.cpp:
        (HTMLButtonElement::focus): 
        (HTMLButtonElement::blur): 
        * khtml/dom/html_form.h: 
        * khtml/ecma/kjs_html.cpp:
        (KJS::HTMLElementFunction::tryCall):
        * khtml/ecma/kjs_html.h:
        (KJS::HTMLElement::):
        * khtml/ecma/kjs_html.lut.h:
        (KJS::):
        * khtml/html/html_formimpl.cpp:
        (DOM::HTMLButtonElementImpl::blur):
        (DOM::HTMLButtonElementImpl::focus):
        * khtml/html/html_formimpl.h:

=== WebCore-415 ===

2005-03-28  David Harrison  <harrison@apple.com>

        Reviewed by Darin.
        
        <rdar://problem/4069161> REGRESSION (8A416-8A419): Safari crash bringing up context menu for non-HTML content in a frame

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject rendererForView:]):
        Nil-check node variable instead of rechecking document variable.

=== WebCore-413 ===

2005-03-27  Darin Adler  <darin@apple.com>

        Reviewed by me, fix by Kida-san.

        - fixed <rdar://problem/4067474> 8A424: Safari immediately quit by Cmd+Ctrll+'D'

        * kwq/WebCoreBridge.mm: (-[WebCoreBridge convertToNSRange:DOM::]):
        Added nil check.

=== Safari-412 ===

2005-03-24  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4052683> After adding/removing stocks from Stocks Widget, stock areas went blank

        The request was being collected before firing it's load handler.
        We now gc protect the request while it's loading.

        Reviewed by Maciej.

        * khtml/ecma/xmlhttprequest.cpp:
        (KJS::XMLHttpRequest::send):
        (KJS::XMLHttpRequest::abort):
        (KJS::XMLHttpRequest::slotFinished):
        (KJS::XMLHttpRequestProtoFunc::tryCall):

=== Safari-411 ===

2005-03-23  Darin Adler   <darin@apple.com>

        Further fix for 4053515.

        Covered cases where text position doesn't lie inside a
        text node.      

        Reviewed by Richard.

        * khtml/editing/visible_text.cpp:
        (khtml::TextIterator::setRangeFromLocationAndLength):

2005-03-23  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4053515> REGRESSION (Mail): Kotoeri input method reconversion does not work in WebViews

        We now use actual document NSRanges to represent both marked text
        ranges and selection ranges.

        Reviewed by Ken.

        * khtml/editing/visible_text.cpp:
        (khtml::TextIterator::rangeLength):
        (khtml::TextIterator::setRangeFromLocationAndLength):
        * khtml/editing/visible_text.h:
        * kwq/WebCoreBridge.h:
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge convertToNSRange:DOM::]):
        (-[WebCoreBridge DOM::convertToDOMRange:]):
        (-[WebCoreBridge selectNSRange:]):
        (-[WebCoreBridge markedTextDOMRange]):
        (-[WebCoreBridge markedTextNSRange]):

2005-03-22  Kevin Decker  <kdecker@apple.com>

        Reviewed by Vicki.

        Fixed <rdar://problem/4062336> REGRESSION (406-407): HTML submenus not working at hrweb.apple.com after going back

        Rolled out the fix for <rdar://problem/4041374> REGRESSION (185-186): unload handlers (at least those added with addEventListener) are broken

        We clearly need a better solution to 4041374. We can't indiscriminately remove event listeners in closeURL() after-all. Since event listeners are registered in a <script> tag, which is evaluated and executed at page load time, this becomes a problem since we don't reevaluate a page's <script> that is in the back/forward cache.  Thus once you leave the page, the listeners are gone for good.  This is the problem.

        * khtml/khtml_part.cpp:
        (KHTMLPart::closeURL):
        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::detach): 

=== Safari-410 ===

2005-03-22  Vicki Murley  <vicki@apple.com>

        - roll the fix for <rdar://problem/4060266> back in, since its 
        now approved by CCC

        * khtml/editing/visible_text.cpp:
        (khtml::TextIterator::handleTextBox):

2005-03-22  Vicki Murley  <vicki@apple.com>

        - roll out the fix for <rdar://problem/4060266> since it was denied by CCC

        * khtml/editing/visible_text.cpp:
        (khtml::TextIterator::handleTextBox):

2005-03-22  David Harrison  <harrison@apple.com>

        Reviewed by Darin.
        
        <rdar://problem/4060266> Double-clicking in Dictionary.app doesn't work for some words (coming just after style change)

        * khtml/editing/visible_text.cpp:
        (khtml::TextIterator::handleTextBox):
        Complete the check of whether to emit space for collapsed space.

2005-03-22  David Harrison  <harrison@apple.com>

        Reviewed by John.
        
        <rdar://problem/4061443> REGRESSION (8A420-8A421): Pasting in the Stickies widget is broken again

        * khtml/editing/htmlediting.cpp:
        (khtml::positionBeforeContainingSpecialElement):
        (khtml::positionAfterContainingSpecialElement):
        Return unchanged Position rather than a null or non-editable one.

=== Safari-409 ===

2005-03-20  Ken Kocienda  <kocienda@apple.com>

        Reviewed by me

        I made an error in this test earlier. It was not testing what I intended. Fixed.

        * layout-tests/editing/unsupported-content/list-delete-001-expected.txt
        * layout-tests/editing/unsupported-content/list-delete-001.html
        
        New tests:

        * layout-tests/editing/unsupported-content/table-delete-001-expected.txt: Added.
        * layout-tests/editing/unsupported-content/table-delete-001.html: Added.
        * layout-tests/editing/unsupported-content/table-delete-002-expected.txt: Added.
        * layout-tests/editing/unsupported-content/table-delete-002.html: Added.
        * layout-tests/editing/unsupported-content/table-delete-003-expected.txt: Added.
        * layout-tests/editing/unsupported-content/table-delete-003.html: Added.
        * layout-tests/editing/unsupported-content/table-type-after-expected.txt: Added.
        * layout-tests/editing/unsupported-content/table-type-after.html: Added.
        * layout-tests/editing/unsupported-content/table-type-before-expected.txt: Added.
        * layout-tests/editing/unsupported-content/table-type-before.html: Added.

2005-03-20  Darin Adler  <darin@apple.com>

        Reviewed by Maciej.

        - fixed <rdar://problem/3923903> REGRESSION (164-165): Repro Safari crash in khtml::RenderLayer::scrollToOffset

        * khtml/rendering/render_layer.cpp: (RenderLayer::scrollToOffset): Check canvas for nil.

2005-03-20  David Harrison  <harrison@apple.com>

        Reviewed by Darin.
        
        <rdar://problem/4055127> Dictionary pop-up panel misplaced at beginning of text blocks (breaks double-clicking in Dictionary.app)

        SimplifiedBackwardsTextIterator::advance() needed to not limit to textnodes
        when checking whether moving back across block boundaries

        VisibleUnits previousBoundary() needed to INIT_DOWN when creating result VisiblePosition

        All editing tests pass.

        * khtml/editing/visible_text.cpp:
        (khtml::SimplifiedBackwardsTextIterator::advance):
        * khtml/editing/visible_units.cpp:
        (khtml::previousBoundary):

2005-03-20  Darin Adler  <darin@apple.com>

        Reviewed by Harrison.

        - fixed <rdar://problem/4059914> when you select all of a frame's content, need to select the frame in the parent document so it can be easily deleted

        * khtml/khtml_part.h: Added selectFrameElementInParentIfFullySelected.
        * khtml/khtml_part.cpp:
        (isFrame): Added.
        (KHTMLPart::setFocusNodeIfNeeded): Changed to not set focus to a frame; was not what this function was
        intended to do, and caused trouble when trying to select a frame element.
        (KHTMLPart::khtmlMouseReleaseEvent): Call selectFrameElementInParentIfFullySelected.
        (KHTMLPart::selectAll): Call selectFrameElementInParentIfFullySelected.
        (KHTMLPart::selectFrameElementInParentIfFullySelected): Added. Selects the frame element in the parent
        if a frame is entirely selected, which makes it easier to delete or replace the frame and is consistent
        with the changes Maciej made recently for other elements.

        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge alterCurrentSelection:direction:granularity:]): Call selectFrameElementInParentIfFullySelected.
        (-[WebCoreBridge alterCurrentSelection:verticalDistance:]): Call selectFrameElementInParentIfFullySelected.

2005-03-20  Darin Adler  <darin@apple.com>

        Reviewed by me, code change by Ken.

        - fixed <rdar://problem/4059852> Deleting from first element of list makes content jump to wrong place

        * khtml/editing/htmlediting.cpp:
        (khtml::isListStructureNode): Added.
        (khtml::DeleteSelectionCommand::moveNodesAfterNode): Check for list nodes as well as table nodes.

2005-03-20  Ken Kocienda  <kocienda@apple.com>

        Reviewed by me
        
        Added tests to cover new "unsupported content" editing code.

        * layout-tests/editing/unsupported-content/list-delete-001-expected.txt: Added.
        * layout-tests/editing/unsupported-content/list-delete-001.html: Added.
        * layout-tests/editing/unsupported-content/list-delete-002-expected.txt: Added.
        * layout-tests/editing/unsupported-content/list-delete-002.html: Added.
        * layout-tests/editing/unsupported-content/list-delete-003-expected.txt: Added.
        * layout-tests/editing/unsupported-content/list-delete-003.html: Added.
        * layout-tests/editing/unsupported-content/list-type-after-expected.txt: Added.
        * layout-tests/editing/unsupported-content/list-type-after.html: Added.
        * layout-tests/editing/unsupported-content/list-type-before-expected.txt: Added.
        * layout-tests/editing/unsupported-content/list-type-before.html: Added.

2005-03-20  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Maciej
        
        Fix for this bug:
        
        <rdar://problem/4059578> Entire list deleted, and caret disappears, when delete key hit at end of list

        The problem is that a new case in the delete code did not consider when the
        downstream end node of the selection might be an ancestor of the upstream start
        node. That is the case in this bug. The downstream end is the body element, and
        this line of code would delete all the children of the downstream end:
            removeChildrenInRangePreservingPosition(m_downstreamEnd.node(), 0, 
                m_downstreamEnd.offset(), m_upstreamStart);

        The fix is to check for this "is ancestor" case, and do some tree logic to find
        the right offset of the downstream end node for the call to
        removeChildrenInRangePreservingPosition().

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::handleGeneralDelete): Fixed as described.

2005-03-19  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Maciej

        Fix for this bug:
       
        <rdar://problem/4059384> Cannot place insertion point correctly in editable text that avoids floating elements

        Note: I strongly suspect this bug blocks a complete solution to this other Tiger/P2:
        <rdar://problem/4055748> AX: Dictionary pop-up panel shows at wrong place on specific parts of particular pages

        * khtml/rendering/render_text.cpp:
        (RenderText::caretRect): Change the y-coordinate used to calculate the available width for a line. Height is wrong.
        Top of the box containing the text where the click is done is correct. Also, add in the x-offset for the start
        of the text box when calculating the available width. If this text box is avoiding a float at the y-coordinate
        for the relevant box, failure to add in the amount of float-avoidance will make the text at the coordinates
        greater than end-of-line minus float-avoidance ineligible for caret placement.

2005-03-19  Darin Adler  <darin@apple.com>

        Reviewed by Ken.

        - fixed <rdar://problem/4057594> REGRESSION (125-406): Unrepro crash in HTMLTokenizer::allDataProcessed after hitting Back button

        * khtml/html/htmltokenizer.cpp: (khtml::HTMLTokenizer::allDataProcessed):
        To get the part safely after calling end, save a guarded pointer to the view.
        The old way could end trying to call a virtual function a part that was destroyed.

2005-03-19  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Darin.

        <rdar://problem/4053506> Pasting Tables and Cells in Mail does not allow editing before or after
        <rdar://problem/4005954> REGRESSION (Mail): After copy/paste of content containing list element cannot go back to entering text at left side of page
        
        * khtml/editing/htmlediting.cpp:
        (khtml::maxDeepOffset):
        (khtml::CompositeEditCommand::removeFullySelectedNodePreservingPosition):
        (khtml::CompositeEditCommand::removeChildrenInRangePreservingPosition):
        (khtml::CompositeEditCommand::removeNodePreservingPosition):
        (khtml::CompositeEditCommand::insertBlockPlaceholder):
        (khtml::CompositeEditCommand::appendBlockPlaceholder):
        (khtml::CompositeEditCommand::forceBlockPlaceholder):
        (khtml::CompositeEditCommand::addBlockPlaceholderIfNeeded):
        (khtml::isSpecialElement):
        (khtml::isFirstVisiblePositionInSpecialElementInFragment):
        (khtml::positionBeforePossibleContainingSpecialElement):
        (khtml::positionAfterPossibleContainingSpecialElement):
        (khtml::ApplyStyleCommand::applyInlineStyle):
        (khtml::DeleteSelectionCommand::initializePositionData):
        (khtml::DeleteSelectionCommand::insertPlaceholderForAncestorBlockContent):
        (khtml::DeleteSelectionCommand::handleGeneralDelete):
        (khtml::DeleteSelectionCommand::calculateTypingStyleAfterDelete):
        (khtml::DeleteSelectionCommand::doApply):
        (khtml::InsertParagraphSeparatorCommand::doApply):
        (khtml::ReplacementFragment::ReplacementFragment):
        (khtml::ReplaceSelectionCommand::doApply):
        * khtml/editing/htmlediting.h:
        * khtml/editing/visible_position.cpp:
        (khtml::isRenderedBR):
        (khtml::VisiblePosition::initDownstream):
        (khtml::isLastVisiblePositionInBlock):
        * khtml/rendering/render_line.cpp:
        (khtml::RootInlineBox::closestLeafChildForXPos):
        * khtml/xml/dom_nodeimpl.cpp:
        (NodeImpl::isBlockFlowOrTable):
        (NodeImpl::isEditableBlock):
        (NodeImpl::enclosingBlockFlowOrTableElement):
        * khtml/xml/dom_nodeimpl.h:
        * khtml/xml/dom_position.cpp:
        (DOM::Position::upstream):
        (DOM::Position::downstream):
        * layout-tests/editing/deleting/delete-at-paragraph-boundaries-003-expected.txt:
        * layout-tests/editing/deleting/delete-at-paragraph-boundaries-004-expected.txt:
        * layout-tests/editing/deleting/delete-select-all-001-expected.txt:
        * layout-tests/editing/deleting/delete-select-all-003-expected.txt:
        * layout-tests/editing/inserting/insert-3786362-fix-expected.txt:

2005-03-19  John Sullivan  <sullivan@apple.com>

        Reviewed by Darin.
        
        - fixed <rdar://problem/4058740> Crash (nil-deref) editing Mail reply 
        message in KWQKHTMLPart::fontForSelection (MailViewer-723)

        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::fontForSelection):
        Add nil check to loop. It shouldn't be necessary, but this crash trace seems to be
        running into it. We're not completely certain, but the check is harmless at worst.

=== Safari-408 ===

2005-03-18  David Harrison  <harrison@apple.com>

        Reviewed by Darin.
        
        <rdar://problem/3584942> AX: Safari Accessibility parent-child mismatch
        
        Also changed WebKit.

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject rendererForView:]):
        New to cover both the WebCore and WebKit NSViews.
        
        (-[KWQAccObject _accessibilityParentForSubview:]):
        Use rendererForView.
        
        * kwq/WebCoreFrameView.h:
        Add WebCoreBridgeHolder protocol to get access to WebKit NSViews.

2005-03-18  David Harrison  <harrison@apple.com>

        Reviewed by Darin, Ken.

