Issue #1: indent/justify* can create non-valid fragments.
When applying a block-level formatting to a text node, Gecko creates a div or
blockquote block around the text node and sets the corresponding "align" or
"style" attribute. This patch checks that the active editing host can contain
such a block-level element.
Issue #2: indent/justify* can modify the active editing host.
On the first child of the editable element, the selection is extended outside of
the active editing host -- which causes a few issues for our test cases.
In this patch, this issue is "solved" by modifying
`nsHTMLEditRules::GetPromotedPoint' for block-level operations.
** About the tests **
Sorry for the long explanation but I prefer to be as sharp as possible when I
have to modify existing unit tests.
This