        <rdar://problem/3735625> AX: add AXPress action if an element has an onclick handler

        * khtml/dom/html_form.cpp:
        (HTMLInputElement::click):
        * khtml/html/html_elementimpl.cpp:
        (HTMLElementImpl::click):
        (HTMLElementImpl::accessKeyAction):
        * khtml/html/html_elementimpl.h:
        * khtml/html/html_formimpl.cpp:
        (DOM::HTMLFormElementImpl::submitClick):
        (DOM::HTMLButtonElementImpl::click):
        (DOM::HTMLButtonElementImpl::accessKeyAction):
        (DOM::HTMLInputElementImpl::click):
        (DOM::HTMLInputElementImpl::accessKeyAction):
        (DOM::HTMLInputElementImpl::defaultEventHandler):
        (DOM::HTMLLabelElementImpl::accessKeyAction):
        (DOM::HTMLSelectElementImpl::accessKeyAction):
        (DOM::HTMLTextAreaElementImpl::accessKeyAction):
        * khtml/html/html_formimpl.h:
        * khtml/html/html_inlineimpl.cpp:
        (HTMLAnchorElementImpl::defaultEventHandler):
        (HTMLAnchorElementImpl::accessKeyAction):
        * khtml/html/html_inlineimpl.h:
        * khtml/rendering/render_form.cpp:
        (RenderFileButton::click):
        * khtml/rendering/render_form.h:
        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::defaultEventHandler):
        * khtml/xml/dom_elementimpl.h:
        (DOM::ElementImpl::accessKeyAction):
        * kwq/DOMHTML.mm:
        (-[DOMHTMLInputElement click]):
        * kwq/KWQButton.h:
        * kwq/KWQButton.mm:
        (QButton::click):
        * kwq/KWQFileButton.h:
        * kwq/KWQFileButton.mm:
        (KWQFileButton::click):
        
        Add accessKeyAction parameter about whether to limit to HTMLElementImpl subclasses that JavaScript wants, or to apply to others as well.
        
        Add click() parameter about whether to send the mousedown and mouseup events in addition to the click event.
        
        * kwq/KWQAccObject.mm:
        (-[KWQAccObject mouseButtonListener]):
        Locate a mousedown, mouseup, or click handler in the current element and its ancestors.
        
        (-[KWQAccObject actionElement]):
        (-[KWQAccObject accessibilityIsIgnored]):
        (-[KWQAccObject accessibilityPerformAction:]):
        Consider mouseButtonListener.

2005-03-18  John Sullivan  <sullivan@apple.com>

        Reviewed by Darin.
        
        - fixed <rdar://problem/4002164> maps that include start and end 
        location don't print right from maps.google.com

        I thought I checked this in yesterday but a ChangeLog conflict aborted my checkin
        without me noticing.

        * khtml/rendering/render_style.cpp:
        (RenderStyle::diff):
        flag name changed from _should_correct_text_color to _force_backgrounds_to_white
        
        * khtml/rendering/render_style.h:
        (khtml::RenderStyle::InheritedFlags::operator==):
        flag name changed from _should_correct_text_color to _force_backgrounds_to_white
        (khtml::RenderStyle::setBitDefaults):
        ditto
        (khtml::RenderStyle::forceBackgroundsToWhite):
        ditto, and method name changed too
        (khtml::RenderStyle::setForceBackgroundsToWhite):
        ditto
        
        * khtml/rendering/render_text.cpp:
        (InlineTextBox::paint):
        updated for name change; also, compare text against white instead of current bg color
        because we no longer actually modify the bg color (previously we would always set
        the bg color to white, so the result is the same)
        
        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::recalcStyle):
        updated for name change
        
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge styleSheetForPrinting]):
        removed this method
        (-[WebCoreBridge reapplyStylesForDeviceType:]):
        removed the code that called styleSheetForPrinting; we no longer use a stylesheet
        for this behavior.

        * khtml/rendering/render_box.cpp:
        (RenderBox::paintBackgroundExtended):
        If forceBackgroundsToWhite flag is set, convert background images and
        background colors to white background color with no background image.
        
2005-03-18  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4056718> Pasting quotes the entire message

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply): After pasting, nodes are moved to the block containing
        the end of the pasted content in certain cases. This move logic used to stop once it moved all the
        siblings of the node following the last node of the pasted-in content. This means that block elements
        could get moved, and if the pasted-in content included a mail blockquote, this could result in
        one quote level getting added. The fix is to stop the move of nodes once a <br>, block element, or
        <table> is seen. This only affected one of the many test cases we have for such scenarios, and 
        the change to that result makes sense given the code change.
        * layout-tests/editing/pasteboard/paste-text-003-expected.txt: This test result changed in a way that
        adequately tests the behavior change, so I did not add a new test.

2005-03-17  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Harrison

        Fix for this bug:
        
        <rdar://problem/4051809> 8A413: Cursor-up in a mail message sometimes gets stuck (with specific reproducible case)

        * khtml/rendering/render_text.cpp:
        (RenderText::positionForCoordinates): Consider two lines: line-above and line-below. If the caret position in line-below
        was at an x coordinate between half way through the x coordinate of the last character on the line-above and the
        end of that same character, this bug would happen since the positioning code would assume that it could create a 
        VisiblePosition with a DOWNSTREAM affinity. Now, I check to see if the character position on the line-above is the 
        last character on that line, and if it is, I use UPSTREAM as the affinity.

=== Safari-407 ===

2005-03-17  David Harrison  <harrison@apple.com>

        Reviewed by Darin, Ken.

        * khtml/editing/htmlediting.cpp:
        (khtml::EditCommandPtr::setEndingSelection):
        Fixed typo so that it calls setEndingSelection rather than setStartingSelection.
        Commented out this unused method, tho, since this is late in Tiger.  Proved unused by successful build after temporarily removing method declaration or implemenation.

2005-03-16  Kevin Decker  <kdecker@apple.com>

        Reviewed by mjs.

        Fixed <rdar://problem/4046665> REGRESSION (403-405): mypage.apple.com login does not work (hits assertion in Development build)

        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge canLoadURL:fromReferrer:hideReferrer:]): Check always came back false when callers would send a nil NSURL to this method. Now we allow the empty url cases, eg., <frame src="">

2005-03-16  Darin Adler  <darin@apple.com>

        Reviewed by Maciej.

        - fixed <rdar://problem/4045203> REGRESSION (125-188): Redundant JS imports crash Safari

        * khtml/html/htmltokenizer.h: Take inWRite bool out of NDEBUG ifdef.
        * khtml/html/htmltokenizer.cpp:
        (khtml::HTMLTokenizer::HTMLTokenizer): Take inWrite bool management code out of NDEBUG ifdef.
        (khtml::HTMLTokenizer::write): Ditto. Don't call end if inWrite is true, since it will be called
        when you return to the body of the outer write() call.
        (khtml::HTMLTokenizer::allDataProcessed): Don't call end() if inWrite is true for the same reason.
        (khtml::HTMLTokenizer::finish): Ditto.

        * kwq/KWQWidget.mm: (QWidget::getOuterView): Removed bogus assertion that has been vexing us of late.

2005-03-16  David Harrison  <harrison@apple.com>

        Reviewed by me (written by Patti Yeh).

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject doAXNextWordEndTextMarkerForTextMarker:]):
        Use LeftWordIfOnBoundary instead of RightWordIfOnBoundary.

        (-[KWQAccObject doAXPreviousWordStartTextMarkerForTextMarker:]):
        Use RightWordIfOnBoundary instead of LeftWordIfOnBoundary.

2005-03-16  David Harrison  <harrison@apple.com>

        Reviewed by Maciej.

        <rdar://problem/4054590> AX: Dictionary panel does not work when page is scrolled on Safari

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject doAXTextMarkerForPosition:]):
        Add in the view's contentsX and contentsY to the point.

2005-03-16  David Harrison  <harrison@apple.com>

        Reviewed by Maciej.

        <rdar://problem/4048506> Deleting from beginning of editable div deletes other document elements
        
        Also changed WebKit.
        
        * khtml/editing/visible_units.h:
        * khtml/editing/visible_units.cpp:
        (khtml::startOfEditableContent):
        (khtml::endOfEditableContent):
        (khtml::inSameEditableContent):
        (khtml::isStartOfEditableContent):
        (khtml::isEndOfEditableContent):
        New.
        
        * kwq/WebCoreBridge.h:
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge canDeleteRange:]):
        New.

2005-03-16  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Darin

        Fix for this bug:
        
        <rdar://problem/4042935> undo doesn't work properly during inline input

        * kwq/WebCoreBridge.h: Declare new method below.
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge replaceMarkedTextWithText:]): New method. Wraps calls to TypingCommand::deleteKeyPressed and
        TypingCommand::insertText to map the way that international text input works onto the typing undo system.

2005-03-16  David Harrison  <harrison@apple.com>

        Reviewed by Darin.

        <rdar://problem/4044336> REGRESSION (8A398-8A409): Option-Delete also deletes space to left of deleted word

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::initializePositionData):
        - skip smart delete if the selection to delete already starts or ends with whitespace
        
        * khtml/khtml_part.cpp:
        (KHTMLPart::handleMousePressEventDoubleClick):
        - preserve selection on double-click when range is already selected
        
        * khtml/xml/dom_position.cpp:
        (DOM::Position::leadingWhitespacePosition):
        (DOM::Position::trailingWhitespacePosition):
        - fix considerNonCollapsibleWhitespace action (logic was reversed)
        - add non-breaking space to the non-collapsable ones
        
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge rangeByExpandingSelectionWithGranularity:]):
        (-[WebCoreBridge rangeByAlteringCurrentSelection:direction:granularity:]):
        - these methods do not set the selection, so remove calls to setSelectionGranularity
          
        (-[WebCoreBridge alterCurrentSelection:direction:granularity:]):
        (-[WebCoreBridge alterCurrentSelection:verticalDistance:]):
        - set the granularity back to character
        - the one exception is that we need to keep word granularity
          to preserve smart delete behavior when extending by word

2005-03-15  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by John.

        <rdar://problem/4053266> Pressing return a few times right after a link makes the new blank lines part of the link
        
        * khtml/editing/htmlediting.cpp:
        (khtml::InsertLineBreakCommand::doApply): Use
        positionOutsideContainingSpecialElement in the right two places.
        (khtml::InsertParagraphSeparatorCommand::doApply): Ditto.

        - move all these helper functions higher in the file
        
        * khtml/editing/htmlediting.cpp:
        (khtml::isSpecialElement):
        (khtml::isFirstVisiblePositionInSpecialElement):
        (khtml::positionBeforeNode):
        (khtml::positionBeforeContainingSpecialElement):
        (khtml::maxRangeOffset):
        (khtml::isLastVisiblePositionInSpecialElement):
        (khtml::positionAfterNode):
        (khtml::positionAfterContainingSpecialElement):
        (khtml::positionOutsideContainingSpecialElement):

2005-03-14  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Ken.

        <rdar://problem/4049925> Pasting right after a link makes pasted content part of the link (without visible style change)
        
        * khtml/editing/htmlediting.cpp:
        (khtml::positionOutsideContainingSpecialElement): made a helper
        function that computes a position outside the outermost containing
        special element if the passed in position is right at the start or
        end of it
        (khtml::InsertTextCommand::prepareForTextInsertion): use new helper here
        (khtml::ReplaceSelectionCommand::doApply): use it here too: this is the fix
        (khtml::positionBeforeNode): made static
        (khtml::positionBeforeContainingSpecialElement): made static
        (khtml::positionAfterNode): made static
        (khtml::positionAfterContainingSpecialElement): made static

2005-03-15  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4053658> Crash getting direction at maps.google.com

        Add non nil style() check.

        Reviewed by Dave Harrison.

        * khtml/rendering/render_table.cpp:
        (RenderTableCell::collapsedRightBorder):

2005-03-15  Kevin Decker  <kdecker@apple.com>

        Reviewed by John
        
        Fixed: <rdar://problem/4041374> REGRESSION (185-186): unload handlers (at least those added with addEventListener) are broken

        The reason why UNLOAD_EVENT wouldn't dispatch was because the code would delete all event listeners at the detach() phase which is prior to closeURL(). 

        This fixes a recent regression from:

 <rdar://problem/3977973> pages on ebay leak referenced JavaScript objects -- over time browsing becomes super-slow

        * khtml/khtml_part.cpp:
        (KHTMLPart::closeURL): After dispatching event handlers, go ahead and remove them from the DOM. 
        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::detach): Took out the call to removeAllEventListenersFromAllNodes(). If we remove all event listeners here then when KHTMLPart::closeURL() checks for even listeners, it will never have any because they'll already be gone. 

2005-03-15  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Vicki

        Fox for this bug:
        
        <rdar://problem/4052642> Each delete keystroke is in its own undo group; not included in undo group with other typing

        Calling -[WebCore setSelectedDOMRange:range affinity:] had the result of "closing" any active set of typing
        keystrokes grouped together in a single undo operation. A change on 27 Jan in WebKit to change the way delete
        keystrokes are handled made this feature regress. Previous to that change, the backwards delete
        key went through separate code that is no longer in the tree that did not set the selection in the way
        it is done now.
        
        The solution is to add an extra argument to the set-selection call. The WebCoreBridge now offers this method:
        -[WebCore setSelectedDOMRange:range affinity:closeTyping:]. Now, callers must indicate whether setting the 
        selection will act to close typing or not.

        * kwq/WebCoreBridge.h: Changed header accordingly to change method shown below.
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge setSelectedDOMRange:affinity:closeTyping:]): Added closeTyping argument to this method.

2005-03-15  John Sullivan  <sullivan@apple.com>

        Reviewed by Vicki.
        
        - fixed <rdar://problem/4052246> crash in KWQKHTMLPart::createPart() trying to display local file in frame

        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::createPart):
        Check part for nil before trying to ref. This was probably a longstanding code flaw revealed by
        the recent security fix.

2005-03-15  Kevin Decker  <kdecker@apple.com>

        Reviewed by Ken and Maciej.

        New fix for <rdar://problem/3667701> crash in KHTMLPart::jScriptEnabled()
        
        The tokenizer has buffers which mean parsing can continue even after loading is supposed to be stopped. If the loading process was aborted, the tokenizer should abort, too.

        * khtml/html/htmltokenizer.cpp:
        (khtml::HTMLTokenizer::HTMLTokenizer): Initialize loadStopped to false.
        (khtml::HTMLTokenizer::write): Go ahead and bail out if loadStopped is true. 
        (khtml::HTMLTokenizer::processToken):
        * khtml/html/htmltokenizer.h: Added loadStopped flag. Changed the view pointer from a standard pointer to a QGuardedPtr.  This fixes the crash.  Now the tokenizer's handle to the view will now automatically nil-out and never dangle.
        * khtml/khtml_part.cpp:
        (KHTMLPart::closeURL): Notify the tokenizer to stop parsing. 
        * khtml/xml/xml_tokenizer.cpp:
        (khtml::XMLTokenizer::XMLTokenizer): Initialize loadStopped to false.
        * khtml/xml/xml_tokenizer.h:
        (khtml::Tokenizer::stopParsing): Added. 

2005-03-14  David Harrison  <harrison@apple.com>

        Reviewed by Darin, Maciej.

        <rdar://problem/4046103> REGRESSION (Mail): clicking after style change sets insertion point incorrectly
        
        Also fixes crash by adding nil check.

        * khtml/editing/htmlediting.cpp:
        (khtml::MoveSelectionCommand::doApply):
        Check the node for nil.
        
        * khtml/khtml_part.cpp:
        (KHTMLPart::khtmlMouseReleaseEvent):
        Use the node from the event rather than from the selection.

2005-03-14  Darin Adler  <darin@apple.com>

        Reviewed by Harrison.

        - fixed <rdar://problem/4049776> Seed: Mail: Disable spellcheck leaves red artifacts

        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::DocumentImpl): Set markers list to be "auto-delete" so they don't all leak.
        (DocumentImpl::removeMarker): Remove markers list for a node when the last marker is removed
        for that node. Otherwise, we can have empty marker lists for each node forever until the
        document goes away.
        (DocumentImpl::removeAllMarkers): Added code to dirty the markers.
        (DocumentImpl::shiftMarkers): Remove unneeded empty check.

        * kwq/WebCoreBridge.h: Added unmarkAllMisspellings for WebKit.
        * kwq/WebCoreBridge.mm: (-[WebCoreBridge unmarkAllMisspellings]): Added. Calls removeAllMarkers.

2005-03-14  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4027928> Tiger_8A394:Acrobat crashes while tried to remove the subscription errors by clicking on "Would you like to remove the subscription" from Tracker details view pane

        A document may be deleted as a consequence of handling an event,
        as was the case with Acrobat.app.  Ensure that the document is still valid
        before passing the event on for further handling.

        * khtml/xml/dom_nodeimpl.cpp:
        (NodeImpl::dispatchUIEvent):

2005-03-14  Ken Kocienda  <kocienda@apple.com>

        Reviewed by me

        Added a couple of comments about setChanged() to this code based on my experiences with 4047028.

        * khtml/css/css_valueimpl.cpp:
        (DOM::CSSMutableStyleDeclarationImpl::addParsedProperties)
        (DOM::CSSMutableStyleDeclarationImpl::merge)

2005-03-14  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4047028> Changing quote levels on stylized text causes it to be sent as colored (Blue). Tiger8A410

        * khtml/css/css_valueimpl.cpp:
        (DOM::CSSMutableStyleDeclarationImpl::removePropertiesInSet): This function now calls setChanged() at the 
        end of its loop if any properties were removed. This makes the style system update correctly in response
        to changes made by this function. The code to paste removes style from the pasted content in a 
        preliminary step, and the fact that the style system did not update properly left unwanted color
        declarations in the document.

2005-03-14  Vicki Murley  <vicki@apple.com>

        - roll out this change for now, since it was denied by CCC

    2005-03-11  David Harrison  <harrison@apple.com>
        
        Reviewed by Darin.
        
        <rdar://problem/4046602> WebCore invokes undefined behavior when the spell checker isn't running

        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::advanceToNextMisspelling):
        (KWQKHTMLPart::markMisspellings):
        Nil check checker.

2005-03-14  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4050403> Mail crashes after pasting and deleting the content of one Excel cell

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::handleGeneralDelete): Add some null checks to the code. This fixes the crash, although
        following the steps described in the bug by John Sullivan on 3/14/05 at 10:49 AM will leave us with a "blank line" in 
        the document that cannot be removed (this is actually an empty table). This is undesirable, however, work Maciej is
        doing to fix the general-case problem of trying to edit constructs we do not handle well in editing should fix this
        particular case, making the deletion of this "blank line" possible. Maciej is doing this work as part of this bug:
        <rdar://problem/4036051> Hard to select (and thus delete) an IFRAME in an editable WebView

2005-03-14  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Darin

        Fix for this bug:
        
        <rdar://problem/4029632> Mail crashes in DOM::NodeImpl::isBlockFlow() after pasting text with alignment style and BR element from Safari
        
        The problem is that removeInlineStyle() can remove nodes, and if either the start or end node of the
        selection at the time of the call to removeInlineStyle() was in a node that got removed, bad things
        would happen. The fix is described below.

        * khtml/editing/htmlediting.cpp:
        (khtml::maxRangeOffset): Moved this static function to a different place in the file so the code below can use it.
        (khtml::ApplyStyleCommand::applyInlineStyle): Calling removeInlineStyle() now has the side effect of 
        setting the command's ending selection. Now resets start and end using the ending selection after the call to
        removeInlineStyle() as it is done elsewhere in this function.
        (khtml::ApplyStyleCommand::removeInlineStyle): Track the removal of the start or end node based on
        the positions passed in. If either the start or the end node is removed as part of style removal, 
        set an appropriate replacement start or end that is still in the document.

=== Safari-406 ===

2005-03-14  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Darin

        I need to roll out Kevin's change to fix 3667701. It breaks contextual fragments, and hence breaks
        paste in editing (among other things).

        * khtml/html/htmltokenizer.cpp: Roll out recent change.
        (khtml::HTMLTokenizer::write) 
        * khtml/html/htmltokenizer.h: Ditto.

2005-03-13  Darin Adler  <darin@apple.com>

        Reviewed by Ken and Maciej.

        - fixed <rdar://problem/4049040> REGRESSION (403-405): security check prevents user stylesheet from loading (Dictionary.app doesn't work at all!)

        * kwq/WebCoreBridge.mm: (-[WebCoreBridge canLoadURL:fromReferrer:hideReferrer:]): Changed to give "applewebdata:"
        documents the same privileges to open local files that "file:" documents have.

2005-03-13  Kevin Decker  <kdecker@apple.com>

        Reviewed by mjs.
        
        Fixed: <rdar://problem/3667701> crash in KHTMLPart::jScriptEnabled()
        
        The problem here was that the tokenizer would continue to receive chunks of data from the loader already
        after the view and part had been destroyed.  Situations like this could arise when clicking on another link 
        while still loading the current view, or during self test where we rapidly open, load, and close browser
        windows very fast. 

        * khtml/html/htmltokenizer.cpp: 
        (khtml::HTMLTokenizer::write): Simple nil check against the view. 
        * khtml/html/htmltokenizer.h: Changed the view pointer from a standard pointer to a QGuardedPtr.  The tokenizer's
        handle to the view will now automatically nil-out and never dangle.

2005-03-13  Darin Adler  <darin@apple.com>

        Reviewed by John and Ken.

        - fixed <rdar://problem/4044347> REGRESSION (Mail): Control-K in particular message moves insertion point to previous line

        Tweaked the deleting code, and added three new deleting layout tests to confirm the new code works.

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::handleSpecialCaseBRDelete): Removed special case with comment that said it was
        for the case where a "selection contains only a BR right after a block ended". This code was being triggered in
        more cases than just that one, and in all the cases I tested, the general delete code works fine.
        (khtml::DeleteSelectionCommand::handleGeneralDelete): Changed the code that decides whether to delete an entire
        block to understand the case where the end block is outside the start block, but contains the start block.
        In that case, we want to delete the entire block. Not deleting the block was causing us to delete just the <br>,
        and not the enclosing <div> in the case in the bug.

        * layout-tests/editing/deleting/delete-line-015-expected.txt: Added.
        * layout-tests/editing/deleting/delete-line-015.html: Added.
        * layout-tests/editing/deleting/delete-line-016-expected.txt: Added.
        * layout-tests/editing/deleting/delete-line-016.html: Added.
        * layout-tests/editing/deleting/delete-line-017-expected.txt: Added.
        * layout-tests/editing/deleting/delete-line-017.html: Added.
        * layout-tests/editing/style/smoosh-styles-002-expected.txt: Updated to improved results. With the code change, the deletion
        now deletes more than it used to. The old results had an empty text node and <h1> element that were both 0-sized, and now
        we delete both of those.

2005-03-13  Darin Adler  <darin@apple.com>

        - fixed <rdar://problem/4049172> REGRESSION (403-405): Gmail: text box in "Invite a friend" section overlaps other sections

        Rolled out fix for <rdar://problem/3952698> Function buttons do not display properly with Telia Webmail

        * khtml/rendering/render_replaced.cpp: (RenderReplaced::calcMinMaxWidth): Back to previous version of this file.

2005-03-12  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Adele.

        <rdar://problem/4046144> RSS pages leave a hole in local file security policy (need to revert feed: exemption)
        
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge canLoadURL:fromReferrer:hideReferrer:]): Revert
        emergency workaround for Safari RSS, now that a new Syndication
        has been submitted.

2005-03-11  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Kevin.

        <rdar://problem/4026787> text typed after a link (pasted or Mail Link to this Page) is part of the link, underlined and colored blue
        
        The concept of this change is every time you type at the very
        start or very end of a link (even if nested in further inner
        elements), the typed text goes outside the link instead of inside.
        
        * khtml/editing/htmlediting.cpp:
        (khtml::InsertTextCommand::prepareForTextInsertion): Check whether
        we are at the first visible position or last visible position of a
        special element. For now this only includes HTML A elements that
        are links (i.e. they have an href).
        (khtml::isSpecialElement): Helper function that identifies special
        elements (for now only links).
        (khtml::isFirstVisiblePositionInSpecialElement): Checks if a given DOM
        position is equivalent to the first visible position in some containing 
        editable special element.
        (khtml::positionBeforeNode): Returns the DOM position immediately
        before a node.
        (khtml::positionBeforeContainingSpecialElement): Gives a DOM
        position immediately before the outermost editable containing
        special element where the passed-in position is equivalent to the
        first visible position.
        (khtml::maxRangeOffset): Helper to get the maximum allowed
        range/position offset for a node, does the right thing based on
        whether the node would use a character offset or child offset.
        (khtml::isLastVisiblePositionInSpecialElement): Similar to above,
        but for end of node instead of start.
        (khtml::positionAfterNode): Ditto.
        (khtml::positionAfterContainingSpecialElement): Ditto.

        Some layout tests changed - I looked over all the diffs and found
        that the only changes were "junk nodes" like empty spans and text
        nodes moving from one spot in the tree to another. These changes
        are all harmless and do not affect layout or future editing.

        * layout-tests/editing/inserting/typing-003-expected.txt:
        * layout-tests/editing/style/remove-underline-across-paragraph-expected.txt:
        * layout-tests/editing/style/remove-underline-across-paragraph-in-bold-expected.txt:
        * layout-tests/editing/style/remove-underline-after-paragraph-expected.txt:
        * layout-tests/editing/style/remove-underline-after-paragraph-in-bold-expected.txt:
        * layout-tests/editing/style/remove-underline-expected.txt:
        * layout-tests/editing/style/remove-underline-from-stylesheet-expected.txt:
        * layout-tests/editing/style/remove-underline-in-bold-expected.txt:
        * layout-tests/editing/style/typing-style-003-expected.txt:
        * layout-tests/editing/style/unbold-in-bold-expected.txt:
        * layout-tests/editing/style/underline-expected.txt:

2005-03-11  Adele Amchan  <adele@apple.com>

        backing out fix for <rdar://problem/4021711> REGRESSION (125-188): blank pages when browsing forum at cooperativeresearch.org - cached external script problem

        This caused the following regressions (that we know of): 
        <rdar://problem/4047445> REGRESSION (Safari-400-403?): Some or all page contents sometimes don't appear (macworld.com)
        <rdar://problem/4046153> 8a409: Problem loading Citibank page in Safari 2 (403)
        <rdar://problem/4047801> REGRESSION (402-403): .Mac homepage links don't work

        * khtml/html/htmltokenizer.cpp:
        (khtml::HTMLTokenizer::scriptHandler):

2005-03-11  David Harrison  <harrison@apple.com>

        Reviewed by Darin.

        <rdar://problem/4046602> WebCore invokes undefined behavior when the spell checker isn't running

        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::advanceToNextMisspelling):
        (KWQKHTMLPart::markMisspellings):
        Nil check checker.

2005-03-11  Ken Kocienda  <kocienda@apple.com>

        Reviewed by me

        * ForwardingHeaders/editing/visible_units.h: Added.

2005-03-11  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4045521> Hitting return key with full line selected does not add blank line as it should

        * khtml/editing/htmlediting.cpp:
        (khtml::InsertParagraphSeparatorCommand::doApply): Removed some "special-case" code from this 
        function that would look for a selection that started and ended in a different block, and would
        then bail right after the deletion of the selection without inserting a paragraph separator.
        This was just wrong. So, the code change is removal only. When the general-case code runs instead
        of the erroneous special-case code, the bug goes away.
        
        New tests:
        
        * layout-tests/editing/inserting/return-key-with-selection-001-expected.txt: Added.
        * layout-tests/editing/inserting/return-key-with-selection-001.html: Added.
        * layout-tests/editing/inserting/return-key-with-selection-002-expected.txt: Added.
        * layout-tests/editing/inserting/return-key-with-selection-002.html: Added.
        * layout-tests/editing/inserting/return-key-with-selection-003-expected.txt: Added.
        * layout-tests/editing/inserting/return-key-with-selection-003.html: Added.

2005-03-11  David Harrison  <harrison@apple.com>

        Reviewed by Darin.

        <rdar://problem/4009446> AX: kAXTextMarkerForPositionParameterizedAttribute not working correctly (required for Dictionary pop-up)

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject accessibilityAttributeValue:]):
        Comment changes.
        
        (-[KWQAccObject doAXTextMarkerForPosition:]):
        Dig into widgets.
        
        (-[KWQAccObject doAXBoundsForTextMarkerRange:]):
        Use the selection's document instead of the top document, to accommodate frames, etc.
        
        (-[KWQAccObject accessibilityAttributeValue:forParameter:]):
        Fixed parameter processing to look for NSValue instead of AXValue.

2005-03-11  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Harrison

        Fix for this bug:
        
        <rdar://problem/3972665> 8A360: HTML message partially truncated on left hand side, text-indent from Script Editor
        
        This was fixed, then regressed with Harrison's fix for this bug:
        
        <rdar://problem/3948453> Can't type accented chars as first character in Stickies widget

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply): My now addresses both problems in a way that they no longer
        clobber each other.

2005-03-10  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for these bugs:
        
        <rdar://problem/4045511> Copying and pasting end-of-paragraph selection puts insertion point in wrong place
        <rdar://problem/4045513> Copying and pasting selection starting at end of paragraph can incorrectly remove line break

        The copy/paste code before this patch had no notion of a "logical newline" at the start of the selection. We have
        had a similar notion for "logical newline" at the end of the selection for quite some time. To fix these bugs, we
        need to introduce the same idea for selection starts.

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplacementFragment::ReplacementFragment): Process the "logical newline" at start as we write it out
        in markup. Set the bit we added to this object to signify we have such a newline.
        (khtml::ReplaceSelectionCommand::doApply): Many, many changes to introduce the new "logical newline" at start concept.
        I also tried to simply the code that sets the start position for inserting content to be pasted. I also improved a
        weakness in the smart-paste code. Now, we check before and after the paste for whether we need to add a leading or
        trailing space. The code previous to this patch only did a "before" check, with the result that we sometimes added
        a second space. In other words, the code did not realize that DOM changes done by pasting could cause formerly
        unrendered whitespace to become rendered. Also moved line placeholder clean up code to its own function.
        (khtml::ReplaceSelectionCommand::removeLinePlaceholderIfNeeded): New helper that further refines the notion
        of when we can remove a line placeholder. The definition is now, "If a line placeholder is at the visible start
        and visible end of its line, keep it; otherwise remove it".
        * khtml/editing/htmlediting.h: Declare new functions. Rework inlines in ReplacementFragment class to account for
        addition of new "logical newline" at start concept.
        (khtml::ReplacementFragment::hasInterchangeNewlineAtStart): New accessor.
        (khtml::ReplacementFragment::hasInterchangeNewlineAtEnd): Renamed from hasInterchangeNewline(), since before we
        only had a bit for the end, hence we did not need to distinguish it from the start.
        * khtml/editing/markup.cpp:
        (khtml::createMarkup): Added code to detect and write out markup for cases where we have a "logical newline" at start.
        * khtml/xml/dom2_rangeimpl.cpp:
        (DOM::RangeImpl::startPosition): New helper.
        (DOM::RangeImpl::endPosition): Ditto.
        * khtml/xml/dom2_rangeimpl.h: Declare new helpers.
        
        New tests:
        * layout-tests/editing/pasteboard/paste-line-endings-001-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-001.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-002-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-002.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-003-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-003.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-004-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-004.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-005-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-005.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-006-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-006.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-007-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-007.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-008-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-008.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-009-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-009.html: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-010-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-line-endings-010.html: Added.

=== Safari-405 ===

2005-03-10  Darin Adler  <darin@apple.com>

        Reviewed by Richard.

        - fixed <rdar://problem/4037700> Every character typed causes stat call for /usr/share/icu/icudt32b_char.brk

        * khtml/rendering/render_text.cpp: (getCharacterBreakIterator): Set boolean "got iterator" to true.

2005-03-10  Darin Adler  <darin@apple.com>

        Reviewed by Ken.

        - fixed <rdar://problem/4042867> "Bigger" changes the font size of too much text when the selection is on a style-change boundary

        * khtml/editing/htmlediting.cpp: (khtml::ApplyStyleCommand::applyRelativeFontStyleChange):
        Advance out of the starting text node if we're at the end of it.

2005-03-10  Darin Adler  <darin@apple.com>

        * khtml/html/htmltokenizer.cpp: (khtml::HTMLTokenizer::write): Fixed assertion for nested calls to write.

2005-03-10  David Harrison  <harrison@apple.com>

        Reviewed by Darin.

        <rdar://problem/4032346> REGRESSION (Mail): changing 1st line of a URL that wraps to two lines doesn't always update 2nd line

        * khtml/rendering/bidi.cpp:
        (khtml::RenderBlock::matchedEndLine):
        Look at first clean line in case line wrap implicitly dirtied it.

2005-03-10  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Vicki.

        <rdar://problem/4046018> REGRESSION (TOT): RSS pages don't display anything
        
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge canLoadURL:fromReferrer:hideReferrer:]): Exempt
        feed: and feeds: pages from the local file security check.

2005-03-10  Darin Adler  <darin@apple.com>

        Change written by Steve Peters, reviewed by me.

        - fixed <rdar://problem/4045924> improve compareBoundaryPoints to make style changes faster

        * khtml/xml/dom2_rangeimpl.cpp: (DOM::RangeImpl::compareBoundaryPoints):
        Do early outs in a couple of the loops to make things faster.

2005-03-10  Jens Alfke  <jens@apple.com>

        Reviewed by rjw.

        Fixes <rdar://problem/4040848> "REGR: Sun security-check demo applet does not load". My earlier fix for 3603191 deferred loading the applet until the </applet> tag is reached, so all <param>s will be available. Unfortunately that meant that if the </applet> tag was missing, the applet would never load at all. So instead, the loading needs to happen when the ID_APPLET tag is popped from the parser stack for any reason.

        I've restored the old NodeImpl::closeRenderer() method. Except it's now virtual, with a no-op base implementation, and is overridden by HTMLAppletElementImpl, replacing its setAllParamsAvailable method that I added a few weeks ago when fixing 3603191.

        This ensures that the applet will get loaded even if there is no explicit </applet> tag.

        The changes to htmlparser.cpp back out my earlier change and restore the lines that were deleted on 8/3/04 when the old closeRenderer was removed.

        * khtml/html/html_objectimpl.cpp:
        (HTMLAppletElementImpl::closeRenderer):
        * khtml/html/html_objectimpl.h:
        * khtml/html/htmlparser.cpp:
        (KHTMLParser::insertNode):
        (KHTMLParser::processCloseTag):
        (KHTMLParser::popOneBlock):
        * khtml/xml/dom_nodeimpl.h:
        (DOM::NodeImpl::closeRenderer):

2005-03-10  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4024929> REGRESSION (Mail): Pasting text with multiple reply levels removes one level instead of all
        
        The code to figure out which node to use to merge into an existing line did not drill into the first inline element
        as needed to make the feature work as user expect. Instead, it looked at the first node, and if it was a block, it
        skipped that node. This worked for some cases (including the important "paste-as-quotation" case) but obviously 
        doesn't work for content quoted more than once.
        
        Now, mergeStartNode() will look for the first node in pasted content that is not a block. It will now also look
        for nodes specially marked by Mail as a node added to make "Paste As Quotation" work. It won't skip those. 
        
        NOTE: This change will break Mail's "Paste As Quotation" feature for TOT WebKit users, but this is only a temporary 
        problem that will exist until we sync up with Mail's pending change to mark nodes as needed in its pasteAsQuotation:
        method.
        
        * khtml/editing/html_interchange.h: Add ApplePasteAsQuotation constant used to check for "marked" blockquotes.
        * khtml/editing/htmlediting.cpp:
        (khtml::ReplacementFragment::mergeStartNode): Look for first node that is either not a block or is marked as
        an ApplePasteAsQuotation node.
        (khtml::isMailPasteAsQuotationNode): New helper that looks for nodes marked with ApplePasteAsQuotation.
        * khtml/editing/htmlediting.h: Updated header for new function.
        
        This test result changed is an acceptable way.
        
        * layout-tests/editing/pasteboard/paste-text-013-expected.txt
        * layout-tests/editing/pasteboard/paste-text-013.html

        NOTE: This change causes a regression in this layout test:

        * layout-tests/editing/pasteboard/paste-text-013.html

        This problem will need to be fixed separately, and this bug has been filed to track this problem:
        <rdar://problem/4045513> Copying and pasting selection starting at end of paragraph can incorrectly remove line break

2005-03-09  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Richard.

        <rdar://problem/4040776> Dashboard (Weather widget) is a memory hog
        
        Change things around so the event listeners for XMLHttpRequest
        mark their JS listener objects instead of holding a hard
        reference, to avoid an unbreakable reference cycle.

        * khtml/ecma/kjs_events.cpp:
        (JSAbstractEventListener::JSAbstractEventListener):
        (JSAbstractEventListener::~JSAbstractEventListener):
        (JSAbstractEventListener::handleEvent):
        (JSAbstractEventListener::eventListenerType):
        (JSUnprotectedEventListener::JSUnprotectedEventListener):
        (JSUnprotectedEventListener::~JSUnprotectedEventListener):
        (JSUnprotectedEventListener::listenerObj):
        (JSUnprotectedEventListener::windowObj):
        (JSUnprotectedEventListener::mark):
        (JSEventListener::JSEventListener):
        (JSEventListener::~JSEventListener):
        (JSEventListener::listenerObj):
        (JSEventListener::windowObj):
        (JSLazyEventListener::JSLazyEventListener):
        * khtml/ecma/kjs_events.h:
        * khtml/ecma/kjs_html.h:
        * khtml/ecma/kjs_window.cpp:
        (Window::getJSEventListener):
        (Window::getJSUnprotectedEventListener):
        * khtml/ecma/kjs_window.h:
        * khtml/ecma/xmlhttprequest.cpp:
        (KJS::XMLHttpRequest::putValue):
        (KJS::XMLHttpRequest::mark):
        * khtml/ecma/xmlhttprequest.h:
        * khtml/khtml_part.h:

2005-03-06  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Darin.

        <rdar://problem/4005575> Arbitrary file disclosure vulnerability due to ability to load local html from remote content
        
        * khtml/ecma/kjs_html.cpp:
        (KJS::HTMLDocument::putValue):
        * khtml/ecma/kjs_window.cpp:
        (Window::put):
        (WindowFunc::tryCall):
        (Location::put):
        (LocationFunc::tryCall):
        * khtml/khtml_part.cpp:
        (KHTMLPart::begin):
        (KHTMLPart::scheduleLocationChange):
        (KHTMLPart::slotRedirect):
        (KHTMLPart::processObjectRequest):
        * khtml/khtml_part.h:
        * khtml/khtmlpart_p.h:
        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::openURLRequest):
        (KWQKHTMLPart::urlSelected):
        (KWQKHTMLPart::createPart):
        * kwq/KWQKHTMLPartBrowserExtension.mm:
        (KHTMLPartBrowserExtension::createNewWindow):
        * kwq/WebCoreBridge.h:
        * kwq/WebCoreBridge.mm:
        (hasCaseInsensitivePrefix):
        (-[WebCoreBridge didNotOpenURL:pageCache:]):
        (-[WebCoreBridge canLoadURL:fromReferrer:hideReferrer:]):

2005-03-09  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4032938> Safari: text layout for MS P Gothic font is corrupted

        Remove our hacked special case now we have our own cache of
        which fonts are fixed  pitch.

        Reviewed by Maciej.

        * kwq/KWQFont.mm:
        (QFont::isFixedPitch):

2005-03-09  David Harrison  <harrison@apple.com>

        Reviewed by Maciej.

        <rdar://problem/4037141> REGRESSION (Mail): Pasting deletes preceding blank lines with certain steps

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::insertPlaceholderForAncestorBlockContent):
        Add check for anonymous text after blocks.

2005-03-08  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for these bugs:
        
        <rdar://problem/4039661> crash in ApplyStyleCommand::applyBlockStyle pasting contents of webpage into Mail or Blot
        <rdar://problem/4039672> hang in moveParagraphContentsToNewBlockIfNecessary after pasting contents of webpage into Blot

        * khtml/editing/htmlediting.cpp:
        (khtml::CompositeEditCommand::moveParagraphContentsToNewBlockIfNecessary): Fix for 4039672. Iteration needs to do
        a traverseNextSibling(), not a traverseNextNode(). The latter might iterate into a child that already got moved, and
        the code can infinite loop as a result.
        (khtml::ApplyStyleCommand::applyBlockStyle): Fix for 4039661. This function iterates over a set of nodes to apply
        block styles. Before, the iteration would go from the start to the end of the selection, and apply block styles
        as it went. However, the act of applying style could confuse the iteration. Now I iterate and store the relevant
        nodes in QPtrList before doing any style manipulation, and then iterate that set of nodes.

2005-03-08  Darin Adler  <darin@apple.com>

        Change written by Hyatt, reviewed by me.

        - fixed <rdar://problem/3952698> Function buttons do not display properly with Telia Webmail

        * khtml/rendering/render_replaced.cpp: (RenderReplaced::calcMinMaxWidth):
        Only use a minWidth of 0 for images with a percentage value. Other replaced elements aren't scalable
        so their minWidth should be their calculated width. This has been wrong all along, but it's a relatively
        safe change because it only affects replaced elements' min-width.

2005-03-08  David Harrison  <harrison@apple.com>

        Reviewed by Ken.

        <rdar://problem/4039006> REGRESSION (Mail): Command-left-arrow goes to the start of the wrong line in particular message

        * khtml/editing/visible_units.cpp:
        (khtml::startOfLine):
        Use firstLeafChild() instead of firstChild().

2005-03-08  Richard Williamson   <rjw@apple.com>

        Fixed by Tom Madden.

        Fixed <rdar://problem/4038586> 8A402: Osaka-Mono text overlaps when typing with input method

        We weren't clearing the "all ascii" flag when setting text on
        RenderText.

        Reviewed by Richard Williamson.

        * khtml/rendering/render_text.cpp:
        (RenderText::setText):

2005-03-08  Darin Adler  <darin@apple.com>

        Reviewed by Ken and Maciej.

        - fixed <rdar://problem/3988809> REGRESSION (Mail): wide space characters are turned into plain old spaces when typed or pasted

        Changed the few places where it matters to use a "collapsible whitespace" concept instead of
        the general whitespace concept. This means treating only ' ' and '\n' specially instead of
        including other space characters, which matches what the space-collapsing logic does in RenderText,
        although it really needs to behave differently based on whitespace mode.

        * khtml/editing/visible_text.h: (khtml::isCollapsibleWhitespace): Added. This returns true only for
        ' ' and '\n' since they are the only characters collapsed once text is in the DOM. But really it
        can't do the job ignoring white-space mode, so some day it must go.

        * khtml/editing/html_interchange.cpp: (convertHTMLTextToInterchangeFormat): Changed to use
        isCollapsibleWhitespace and removed unnecessary calls to latin1().

        * khtml/editing/htmlediting.cpp:
        (khtml::isNBSP): Removed unnecessary creation/destruction of QChar each time this is called.
        (khtml::nextCharacterIsCollapsibleWhitespace): Renamed from isWS and changed to use isCollapsibleWhitespace.
        (khtml::DeleteSelectionCommand::fixupWhitespace): Use isCollapsibleWhitespace instead of isWS,
        since we only want to do our NBSP tricks for collapsible whitespace.
        (khtml::InsertTextCommand::input): Changed use of isTab to instead say == "\t" since that does the
        same thing and is arguably just as clear. Changed use of isWS to == " " since the input text can't
        include any "\n" characters, and really a plain old space is the only collapsible thing that can
        be passed in. Use isCollapsibleWhitespace instead of isWS since we want to do NBSP tricks only for
        collapsible whitespace, not all whitespace.
        (khtml::InsertTextCommand::insertSpace): Use isCollapsibleWhitespace instead of isWS since we want
        to do NBSP tricks only for collapsible whitespace, not all whitespace.
        (khtml::RebalanceWhitespaceCommand::doApply): Ditto.

        * khtml/xml/dom_position.h: Changed treatNBSPAsWhiteSpace parameters to considerNonCollapsibleWhitespace
        parameters. The most common callers are only interested in collapsible whitespace, and the smart copy
        and paste callers want to include all whitespace, including non-breaking spaces and all the Unicode spaces.
        * khtml/xml/dom_position.cpp:
        (DOM::Position::leadingWhitespacePosition): Rename the parameter, and use either QChar::isSpace or
        isCollapsibleWhitespace depending on the boolean passed in.
        (DOM::Position::trailingWhitespacePosition): Ditto.

2005-03-07  John Sullivan  <sullivan@apple.com>

        Reviewed by Darin.
        
        - <rdar://problem/4040868> REGRESSION (Mail, 403-403+): Drag and drop deletes text, 
        many other bad editing problems

        * khtml/rendering/render_text.cpp:
        (RenderText::caretMaxOffset):
        A "-" should have been a "+" in this method that was tweaked an hour ago. Editing
        was completely horked.

2005-03-07  Richard Williamson   <rjw@apple.com>

        Additional nil check from fix for 4040749.

        * khtml/rendering/render_box.cpp:
        (RenderBox::setStyle):

2005-03-07  Richard Williamson   <rjw@apple.com>

        Additional nil check from fix for 4040749.

        * khtml/rendering/render_layer.cpp:
        (RenderLayer::updateLayerPosition):

2005-03-07  David Harrison  <harrison@apple.com>

        Reviewed by Darin.

        <rdar://problem/4033202> REGRESSION (Mail): Can't arrow up at a particular spot in a particular Mail message

        Doublecheck that new position is really on a different line, because the VisiblePosition constructor does not.
        Filed <rdar://problem/4040763> for that problem.

        * khtml/editing/visible_units.cpp:
        (khtml::previousLinePosition):

2005-03-07  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4040749> REGRESSION (125-178): opacity style not working, breaks fading images on okcupid.com

        When layers are dynamically created/removed as a result of changing opacity they weren't being correctly
        sized and positioned.  This happens whenever opacity goes from < 1 to 1.

        Reviewed by Darin.

        * khtml/rendering/render_box.cpp:
        (RenderBox::setStyle):
        * khtml/rendering/render_layer.cpp:
        (RenderLayer::updateLayerPosition):

2005-03-07  Christy Warren  <kali@apple.com>

        Reviewed by Darin

        * khtml/rendering/render_text.cpp:
        (RenderText::caretMinOffset): modified to handle BIDI case by checking all text boxes for min offset
        (RenderText::caretMaxOffset): modified to handle BIDI case by checking all text boxes for max offset

2005-03-07  David Harrison  <harrison@apple.com>

        Reviewed by Ken.

        <rdar://problem/4029225> REGRESSION (Mail): Crash if hit return after dragging in attachment - DeleteSelectionCommand::initializePositionData

        Work around the fact that the height() of a BR is 0 if there are no text elements on the line, even if there are replaced elements.
        Filed <rdar://problem/4040358> RenderBR height() is not accurate.
        
        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply):
        When height is 0, double check that the placeholder is the first position on the line.

=== Safari-403 ===

2005-03-07  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Vicki
    
        Fix for this bug:
        
        <rdar://problem/4040136> Expose SPI for WebCore's functions to create document fragments from plain text and markup strings

        * kwq/DOMHTML.mm: Add two new SPI functions.
        (-[DOMHTMLDocument _createDocumentFragmentWithMarkupString:baseURLString:])
        (-[DOMHTMLDocument _createDocumentFragmentWithText:])
        * kwq/DOMPrivate.h: Declare the new SPI here, a privately-exported header, so Mail can make use of it.

2005-03-07  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4039676> REGRESSION (Mail): in reply, <cr> on pasted text (without trailing newline) is eaten when pasted above date

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply): Code to "eat" a <br> element, that was creating an otherwise empty line, was running
        when it should not. If the content being pasted in does not end in a "logical" newline itself, then the <br> already
        in the content needs to be preserved. A simple one-line change now sees to this.
        
        * layout-tests/editing/pasteboard/paste-text-010-expected.txt: This test now has an extra, but harmless,
        <br> at the end of a paragraph. This makes sense given the code change.

        New test:
        * layout-tests/editing/pasteboard/paste-text-018-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-text-018.html: Added.

2005-03-07  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4035648> REGRESSION (Mail): line feed in source HTML file causes bad copy/paste behavior

        The createMarkup() function in markup.cpp iterates over the nodes in a range,
        and does some bookkeeping to figure out when to add close tags to the markup.
        Some code added at the start of the loop to prevent markup from being written
        for unrendered nodes short-circuited the rest of the loop, and so prevented
        the close-tag-writing code from running when it should.

        This is why the "plain" text wound up inside of the bold tag in the example
        above. The addition of the unrendered return character caused an incorrect
        delay in the close tag for the bold element from being written out, with the
        result being that it wound up including additional content.

        The fix is to add checks for node renderers throughout the loop at the points
        where markup is written out for each node. This allows the additional close
        tag logic to run as needed.

        All layout tests pass with this change.

        * khtml/editing/markup.cpp:
        (khtml::createMarkup)

        New test:
        
        * layout-tests/editing/pasteboard/paste-4035648-fix-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-4035648-fix.html: Added.

2005-03-06  Christy Warren  <kali@appple.com>

        Reviewed by Ken

        * khtml/rendering/render_text.cpp:
        (lastRendererOnPrevLine): helper for RenderText::caretRect
        (RenderText::caretRect): added code to properly handle bidi ordered text boxes

2005-03-06  Christy Warren  <kali@apple.com>

        Reviewed by Darin.

        * khtml/editing/visible_position.cpp: Eliminated code that cuts off searching through the text boxes based on an assumption that breaks under bidi
        (khtml::VisiblePosition::isCandidate):
        * kwq/KWQFontMetrics.mm:
        (QFontMetrics::checkSelectionPoint): Made the initialization of the WebCoreStyle honor the reversed flag

2005-03-05  John Sullivan  <sullivan@apple.com>

        Reviewed by Maciej.
        
        - fixed <rdar://problem/4038417> Mail crashed in StyleChange::checkForLegacyHTMLStyleChange 
        when composing a reply

        * khtml/editing/htmlediting.cpp:
        (khtml::StyleChange::checkForLegacyHTMLStyleChange):
        Added missing nil check.

2005-03-05  Darin Adler  <darin@apple.com>

        Reviewed by Don.

        - fixed <rdar://problem/4038478> Crash in renderer viewing RSS feed at feed://wvs.topleftpixel.com/index.rdf

        * khtml/xml/dom_docimpl.cpp: (widgetForNode): Check for nil before going from node to renderer.

2005-03-05  Darin Adler  <darin@apple.com>

        Reviewed by Richard.

        - fixed <rdar://problem/4037700> Every character typed causes stat call for /usr/share/icu/icudt32b_char.brk

        * khtml/rendering/render_text.cpp:
        (getCharacterBreakIterator): Added. Helper that sets up an iterator for the passed-in text.
        Shares a single global iterator (fast, albeit not thread-safe).
        (RenderText::previousOffset): Changed to call getCharacterBreakIterator.
        (RenderText::nextOffset): Ditto.

2005-03-05  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Darin

        Fix for this bug:
        
        <rdar://problem/4038267> REGRESSION (Mail): Crash copying and pasting end of paragraph

        Code to handle content that has a "logical" newline at the end of the pasted content, and the code
        to adjust the selection at the end of the paste operation made an assumption that at least one
        node had been inserted by the paste command. This is not necessarily true in the case where the sole content
        in the pasted content is one of these "logical" newlines. Adjust some code around so that we don't deref
        null, but still adjust the selection correctly for this case. In each of the two functions below, some
        null checks have been added, and some code has been rearranged a little bit to continue on through
        the end of completeHTMLReplacement, even if no nodes have been inserted. The patch looks bigger and more
        complicated than the conceptual change.

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply)
        (khtml::ReplaceSelectionCommand::completeHTMLReplacement)
        
        * layout-tests/editing/pasteboard/paste-4038267-fix-expected.txt: Added.
        * layout-tests/editing/pasteboard/paste-4038267-fix.html: Added.

2005-03-05  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - fixed <rdar://problem/4025918> images copied from Safari with relative src URLs aren't pasted into Mail messages (KURL resolves base URLs incorrectly)

        * kwq/KWQKURL.mm: (KURL::KURL): Add a slash at the start of the path if a relative part is adding
        a path onto a URL that has "pre-path" bits like host name, but no path yet. This doesn't come up
        for http because in that case we add a trailing "/" as part of canonicalization.

2005-03-04  John Sullivan  <sullivan@apple.com>

        Reviewed by Kevin.
        
        - fixed <rdar://problem/4033705> REGRESSION (Mail): Copy/Paste from Excel 
        crashes Mail in KWQKHTMLPart::fontForSelection(bool*) const + 232

        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::fontForSelection):
        Check for nil startNode, not just nil renderer.

2005-03-04  David Harrison  <harrison@apple.com>

        Reviewed by Ken, Richard.

        <rdar://problem/3996383> REGRESSION (Mail): Deleting all of first line also deletes line ending

        Problem was the placeholder check was based on 0 height block, but the block in this
        case is the BODY, which has other content even though the paragraph is gone.
        
        * khtml/editing/htmlediting.cpp:
        (khtml::CompositeEditCommand::insertBlockPlaceholder):
        Now does insert instead of append!
    
        (khtml::CompositeEditCommand::appendBlockPlaceholder):
        New.  Actually does append.
        
        (khtml::CompositeEditCommand::addBlockPlaceholderIfNeeded):
        Renamed from insertBlockPlaceholderIfNeeded because it can insert or append.  Also accepts "force insert" flag.
        
        (khtml::CompositeEditCommand::removeBlockPlaceholder):
        Renamed for succinctness from removeBlockPlaceholderIfNeeded.
        
        (khtml::DeleteSelectionCommand::insertPlaceholderForAncestorBlockContent):
        (khtml::DeleteSelectionCommand::moveNodesAfterNode):
        Use renamed methods.
        
        (khtml::DeleteSelectionCommand::doApply):
        Use addBlockPlaceholderIfNeeded "force insert" flag if deleting whole paragraphs did not leave an empty one.
        
        (khtml::InsertParagraphSeparatorCommand::doApply):
        (khtml::InsertParagraphSeparatorInQuotedContentCommand::doApply):
        (khtml::InsertTextCommand::input):
        Use renamed methods.
        
        * khtml/editing/htmlediting.h:
        Renamed and new methods per above.
        

2005-03-04  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4032840> REGRESSION (Mail): crash in RemoveNodeCommand after pasting attachment at end of message

        Reviewed by darin.

        * khtml/editing/visible_units.cpp:
        (khtml::endOfParagraph): don't consider text nodes that have no rendered characters

2005-03-04  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4034764> REGRESSION(125-188)Viewing text/plain page and going back/forward corrupts HTML pages (google.com)

        The parse mode wasn't be restored to the document when going
        back.

        Reviewed by Hyatt.

        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::openURLFromPageCache):
        * kwq/KWQPageState.h:
        * kwq/KWQPageState.mm:
        (-[KWQPageState initWithDocument:URL:windowProperties:locationProperties:interpreterBuiltins:]):

2005-03-04  Ken Kocienda  <kocienda@apple.com>

        Reviewed by me

        * layout-tests/editing/style/create-block-for-style-012-expected.txt: Shame on me!
        I landed my last fix without running layout tests. This one changes in a subtle, but
        acceptable way.

2005-03-04  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Harrison

        Fix for this bug:
        
        <rdar://problem/4032543> REGRESSION (Mail): Mail hangs when quoted text is pasted twice
        
        This code change fixes the bug in a non-obvious way. The root cause of the problem was
        that a VisiblePosition created using an affinity originating in Mail code caused 
        two VisiblePosition objects that should have been equal to differ only in their 
        affinities, which in turn caused us to run a code path that should not have run.

        * khtml/editing/visible_position.cpp:
        (khtml::VisiblePosition::VisiblePosition): Added copy constructor.
        (khtml::VisiblePosition::next): Factored out inline code that used to be here into new
        setAffinityUsingLinePosition() function.
        (khtml::isEqualIgnoringAffinity): New helper to handle cases when affinity in equality check does
        not matter. However, we want to know about such cases where a VisiblePosition differs only by affinity, 
        and the code will assert in development when this happens.
        (khtml::isNotEqualIgnoringAffinity): Ditto, but not. :)
        (khtml::setAffinityUsingLinePosition): New helper function mentioned above. This will "correct"
        upstream affinity to downstream if the affinity does not make a difference for the position.
        * khtml/editing/visible_position.h:
        * khtml/editing/visible_range.h: Wacky bug. The operator== for this class took VisiblePosition classes!
        * khtml/editing/visible_units.cpp:
        (khtml::isStartOfParagraph): Now performs equality check without regard to affinity.
        (khtml::isEndOfParagraph): Ditto.
        (khtml::isStartOfBlock): Ditto.
        (khtml::isEndOfBlock): Ditto.
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge setSelectedDOMRange:affinity:]): Adjusts the affinity using setAffinityUsingLinePosition()
        if necessary.

2005-03-04  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - fixed <rdar://problem/4036817> REGRESSION: ctrl-y broken when a line + carriage return cut

        * khtml/editing/htmlediting.h: Added insertTextRunWithoutNewlines.
        * khtml/editing/htmlediting.cpp:
        (khtml::InsertTextCommand::input): Added assertion to make sure strings with newlines don't get
        down to this level.
        (khtml::TypingCommand::insertText): Added. Takes the name of the old function, but is a new
        level that breaks runs into lines and inserts each one separately.
        (khtml::TypingCommand::insertTextRunWithoutNewlines): Renamed old insertText to this. 

        * khtml/dom/dom_string.h: Made substring be a const member function.
        * khtml/dom/dom_string.cpp: (DOM::DOMString::substring): Made this const.

2005-03-04  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - fixed <rdar://problem/4020413> REGRESSION (Mail): can't use fonts with names that start with "#" in Mail (Korean fonts)

        * khtml/css/css_valueimpl.cpp:
        (DOM::isLegalIdentifier): Added. Commented out and not used.
        (DOM::quoteStringIfNeeded): Quotes the string if needed. For now only if it starts with "#".
        (DOM::CSSPrimitiveValueImpl::cssText): Call quoteStringIfNeeded when asked for cssText for an arbitrary string, since we
        need text you can re-parse.
        (DOM::FontFamilyValueImpl::cssText): Ditto.

        * khtml/editing/markup.cpp:
        (khtml::startMarkup): Added comments about lack of quoting for attributes.
        (khtml::createMarkup): Ditto.

2005-03-04 Adele Amchan  <adele@apple.com>
 
       Reviewed by Maciej

        Fix for <rdar://problem/4021711> REGRESSION(125-188) blank pages when browsing forum at cooperativeresearch.org - cached external script problem

        * khtml/html/htmltokenizer.cpp: (khtml::HTMLTokenizer::scriptHandler): set flag needToRefCachedScript so we can make sure 
        notifyFinished is called after pendingSrc is set up with the right data from the cached script

2005-03-04  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4029632> Tiger 8A398:- Mail crashes in DOM::NodeImpl::isBlockFlow() after pasting text with alignment style and BR element from Safari

        * khtml/editing/htmlediting.cpp:
        (khtml::CompositeEditCommand::moveParagraphContentsToNewBlockIfNecessary): VisiblePosition constructor
        was failing to yield a position for a node just pasted into the document since a layout was needed
        for the calculation to come out right. Layout added. Crash gone.

=== Safari-402 ===

2005-03-03  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/3992803> Cannot navigate through list items with the keyboard, you're stuck in the same <li> block unless you click out

        Reviewed by harrison.

        * khtml/rendering/render_line.cpp:
        (khtml::RootInlineBox::closestLeafChildForXPos): avoid returning list markers when possible. Also improved heuristic for returning the best InlineBox given the passed X position.


2005-03-03  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John
        
        Fix for this bug:
        
        <rdar://problem/4035198> Pasting text with different styles does not get reproducible results

        I had a good idea yesterday to improve the way we "fixup" styles after
        pasting, but i chose a poor data structure to do it, a map of
        nodes-to-styles. In the fixup step, I iterated over the map elements and
        did the fixup. However, since the order in which the items would come
        out of the map is indeterminate, we got unpredictable results.

        So, the concept was good, but the implementation was flawed. I have
        fixed this mapping to be a list instead, so the order that nodes are
        evaluated in the fixup step is document order. This works nicely.

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplacementFragment::~ReplacementFragment): No longer need to explicity deref nodes and
        styles saved away for later fixup. This is now handled by the new NodeDesiredStyle class.
        (khtml::ReplacementFragment::computeStylesUsingTestRendering): Now calls computeAndStoreNodeDesiredStyle,
        function renamed from mapDesiredStyleForNode.
        Now accepts a QValueList<NodeDesiredStyle> in place of the old map.
        (khtml::NodeDesiredStyle::NodeDesiredStyle): New class that represents a node-to-style mapping.
        (khtml::NodeDesiredStyle::~NodeDesiredStyle): Ditto.
        (khtml::NodeDesiredStyle::operator=): Ditto.
        (khtml::ReplaceSelectionCommand::doApply): Now calls computeAndStoreNodeDesiredStyle,
        function renamed from mapDesiredStyleForNode.
        (khtml::ReplaceSelectionCommand::fixupNodeStyles): Now operates on a QValueList<NodeDesiredStyle> in 
        place of the old map.
        (khtml::computeAndStoreNodeDesiredStyle): Renamed from mapDesiredStyleForNode.  Now operates on a 
        QValueList<NodeDesiredStyle> in place of the old map.
        * khtml/editing/htmlediting.h:
        (khtml::NodeDesiredStyle): New class that represents a node-to-style mapping.
        (khtml::ReplacementFragment::desiredStyles): Now returns a QValueList<NodeDesiredStyle> in place of the old map.

        * layout-tests/editing/style/typing-style-003-expected.txt: Results changed in an acceptable way.

2005-03-02  Darin Adler  <darin@apple.com>

        Reviewed by Maciej.

        - fixed <rdar://problem/4029741> REGRESSION (188-189): <input type=text> are cleared when you hide/show them

        * khtml/html/html_formimpl.h: Added detach functions for both input and text area elements, since
        both have m_valueMatchesRenderer flags.
        * khtml/html/html_formimpl.cpp:
        (DOM::HTMLInputElementImpl::detach): Set m_valueMatchesRenderer to false, since the renderer is going away.
        (DOM::HTMLTextAreaElementImpl::detach): Ditto.

2005-03-02  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Hyatt

        Fix for this bug:
        
        <rdar://problem/4006151> in reply, caret moves to next line after toggling bold style then typing

        * khtml/editing/htmlediting.cpp:
        (khtml::ApplyStyleCommand::applyInlineStyle): Defer cleaning up empty style until the end of the
        function. In some situations, the render tree can get confused when we do this removal up front.
        I wish I understood the reasons more deeply, but this bit of code shuffling seems harmless, and 
        fixes the bug.

2005-03-02  Darin Adler  <darin@apple.com>

        Reviewed by Maciej.

        - fixed <rdar://problem/4024966> crash happened twice in [KWQTextField setHasFocus:] + 0xe8

        Added more nil checks to the widget implementations. Anywhere a widget pointer is used, do a nil check,
        rather than relying on high level assumptions about which calls can and can't destroy the HTML element
        (and hence the widget). The case in the bug seems to be a case where scrolling destroyed the element.
        We can't be sure this fixes the bug, but we can be relatively sure we didn't introduce any new problems,
        because this just avoids nil-dereferencing.

        * kwq/KWQButton.mm:
        (-[KWQButton becomeFirstResponder]):
        (-[KWQButton nextKeyView]):
        (-[KWQButton previousKeyView]):
        * kwq/KWQListBox.mm:
        (-[KWQTableView mouseDown:]):
        (-[KWQTableView becomeFirstResponder]):
        (-[KWQTableView numberOfRowsInTableView:]):
        (-[KWQTableView tableViewSelectionDidChange:]):
        (-[KWQTableView tableView:shouldSelectRow:]):
        (-[KWQTableView selectionShouldChangeInTableView:]):
        * kwq/KWQScrollBar.mm:
        (-[KWQScrollBar scroll:]):
        * kwq/KWQSlider.mm:
        (-[KWQSlider mouseDown:]):
        (-[KWQSlider slide:]):
        (-[KWQSlider becomeFirstResponder]):
        (-[KWQSlider nextKeyView]):
        (-[KWQSlider previousKeyView]):
        * kwq/KWQTextArea.mm:
        (-[KWQTextAreaTextView becomeFirstResponder]):
        (-[KWQTextAreaTextView resignFirstResponder]):
        (-[KWQTextAreaTextView mouseDown:]):
        * kwq/KWQTextField.mm:
        (-[KWQTextFieldController action:]):
        (-[KWQTextFieldController controlTextDidEndEditing:]):
        (-[KWQTextFieldController controlTextDidChange:]):
        (-[KWQTextFieldController textView:didHandleEvent:]):
        (-[KWQTextFieldController setHasFocus:]):

2005-03-02  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4031483> Leak (RenderObject::createInlineBox) reproducible with Stock widget

        We were leaking inline block line boxes.  Argh!

        Reviewed by Hyatt.

        * khtml/rendering/render_flow.cpp:
        (RenderFlow::dirtyLineBoxes):
        * khtml/rendering/render_line.cpp:
        (khtml::InlineBox::deleteLine):

2005-03-02  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/3976872> REGRESSION (Mail): Pasted plain text doesn't get the proper style if pasted into newlines

        Reviewed by kocienda.

        * khtml/editing/htmlediting.cpp:
        (khtml::EditCommand::styleAtPosition): new, factored out from calculateStyleBeforeInsertion
        (khtml::InsertParagraphSeparatorCommand::calculateStyleBeforeInsertion): call styleAtPosition
        (khtml::ReplaceSelectionCommand::ReplaceSelectionCommand): clear new m_insertionStyle ivar
        (khtml::ReplaceSelectionCommand::~ReplaceSelectionCommand): deref new m_insertionStyle  ivar
        (khtml::ReplaceSelectionCommand::doApply): store the style so it later be applied when matching style
        (khtml::ReplaceSelectionCommand::completeHTMLReplacement): apply style from m_insertionStyle when matching style
        * khtml/editing/htmlediting.h:

2005-03-02  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Darin.

        <rdar://problem/4031718> REGRESSION (401-401+): Safari reproducible crash setting up scope in JSLazyEventListener::parseCode authenticating to bugweb
        
        * khtml/ecma/kjs_events.cpp:
        (JSLazyEventListener::parseCode): If originalNode is NULL, don't
        mess with the scope chain.
        
        * khtml/html/html_baseimpl.cpp:
        (HTMLBodyElementImpl::parseHTMLAttribute): For handlers that are
        delcared on body but set on the document, pass NULL to avoid
        swizzling the scope chain. It turns out that this is what browsers
        do, and it finesses the crash.

2005-03-02  David Harrison  <harrison@apple.com>

        Reviewed by Ken.

        <rdar://problem/3948453> Can't type accented chars as first character in Stickies widget

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply):

2005-03-02  Ken Kocienda  <kocienda@apple.com>
        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4020574> REGRESSION (Mail): copy/paste first part of reply-quoted text alters downstream style
        
        The problem was that the operation to move nodes following the newly-pasted nodes did not preserve the
        style of these moved nodes. I have generalized some of the functions that compute and preserve styles
        for nodes and then apply these styles after a DOM operation.

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplacementFragment::~ReplacementFragment): Call new derefNodesAndStylesInMap() helper function
        in place of old code that had this deref'ing inline.
        (khtml::ReplacementFragment::computeStylesUsingTestRendering): Now calls new mapDesiredStyleForNode() helper
        function place of old code that had this style computation inline.
        (khtml::ReplacementFragment::removeStyleNodes): Updated comment for new helper name.
        (khtml::ReplaceSelectionCommand::doApply): Now calls new helpers in place of helpers whose names were changed,
        or in place of pre-refactored inline code.
        (khtml::ReplaceSelectionCommand::fixupNodeStyles): Renamed from applyStyleToInsertedNodes(). Now generalized
        to take the map of nodes to use for the fixup. This makes it possible to call this code with different maps,
        and that is needed to fix the bug.
        (khtml::mapDesiredStyleForNode): New helper function to compute the inheritable styles for a given node
        and map this style to the given node in the given map. This function now also includes the code that was
        in the removeBlockquoteColorsIfNeeded(). This latter helper has now been removed.
        (khtml::derefNodesAndStylesInMap): Simple helper to deref map members.
        * khtml/editing/htmlediting.h:
        (khtml::ReplacementFragment::desiredStyles): New helper to return map of nodes-to-desiredStyles.

2005-03-01  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Hyatt

        Fix for this bug:

        <rdar://problem/4011358> REGRESSION(Mail): after dragging text to mail message and deleting, typing or dragging new text doesn't work

        * khtml/editing/visible_position.cpp:
        (khtml::VisiblePosition::isCandidate): Insertion point disappearing was due to a failed
        check here after deleting all the content in the body element. The special "empty block"
        needs to add check for DOM children. A block may have straggling anonymous render children in
        some cases, and so the check needs to be (!DOM-kids || !render-kids).

        Also, I botched the last checkin trying to split these two patches in my tree. Wrong layout tests 
        got checked in, I checked in a conflict marker in the Changelog, etc. I fixed everything with this checkin.

2005-03-01  Ken Kocienda  <kocienda@apple.com>
 
        Reviewed by Maciej

        Improved fix for this bug:
        
        <rdar://problem/3996605> Insert paragraph command puts new block in wrong place, creating difficult-to-handle HTML

        Maciej and I discussed this situation at length, and we came up with a better fix than I did earlier.

        * khtml/editing/htmlediting.cpp:
        (khtml::InsertParagraphSeparatorCommand::doApply): Simplify one special case so that it only handles the 
        "last in block" situation. Remove special case for "downstream node is in different block" and handle
        this case with a little bit of special code in the general insertion case.

        Results studied to make sure there were no problems.

        * layout-tests/editing/deleting/delete-3959464-fix-expected.txt
        * layout-tests/editing/inserting/insert-div-001-expected.txt
        * layout-tests/editing/inserting/insert-div-002-expected.txt
        * layout-tests/editing/inserting/insert-div-004-expected.txt
        * layout-tests/editing/inserting/insert-div-005-expected.txt
        * layout-tests/editing/inserting/insert-div-009-expected.txt
        * layout-tests/editing/inserting/insert-div-014-expected.txt
        * layout-tests/editing/inserting/insert-div-018-expected.txt
        * layout-tests/editing/inserting/insert-div-024-expected.txt
        * layout-tests/editing/pasteboard/paste-text-011-expected.txt
        * layout-tests/editing/pasteboard/paste-text-013-expected.txt
        * layout-tests/editing/pasteboard/paste-text-015-expected.txt
        * layout-tests/editing/style/block-style-004-expected.txt
        * layout-tests/editing/style/block-style-005-expected.txt
        * layout-tests/editing/style/block-style-006-expected.txt

        New test:
        
        * layout-tests/editing/inserting/insert-div-027.html
        * layout-tests/editing/inserting/insert-div-027-expected.txt

2005-03-01  David Hyatt  <hyatt@apple.com>

        Fix for 4030890, regression with <sup> on Google.  Fix some bogus != comparison checks in verticalPositionHint.
        
        Reviewed by John

        (RenderObject::getVerticalPosition):
        
2005-03-01  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4030669> smart delete does not delete spaces from pasted content

        Reviewed by rjw.

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::initializePositionData): pass true for treatNBSPAsWhiteSpace to leadingWhitespacePosition and trailingWhitespacePosition when testing whether or not we have to delete those characters as well.
        * khtml/editing/jsediting.cpp: pass true for smartDelete when the selection granularity is WORD. This allows us to write smart delete layout tests.
        * layout-tests/editing/deleting/smart-delete-001-expected.txt: Added.
        * layout-tests/editing/deleting/smart-delete-001.html: Added.
        * layout-tests/editing/deleting/smart-delete-002-expected.txt: Added.
        * layout-tests/editing/deleting/smart-delete-002.html: Added.

2005-03-01  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4029772> Weather widgets use a lot of memory (more that other widgets)

        The string objects created by KWQHeaderStringFromDictionary() leaked.

        Reviewed by John Louch.

        * kwq/KWQLoader.mm:
        (KWQHeaderStringFromDictionary):

2005-03-01  Jens Alfke  <jens@apple.com>

        Reviewed by Darin.

        <rdar://problem/4004531> Simple Sun Signed Applet throws exceptions; doesn't run
        Two fixes for the way we extract the parameter-y goodness from <object> tags.

        * khtml/rendering/render_frames.cpp:
        (RenderPartObject::updateWidget):

2005-03-01  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/3990258> REGRESSION (125.12-181): top of picture that spans two pages is cut off on print from Safari

        The problem was due to incorrect "object truncation" in  RenderFlow::paintLines.  
        We attempt to push objects down if they don't fit on a page at
        paint time.  If the attempt to push object down failed we just
        didn't paint at all.

        Reviewed by Hyatt.

        * khtml/rendering/render_flow.cpp:
        (RenderFlow::paintLines):

2005-03-01  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4030669> smart delete does not delete spaces from pasted content

        Reviewed by rjw.

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::initializePositionData): pass true for treatNBSPAsWhiteSpace to leadingWhitespacePosition and trailingWhitespacePosition when testing whether or not we have to delete those characters as well.
        * khtml/editing/jsediting.cpp: pass true for smartDelete when the selection granularity is WORD. This allows us to write smart delete layout tests.
        * layout-tests/editing/deleting/smart-delete-001-expected.txt: Added.
        * layout-tests/editing/deleting/smart-delete-001.html: Added.
        * layout-tests/editing/deleting/smart-delete-002-expected.txt: Added.
        * layout-tests/editing/deleting/smart-delete-002.html: Added.

2005-03-01  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Vicki

        Fix for this bug:
        
        <rdar://problem/4030068> Trailing <return> gets eaten when pasted at the end of a document

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply): Add code to handle formerly-unhandled end-of-document case.

2005-03-01  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Vicki.

        <rdar://problem/4003774> REGRESSION(125-181): JavaScript problems @ Yankee/Dixie quiz
        
        Reworked how scopes are set up for event handlers to match other
        browser. This includes the following changes:

        - Special scope entries are set up at the time the event handler
        is created, not at the time it fires.

        - Special scope is only set up for event handlers set in the html
        source through an html attribute, not for handlers set using
        addEventHandler or setting JS properties like onclick through
        JavaScript.

        - Special scope is based on the DOM node on which the handler is
        an attribute, not the event target.

        This fixes the regression while allowing the fix to
        <rdar://problem/3798453> (DIG: getting variable with same name as
        DOM element attribute gets attribute value instead) to keep
        working correctly.
        
        * khtml/ecma/kjs_events.cpp:
        (JSEventListener::handleEvent):
        (JSLazyEventListener::JSLazyEventListener):
        (JSLazyEventListener::parseCode):
        * khtml/ecma/kjs_events.h:
        * khtml/ecma/kjs_proxy.cpp:
        (KJSProxyImpl::createHTMLEventHandler):
        * khtml/ecma/kjs_proxy.h:
        * khtml/ecma/kjs_window.cpp:
        (Window::getJSLazyEventListener):
        * khtml/ecma/kjs_window.h:
        * khtml/html/html_baseimpl.cpp:
        (HTMLBodyElementImpl::parseHTMLAttribute):
        (HTMLFrameElementImpl::parseHTMLAttribute):
        (HTMLFrameSetElementImpl::parseHTMLAttribute):
        * khtml/html/html_elementimpl.cpp:
        (HTMLElementImpl::parseHTMLAttribute):
        * khtml/html/html_formimpl.cpp:
        (DOM::HTMLFormElementImpl::parseHTMLAttribute):
        (DOM::HTMLButtonElementImpl::parseHTMLAttribute):
        (DOM::HTMLInputElementImpl::parseHTMLAttribute):
        (DOM::HTMLLabelElementImpl::parseHTMLAttribute):
        (DOM::HTMLSelectElementImpl::parseHTMLAttribute):
        (DOM::HTMLTextAreaElementImpl::parseHTMLAttribute):
        * khtml/html/html_imageimpl.cpp:
        (HTMLImageElementImpl::parseHTMLAttribute):
        * khtml/html/html_objectimpl.cpp:
        (HTMLObjectElementImpl::parseHTMLAttribute):
        * khtml/khtml_part.cpp:
        (KHTMLPart::createHTMLEventListener):
        * khtml/khtml_part.h:
        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::createHTMLEventListener):
        * khtml/xml/dom_docimpl.h:

=== Safari-401 ===

2005-03-01  Chris Blumenberg  <cblu@apple.com>

        Backed out my patch since the tree was closed.

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::initializePositionData):
        * khtml/editing/jsediting.cpp:

2005-03-01  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4030669> smart delete does not delete spaces from pasted content

        Reviewed by rjw.

        * khtml/editing/htmlediting.cpp:
        (khtml::DeleteSelectionCommand::initializePositionData): pass true for treatNBSPAsWhiteSpace to leadingWhitespacePosition and trailingWhitespacePosition when testing whether or not we have to delete those characters as well.
        * khtml/editing/jsediting.cpp: pass true for smartDelete when the selection granularity is WORD. This allows us to write smart delete layout tests.
        * layout-tests/editing/deleting/smart-delete-001-expected.txt: Added.
        * layout-tests/editing/deleting/smart-delete-001.html: Added.
        * layout-tests/editing/deleting/smart-delete-002-expected.txt: Added.
        * layout-tests/editing/deleting/smart-delete-002.html: Added.

2005-03-01  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Vicki

        Fix for this bug:
        
        <rdar://problem/4030068> Trailing <return> gets eaten when pasted at the end of a document

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply): Add code to handle formerly-unhandled end-of-document case.

2005-03-01  David Harrison  <harrison@apple.com>

        Reviewed by Chris.

        <rdar://problem/3915560> Mail would like an SPI to enable "smart" text paste/drop

        * kwq/WebCoreBridge.h:
        * kwq/WebCoreBridge.mm:
        (-[WebCoreBridge smartDeleteRangeForProposedRange:]):
        (-[WebCoreBridge smartInsertForString:replacingRange:beforeString:afterString:]):
        New.

2005-03-01  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4030404> selection granularity should be set when extending selection via JS

        Reviewed by kocienda.

        * khtml/ecma/kjs_window.cpp:
        (SelectionFunc::tryCall): set the granularity on the part. This will allow us to write smart paste layout tests.
        * layout-tests/editing/pasteboard/smart-paste-001-expected.txt: Added.
        * layout-tests/editing/pasteboard/smart-paste-001.html: Added.
        * layout-tests/editing/pasteboard/smart-paste-002-expected.txt: Added.
        * layout-tests/editing/pasteboard/smart-paste-002.html: Added.
        * layout-tests/editing/pasteboard/smart-paste-003-expected.txt: Added.
        * layout-tests/editing/pasteboard/smart-paste-003.html: Added.
        * layout-tests/editing/pasteboard/smart-paste-004-expected.txt: Added.
        * layout-tests/editing/pasteboard/smart-paste-004.html: Added.
        * layout-tests/editing/pasteboard/smart-paste-005-expected.txt: Added.
        * layout-tests/editing/pasteboard/smart-paste-005.html: Added.
        * layout-tests/editing/pasteboard/smart-paste-006-expected.txt: Added.
        * layout-tests/editing/pasteboard/smart-paste-006.html: Added.
        * layout-tests/editing/pasteboard/smart-paste-007-expected.txt: Added.
        * layout-tests/editing/pasteboard/smart-paste-007.html: Added.

2005-03-01  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4029934> smart paste with plain text can add too many spaces

        Reviewed by kocienda.

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::doApply): pass true for treatNBSPAsWhitespace to leadingWhitespacePosition and trailingWhitespacePosition
        * khtml/xml/dom_position.cpp:
        (DOM::isWS): take treatNBSPAsWhitespace param
        (DOM::Position::leadingWhitespacePosition): ditto
        (DOM::Position::trailingWhitespacePosition): ditto
        * khtml/xml/dom_position.h:

2005-03-01  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/3996605> Insert paragraph command puts new block in wrong place, creating difficult-to-handle HTML

        * khtml/editing/htmlediting.cpp:
        (khtml::InsertParagraphSeparatorCommand::doApply): Look for the upstream-most block to insert after
        when at the visible end of a block. This helps to avoid some undesirable sequences of markup which
        Dave says will be vary hard to render. Changing the command in this way avoids the "limitations"
        of the render tree by not asking it to render markup we do not want to make anyway.
        
        All these tests change, but either in insignificant ways, or for the better.
        
        * layout-tests/editing/deleting/delete-3959464-fix-expected.txt:
        * layout-tests/editing/inserting/insert-div-001-expected.txt:
        * layout-tests/editing/inserting/insert-div-002-expected.txt:
        * layout-tests/editing/inserting/insert-div-004-expected.txt:
        * layout-tests/editing/inserting/insert-div-005-expected.txt:
        * layout-tests/editing/inserting/insert-div-009-expected.txt:
        * layout-tests/editing/inserting/insert-div-024-expected.txt:
        * layout-tests/editing/pasteboard/paste-text-011-expected.txt:
        * layout-tests/editing/pasteboard/paste-text-013-expected.txt:
        * layout-tests/editing/pasteboard/paste-text-015-expected.txt:
        * layout-tests/editing/style/block-style-004-expected.txt:
        * layout-tests/editing/style/block-style-005-expected.txt:
        * layout-tests/editing/style/block-style-006-expected.txt:
        * layout-tests/editing/style/remove-underline-across-paragraph-expected.txt:
        * layout-tests/editing/style/remove-underline-across-paragraph-in-bold-expected.txt:
        * layout-tests/editing/style/remove-underline-after-paragraph-expected.txt:
        * layout-tests/editing/style/remove-underline-after-paragraph-in-bold-expected.txt:

        New test to check specific problem mentioned in the bug.

        * layout-tests/editing/inserting/insert-div-026-expected.txt: Added.
        * layout-tests/editing/inserting/insert-div-026.html: Added.

2005-02-28  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Darin.

        <rdar://problem/4002864> REGRESSION(125-146) getElementById in onload fails in a test case involving external resources
        
        Moved management of elementById hashtable from attach/detach to
        insertedIntoDocument/removedFromDocument, to avoid being thrown
        off by temporary detaches due to style recalcs.
        
        * khtml/xml/dom_elementimpl.cpp:
        (ElementImpl::insertedIntoDocument):
        (ElementImpl::removedFromDocument):
        (ElementImpl::attach):
        (ElementImpl::updateId):
        * khtml/xml/dom_elementimpl.h:

        Make sure that insertedIntoDocument is called before firing any
        DOM events.
        
        * khtml/xml/dom_nodeimpl.cpp:
        (NodeBaseImpl::dispatchChildInsertedEvents): 

2005-02-28  David Hyatt  <hyatt@apple.com>

        Fix for 4028999, safari crashes when resetting if mallocsribble is on.  Clip rects were being cleared using dead
        render objects.  Change the ordering.
        
        Reviewed by rjw

        * khtml/rendering/render_box.cpp:
        (RenderBox::detach):
        * khtml/rendering/render_layer.cpp:
        (RenderLayer::~RenderLayer):

2005-02-28  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4024786> REGRESSION (Mail): "Smart paste" plain-text word into Blot leaves insertion point misplaced

        Reviewed by kocienda.

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplaceSelectionCommand::completeHTMLReplacement): Call updateLayout so caretMinOffset and caretMaxOffset return correct values.

2005-02-28  John Sullivan  <sullivan@apple.com>

        Reviewed by Ken.
        
        - WebCore part of fix for <rdar://problem/4023490> REGRESSION (125-185): Tabbing through links 
        on frameset page gets stuck at end (tivofaq.com)

        * kwq/WebCoreBridge.h:
        add nextValidKeyViewOutsideWebFrameViews (code is in WebKit)

2005-02-28  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/3972665> 8A360: HTML message partially truncated on left hand side, text-indent from Script Editor

        * khtml/css/css_computedstyle.cpp: inheritableProperties array now defined in css_valueimpl.cpp.
        * khtml/css/css_valueimpl.cpp: Define inheritableProperties array here.
        (DOM::CSSMutableStyleDeclarationImpl::copyBlockProperties): Use new name for blockProperties, and use the new
        constant for the number of items in the array.
        (DOM::CSSMutableStyleDeclarationImpl::removeBlockProperties): Ditto.
        (DOM::CSSMutableStyleDeclarationImpl::removeInheritableProperties): New function.
        * khtml/css/css_valueimpl.h: Declare inheritableProperties array and numInheritableProperties extern so they
        can be defined in css_valueimpl.cpp and used in css_computedstyle.cpp.
        * khtml/editing/htmlediting.cpp:
        (khtml::ReplacementFragment::removeStyleNodes): This code was misguided, and removed too much style from HTML
        elements. Now, it removes from HTML elements only the styles that we replace later with a call to applyStyle().
        Also, add ID_B to list of inline "style" nodes we are willing to remove. Leaving it off was an oversight.

        * layout-tests/editing/pasteboard/paste-text-011-expected.txt: ID_B fix made this result change, without any
        visible change in the test.
        
        New test:
        
        * layout-tests/editing/style/smoosh-styles-003.html
        * layout-tests/editing/style/smoosh-styles-003-expected.txt

2005-02-28  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4026985> CrashTracer: ...14 crashes at com.apple.WebCore: -[KWQPageState invalidate] + 32

        Added more nil checking and ASSERTS.
        Without a reproducible case this is hard to definitively resolve.

        Reviewed by John Sullivan.

        * kwq/KWQPageState.mm:
        (-[KWQPageState invalidate]):

2005-02-28  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/4027702> 3.5% performance regression btwn Safari-188 and Safari-400

        I inadvertently checked in some debugging code that disabled
        style sharing.  Backed out that change.

        * khtml/css/cssstyleselector.cpp:
        (khtml::CSSStyleSelector::styleForElement):

2005-02-28  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4026639> www.bmw.ca configurator does not work with Safari

        Reviewed by john.

        * khtml/ecma/kjs_html.cpp:
        (KJS::HTMLElement::tryGet): when frameset.<name of frame child> is called, return the window object of the frame child

2005-02-28  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Chris

        Fix for this bug:
        
        <rdar://problem/4026906> Paste of HTML table content can break table structure

        * khtml/editing/htmlediting.cpp:
        (khtml::ReplacementFragment::pruneEmptyNodes): Call new isProbablyTableStructureNode() function to prevent
        removal of empty table structure nodes.
        (khtml::ReplacementFragment::removeUnrenderedNodesUsingTestRendering): Ditto.
        (khtml::isProbablyTableStructureNode): New helper function.
        * khtml/editing/htmlediting.h: Declare new helper.

2005-02-28  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4023566> Stickies: Crash in ReplacementFragment::insertFragmentForTestRendering on paste

        Reviewed by kocienda.

        * khtml/editing/htmlediting.cpp:
        (khtml::ApplyStyleCommand::addInlineStyleIfNeeded): ref and deref the element while it is "floating"
        (khtml::ReplacementFragment::insertFragmentForTestRendering): ditto
        (khtml::floatRefdElement): new, keeps an element alive while its ref count is 0
        (khtml::createDefaultParagraphElement): removed commented out code
        (khtml::createBlockPlaceholderElement): ref the element and return it as "floating"
        (khtml::createFontElement): ditto
        (khtml::createStyleSpanElement): ditto
        * khtml/editing/htmlediting.h:

2005-02-27  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Vicki.

        <rdar://problem/3993557> REGRESSION (125-180-ish): getElementsByTagName no longer works with namespace designations
        
        * khtml/xml/dom_nodeimpl.cpp:
        (NodeBaseImpl::getElementsByTagNameNS): When no namespace is specified, find elements
        of the specified name in any namespace to match Mozilla and earlier Safari behavior.

2005-02-25  Darin Adler  <darin@apple.com>

        Reviewed by Adele.

        - fixed <rdar://problem/4025618> Crash while searching at hollywoodvideo.com

        * khtml/html/html_formimpl.h: Added valueWithDefault.
        * khtml/html/html_formimpl.cpp:
        (DOM::HTMLInputElementImpl::appendFormData): Call valueWithDefault instead of going at the render object
        to try to get the default value; there may be no render object if this is display:none.
        (DOM::HTMLInputElementImpl::valueWithDefault): Added. Knows about the defaults for "submit" and "reset"
        buttons; otherwise just returns the value as-is.

        * khtml/rendering/render_form.h: Removed the defaultLabel member functions.
        * khtml/rendering/render_form.cpp:
        (RenderSubmitButton::rawText): Call valueWithDefault instead of using defaultLabel function here on
        the render side. The DOM needs to know how to deal with the default anyway for form submission.
        (RenderSubmitButton::defaultLabel): Removed.
        (RenderResetButton::defaultLabel): Removed.
        (RenderPushButton::defaultLabel): Removed.

2005-02-25  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - fixed <rdar://problem/4025088> window onblur and onfocus don't fire when text field has focus

        * kwq/KWQKHTMLPart.h: Added setWindowHasFocus function and m_windowHasFocus data member.
        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::setDisplaysWithFocusAttributes): Took out the code that sends the focus and blur events.
        (KWQKHTMLPart::setWindowHasFocus): Put that code here instead.

        * kwq/WebCoreBridge.h: Added setWindowHasFocus: method to the bridge.
        * kwq/WebCoreBridge.mm: (-[WebCoreBridge setWindowHasFocus:]): Added. Calls method on the part.

2005-02-25  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - re-fixed <rdar://problem/3665430> horizontal scroll bar of text area does not show, even when text is wide in "no wrap" mode

        * kwq/KWQTextArea.mm: (-[KWQTextArea _updateTextViewWidth]): Don't change the text view width
        to match the text area's width in the "wrap" case.

2005-02-25  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Hyatt

        Fix for this bug:
        
        <rdar://problem/4021518> 8A394 Mail crashes during paste: khtml::RootInlineBox::closestLeafChildForXPos

        * khtml/editing/visible_units.cpp:
        (khtml::previousLinePosition): Adding an updateLayout call at the start of the function fixes the crash, since
        we caught line boxes in a not-completely-updated state.
        (khtml::nextLinePosition): Ditto.

2005-02-25  David Hyatt  <hyatt@apple.com>

        Fix for 4010774, make sure to avoid an O(N^2) algorithm in nextRenderer() that is triggered when large
        DOM subtrees are inserted into documents via one insert/append call.
        
        Reviewed by kocienda

        * khtml/xml/dom_nodeimpl.cpp:
        (NodeImpl::nextRenderer):

2005-02-25  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/3382926> Bidi neutrals at RTL/LTR boundaries not handled correctly.

        Second pass at fixing 3382926 w/o causing layout regressions.  Same concept:
        if directionality of text's element is RTL and first character has neutral directionality
        then set the initial directionality to RTL.

        Reviewed by Hyatt.

        * khtml/rendering/bidi.cpp:
        (khtml::RenderBlock::bidiReorderLine):
        (khtml::RenderBlock::determineStartPosition):

=== Safari-400 ===

2005-02-25  Ken Kocienda  <kocienda@apple.com>

        Reviewed by me

        Roll out Chris' change to fix this bug:
        
        <rdar://problem/4023566> Stickies: Crash in ReplacementFragment::insertFragmentForTestRendering on paste
        
        That code change is responsible for all these new crashers:
        
        <rdar://problem/4025177> crash copying safari.apple.com into Blot document
        <rdar://problem/4025184> crash in DOM::NodeImpl::parentNode copying "New!" from google.com to Blot
        <rdar://problem/4025214> crash in DOM::NodeImpl::getRect loading paste-match-style-001.html

        Since we wish to close the tree for a build right now, I am rolling out rather than investigating.

        * khtml/editing/htmlediting.cpp:
        (khtml::ApplyStyleCommand::addInlineStyleIfNeeded)
        (khtml::ReplacementFragment::insertFragmentForTestRendering)
        (khtml::createDefaultParagraphElement)
        (khtml::createBlockPlaceholderElement)
        (khtml::createFontElement)
        (khtml::createStyleSpanElement)
        * khtml/editing/htmlediting.h

2005-02-25  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/4020108> Pasting text into message makes Mail crash reproducibly

        Note that, even with this fix, development build will crash until this bug is fixed:
        <rdar://problem/4024996> Applying block styles can cause assertion failure in inline style removal
        
        This will not crash deployment builds, so I am going to land.

        * khtml/editing/htmlediting.cpp:
        (khtml::ApplyStyleCommand::applyBlockStyle): Applying block styles can make the loop to reach beyondEnd
        fail since the structure of the document can change. Cache the next node first before operating on it,
        as we do elsewhere.

2005-02-25  Vicki Murley <vicki@apple.com>

        - recommit this change, since rolling it out did NOT fix the performance regression!

    2005-02-23  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - fixed <rdar://problem/4011405> REGRESSION (180-181): Unconfirmed text disappears when text focus moves

        The key was to change things around so that we don't push text from the DOM to the widget
        unless the DOM has actually been changed. This prevents the code path that wipes out inline input
        during the blur process.

        * khtml/html/html_formimpl.cpp:
        (DOM::HTMLInputElementImpl::HTMLInputElementImpl): Start m_valueMatchesRenderer as false.
        (DOM::HTMLInputElementImpl::parseHTMLAttribute): Set m_valueMatchesRenderer to false when a
        new value is set here.
        (DOM::HTMLInputElementImpl::setValue): Set m_valueMatchesRenderer to false when a new value
        is set here.
        (DOM::HTMLInputElementImpl::setValueFromRenderer): Added. Sets m_value, sets m_valueMatchesRenderer
        to true, and also sends out the input event. It's better to have this here than in the renderer code.
        (DOM::HTMLTextAreaElementImpl::HTMLTextAreaElementImpl): Start m_valueIsValid as false (replaces
        m_dirtyvalue) and m_valueMatchesRenderer as false.
        (DOM::HTMLTextAreaElementImpl::updateValue): Added. Factored this out from the value function. Uses
        the new booleans and keeps them up to date, specifically setting m_valueMatchesRenderer based on
        where the value came from.
        (DOM::HTMLTextAreaElementImpl::value): Updated to call updateValue to do most of the work.
        (DOM::HTMLTextAreaElementImpl::setValue): Set both m_valueIsValid and m_valueMatchesRenderer.
        (DOM::HTMLTextAreaElementImpl::setDefaultValue): Take parameter by reference.

        * khtml/html/html_formimpl.h: Added setValueFromRenderer, valueMatchesRenderer, setValueMatchesRenderer,
        and m_valueMatchesRenderer to input element. For textarea element, made some parameters pass DOMString
        by reference, and added invalidateValue, updateValue, valueMatchesRenderer, and setValueMatchesRenderer.

        * khtml/rendering/render_form.cpp:
        (RenderLineEdit::updateFromElement): Don't re-get the value from the DOM if valueMatchesRenderer
        is true.
        (RenderLineEdit::slotTextChanged): Call setValueFromRenderer instead of manipulating the DOM
        directly.
        (RenderTextArea::detach): Call updateValue instead of calling value for its side effect.
        (RenderTextArea::handleFocusOut): Ditto.
        (RenderTextArea::updateFromElement): Call updateValue and then not re-get the value from the
        DOM if valueMatchesRenderer is true.
        (RenderTextArea::slotTextChanged): Call invalidateValue instead of directly setting m_dirtyvalue to true.

2005-02-25  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4023566> Stickies: Crash in ReplacementFragment::insertFragmentForTestRendering on paste

        Reviewed by darin.

        * khtml/editing/htmlediting.cpp:
        (khtml::ApplyStyleCommand::addInlineStyleIfNeeded):
        (khtml::ReplacementFragment::insertFragmentForTestRendering):
        (khtml::floatRefdElement):
        (khtml::createDefaultParagraphElement):
        (khtml::createBlockPlaceholderElement):
        (khtml::createFontElement):
        (khtml::createStyleSpanElement):
        * khtml/editing/htmlediting.h:

2005-02-25  Darin Adler  <darin@apple.com>

        * kwq/character-sets.txt: Checked in updated file. This new file has no effect, because none of the
        changes affect character sets that we support, but it's good to have the latest file in here, and
        completely safe because I checked that the generated files have not changed.

2005-02-25  Vicki Murley <vicki@apple.com>

        Reviewed by Darin.
 
        - back out this change, since it causes a 3.5% performance regression

    2005-02-23  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - fixed <rdar://problem/4011405> REGRESSION (180-181): Unconfirmed text disappears when text focus moves

        The key was to change things around so that we don't push text from the DOM to the widget
        unless the DOM has actually been changed. This prevents the code path that wipes out inline input
        during the blur process.

        * khtml/html/html_formimpl.cpp:
        (DOM::HTMLInputElementImpl::HTMLInputElementImpl): Start m_valueMatchesRenderer as false.
        (DOM::HTMLInputElementImpl::parseHTMLAttribute): Set m_valueMatchesRenderer to false when a
        new value is set here.
        (DOM::HTMLInputElementImpl::setValue): Set m_valueMatchesRenderer to false when a new value
        is set here.
        (DOM::HTMLInputElementImpl::setValueFromRenderer): Added. Sets m_value, sets m_valueMatchesRenderer
        to true, and also sends out the input event. It's better to have this here than in the renderer code.
        (DOM::HTMLTextAreaElementImpl::HTMLTextAreaElementImpl): Start m_valueIsValid as false (replaces
        m_dirtyvalue) and m_valueMatchesRenderer as false.
        (DOM::HTMLTextAreaElementImpl::updateValue): Added. Factored this out from the value function. Uses
        the new booleans and keeps them up to date, specifically setting m_valueMatchesRenderer based on
        where the value came from.
        (DOM::HTMLTextAreaElementImpl::value): Updated to call updateValue to do most of the work.
        (DOM::HTMLTextAreaElementImpl::setValue): Set both m_valueIsValid and m_valueMatchesRenderer.
        (DOM::HTMLTextAreaElementImpl::setDefaultValue): Take parameter by reference.

        * khtml/html/html_formimpl.h: Added setValueFromRenderer, valueMatchesRenderer, setValueMatchesRenderer,
        and m_valueMatchesRenderer to input element. For textarea element, made some parameters pass DOMString
        by reference, and added invalidateValue, updateValue, valueMatchesRenderer, and setValueMatchesRenderer.

        * khtml/rendering/render_form.cpp:
        (RenderLineEdit::updateFromElement): Don't re-get the value from the DOM if valueMatchesRenderer
        is true.
        (RenderLineEdit::slotTextChanged): Call setValueFromRenderer instead of manipulating the DOM
        directly.
        (RenderTextArea::detach): Call updateValue instead of calling value for its side effect.
        (RenderTextArea::handleFocusOut): Ditto.
        (RenderTextArea::updateFromElement): Call updateValue and then not re-get the value from the
        DOM if valueMatchesRenderer is true.
        (RenderTextArea::slotTextChanged): Call invalidateValue instead of directly setting m_dirtyvalue to true.

2005-02-25  Darin Adler  <darin@apple.com>

        Reviewed by Chris.

        - fixed <rdar://problem/4024786> REGRESSION (Mail): "Smart paste" plain-text word into Blot leaves insertion point misplaced

        * khtml/editing/htmlediting.cpp: (khtml::ReplaceSelectionCommand::doApply):
        Use the insertNodeAfterAndUpdateNodesInserted and insertNodeBeforeAndUpdateNodesInserted
        functions to add the leading and trailing spaces for smart paste.

2005-02-25  David Hyatt  <hyatt@apple.com>

        Back out fix for 3382926, since it breaks LTR text inside RTL contexts.
        
        Reviewed by john

        * khtml/rendering/bidi.cpp:
        (khtml::BidiIterator::direction):

2005-02-25  David Hyatt  <hyatt@apple.com>

        Fix for 3975039, scrolling is slow in huge RSS views.  Optimize the calculation of clip rects for overflow:hidden
        layers.  Also optimize layer movement when scrolling overflow sections.
        
        Reviewed by darin

        * khtml/rendering/render_layer.cpp:
        (throw):
        (ClipRects::operator delete):
        (ClipRects::detach):
        (RenderLayer::RenderLayer):
        (RenderLayer::~RenderLayer):
        (RenderLayer::updateLayerPosition):
        (RenderLayer::removeOnlyThisLayer):
        (RenderLayer::insertOnlyThisLayer):
        (RenderLayer::scrollToOffset):
        (RenderLayer::hitTest):
        (RenderLayer::calculateClipRects):
        (RenderLayer::calculateRects):
        (RenderLayer::containsPoint):
        (RenderLayer::clearClipRects):
        (RenderLayer::clearClipRect):
        * khtml/rendering/render_layer.h:
        (khtml::ClipRects::m_refCnt):
        (khtml::ClipRects::overflowClipRect):
        (khtml::ClipRects::fixedClipRect):
        (khtml::ClipRects::posClipRect):
        (khtml::ClipRects::ref):
        (khtml::ClipRects::deref):
        (khtml::RenderLayer::clipRects):
        * khtml/rendering/render_object.cpp:
        (RenderObject::setStyle):

2005-02-24  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Darin and Dave a while ago.

        <rdar://problem/3996685> REGRESSION: Crash in KWQVectorImpl::at loading http://maps.google.com/mapfiles/homepanel.xsl
        
        * kwq/WebCoreBridge.mm:
        (formElementFromDOMElement): Check for isHTMLElement() as well as
        id() == ID_FORM. This seems like an impossible situation, but
        papering over it seems more expedient for the time being.

2005-02-24  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/3382926> Bidi neutrals at RTL/LTR boundaries not handled correctly.

        Use mirror characters correctly when rendering with RTL directionality.

        Reviewed by Hyatt.

        * khtml/rendering/bidi.cpp:
        (khtml::BidiIterator::direction):

2005-02-24  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/3985889> REGRESSION (125-180): setting <img> src to GIF that already animated does not animate; just shows final frame
        
        Reviewed by Hyatt.

        * WebCore.pbproj/project.pbxproj:
        * khtml/html/html_imageimpl.cpp:
        (HTMLImageLoader::updateFromElement):
        * khtml/rendering/render_image.cpp:
        (RenderImage::resetAnimation):
        * khtml/rendering/render_image.h:
        * khtml/rendering/render_list.cpp:
        (RenderListMarker::setStyle):
        (RenderListMarker::paint):
        * khtml/rendering/render_list.h:
        * kwq/KWQPixmap.h:
        * kwq/KWQPixmap.mm:
        (QPixmap::QPixmap):
        (QPixmap::resetAnimation):
        * kwq/WebCoreImageRenderer.h:

2005-02-24  Chris Blumenberg  <cblu@apple.com>

        Fixed: <rdar://problem/4020110> Safari crashes in setAllData while taking a www.zoomerang.com survey

        Reviewed by rjw.

        * kwq/KWQResourceLoader.mm:
        (-[KWQResourceLoader finishJobAndHandle:]): clear the job after we've deleted to avoid reentrancy

2005-02-24  Darin Adler  <darin@apple.com>

        Reviewed by Vicki.

        - fixed <rdar://problem/4023360> REGRESSION (186-187): image file upload is broken at pep.apple.com

        * khtml/html/html_formimpl.cpp: (DOM::HTMLInputElementImpl::appendFormData): Return true after
        setting up the form data for an uploaded file. The old code would fall through to the ISINDEX
        case and send double form data (the filename instead of the file contents the second time).

2005-02-24  David Harrison  <harrison@apple.com>

        Reviewed by Ken.
        
        <rdar://problem/3990849> AX: textMarkerRange for an AXUIElement within an AXWebArea

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject textMarkerRange]):
        (-[KWQAccObject accessibilityParameterizedAttributeNames]):
        (-[KWQAccObject doAXTextMarkerRangeForUIElement:]):
        (-[KWQAccObject accessibilityAttributeValue:forParameter:]):
        Added AXTextMarkerRangeForUIElement, or you can send textMarkerRange to the UIElement itself.

2005-02-24  David Harrison  <harrison@apple.com>

        Reviewed by Vicki.

        <rdar://problem/4004279> 3 AXSelectedTextChanged notifications are firing each time I type a character

        * khtml/xml/dom_docimpl.cpp:
        (DocumentImpl::updateSelection):
        Send notification only if the selection is not null.  This safely ignores transitory selections set during editing.

2005-02-24  Darin Adler  <darin@apple.com>

        Reviewed by John.

        - fixed <rdar://problem/3987619> in some cases, text doesn't resize with Format->Style->Bigger/Smaller

        * khtml/editing/htmlediting.cpp:
        (khtml::ApplyStyleCommand::applyRelativeFontStyleChange): Only call nodeFullySelected for non-text nodes.
        Text nodes are already split so they're either in the range and full selected or out of the range.
        And nodeFullySelected doesn't work for text nodes.
        (khtml::ApplyStyleCommand::nodeFullySelected): Add an assertion, since this function only works for elements,
        not text nodes.
        (khtml::ApplyStyleCommand::nodeFullyUnselected): Ditto.

        - make big improvement in <rdar://problem/3953636> Mail hung for ~10sec changing font of 84328 characters: khtml::ApplyStyleCommand::nodeFullySelected

        * khtml/xml/dom2_rangeimpl.cpp: (DOM::RangeImpl::compareBoundaryPoints): Improve algorithm based on suggestion
        from Nate Begeman.

        - fixed <rdar://problem/4020305> REGRESSION (185-186): loading image in new window using document.write fails

        * khtml/xml/dom_docimpl.cpp: (DocumentImpl::open): If there is no parent document, don't blow away the base URL.

        - fixed <rdar://problem/4021701> REGRESSION (188-188+): form not submitted after pressing <return> at http://hrweb.apple.com
        
        * khtml/xml/dom_nodeimpl.cpp: (NodeImpl::dispatchMouseEvent): Send activate event in the case where the event
        sent is a KHTML_CLICK_EVENT, not CLICK_EVENT.

2005-02-23  Kevin Decker  <kdecker@apple.com>

        Reviewed by mjs  
        
        Fixed <rdar://problem/4020747> REGRESSION: stray </applet> tags crash Safari

        * khtml/html/htmlparser.cpp:
        (KHTMLParser::processCloseTag):  Made a typesafe check that prevents crashes
        whenever there is a closing applet tag without an actual <applet> in the first place. 
        Now verifies the current token is ID_APPLET before casting it to an HTMLAppletElementImpl.
         

2005-02-23  Ken Kocienda  <kocienda@apple.com>

        Reviewed by John

        Fix for this bug:
        
        <rdar://problem/3977962> font loses bold style after pasting next to existing text and pressing return

        * khtml/editing/htmlediting.cpp:
        (khtml::InsertParagraphSeparatorCommand::calculateStyleBeforeInsertion): Problem here was
        that we were doing work in cases where we should not, and content whose style would have
        been correct if we had done nothing was getting clobbered. It turns out that extra work
        to apply style to the new paragraph added in this command only needs to be done if we're 
        at the boundaries of a paragraph. Otherwise, content that is moved as part of the work 
        of the command will lend their styles to the new paragraph without any extra work needed.
        So, make this position check and return unless at a paragraph boundary.

        New test:

        * layout-tests/editing/style/block-styles-007-expected.txt
        * layout-tests/editing/style/block-styles-007.html

2005-02-23  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Hyatt

        Fix for this bug:
        
        <rdar://problem/4017641> REGRESSION (Mail): you can only bold/unbold a selection starting from end of line once

        Problem is with the way we figure out whether to add or remove a style based on
        the current selection. In this case, the code is looking at the end of the
        previous line, which is not bold, and deduces incorrectly that the operation is
        a "make bold". Then the style code runs to make bold, but there is nothing on
        the end of the previous line to embolden, so we get into a cycle where the same
        thing happens each time cmd-b is hit.

        * khtml/khtml_part.cpp:
        (KHTMLPart::selectionComputedStyle):  Call editingStartPosition() to get the right position for the font determination.
        * khtml/xml/dom2_rangeimpl.cpp:
        (DOM::RangeImpl::editingStartPosition): New helper function that "does the right thing" based on whether the
        selection is a caret or a range, moving upstream for the former, and downstream for the latter.
        * khtml/xml/dom2_rangeimpl.h:
        * kwq/KWQKHTMLPart.mm:
        (KWQKHTMLPart::fontForSelection): Call editingStartPosition() to get the right position for the font determination.

        New tests:

        * layout-tests/editing/style/style-boundary-001-expected.txt
        * layout-tests/editing/style/style-boundary-001.html
        * layout-tests/editing/style/style-boundary-002-expected.txt
        * layout-tests/editing/style/style-boundary-002.html
        * layout-tests/editing/style/style-boundary-003-expected.txt
        * layout-tests/editing/style/style-boundary-003.html
        * layout-tests/editing/style/style-boundary-004-expected.txt
        * layout-tests/editing/style/style-boundary-004.html

2005-02-23  Richard Williamson   <rjw@apple.com>

        Fixed <rdar://problem/3985579> 8A367: Dashboard: Stock widget not visible when click remove to remove single char ticker symbol

        Explicitly remove scroll bar views when removing them from 
        the overflow.

        Don't paint synchronously when the scroll position changes,
        this caused funky clip problems.

        Reviewed by Hyatt.

        * khtml/rendering/render_layer.cpp:
        (RenderLayer::scrollToOffset):
        (RenderLayer::setHasHorizontalScrollbar):
        (RenderLayer::setHasVerticalScrollbar):
        (RenderLayer::updateScrollInfoAfterLayout):

2005-02-23  Ken Kocienda  <kocienda@apple.com>

        Reviewed by Hyatt

        Fix for this bug:
        
        <rdar://problem/3959996> REGRESSION (Mail): cursor moves to beginning of document when click is past end

        * khtml/rendering/render_block.cpp:
        (khtml::RenderBlock::positionForCoordinates): Skip blocks that are invisible or have no height when 
        looking for a child to pass off to. And save away the last visible block with a height to pass off
        to if there is no child at the right y-coordinate.

2005-02-23  David Harrison  <harrison@apple.com>

        Reviewed by Darin.

        <rdar://problem/4010059> BoundsForTextMarkerRange does not update with scrolled web area

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject doAXBoundsForTextMarkerRange:]):
        Adjust for scrolling.

2005-02-22  Maciej Stachowiak  <mjs@apple.com>

        Reviewed by Ken.

        <rdar://problem/3949790> hitting return after pasted styled line results in extra content getting the style
        
        * khtml/editing/htmlediting.cpp:
        (khtml::InsertParagraphSeparatorCommand::doApply): In the case
        where the start block is the root, insert the newly created DIV at
        the end of the root block instead of after the last sibling in the
        start node, since the start node could be inside other
        style-affecting nodes and we don't want to reparent its cousins
        into that.

2005-02-23  David Harrison  <harrison@apple.com>

        Reviewed by Vicki.

        <rdar://problem/4014691> switch to correctly spelled NSAccessibilityForegroundColorTextAttribute constant
        
        Also removed two older, now unneeded, wrappers.

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject visiblePositionForStartOfTextMarkerRange:]):
        (-[KWQAccObject visiblePositionForEndOfTextMarkerRange:]):
        Removed these older, now unneeded, wrappers.
        
        (NSAccessibilityForegroundColorTextAttributeWrapper):
        New wrapper for NSAccessibilityForegroundColorTextAttribute.
        
        (AXAttributeStringSetStyle):
        Use NSAccessibilityForegroundColorTextAttributeWrapper.
        
        (-[KWQAccObject doAXTextMarkerRangeForUnorderedTextMarkers:]):
        Remove uses of visiblePositionForStartOfTextMarkerRange and visiblePositionForEndOfTextMarkerRange.

2005-02-23  David Harrison  <harrison@apple.com>

        Reviewed by Vicki.

        <rdar://problem/3524784> AX hit test doesn't return info when done in empty space of content area

        * kwq/KWQAccObject.mm:
        (-[KWQAccObject accessibilityHitTest:]):
        Return unignored object.

2005-02-23  Darin Adler  <darin@apple.com>

        Reviewed by Ken.

        - fixed <rdar://problem/4006509> REGRESSION (171-172): Setting CSS -khtml-user-modify property triggers crash

        * khtml/css/cssstyleselector.cpp: (khtml::CSSStyleSelector::applyProperty):
        Removed the code that changes the style of the element. This was never needed, and caused the
        style to be modified while we were iterating it.

2005-02-23  Ken Kocienda  <kocienda@apple.com>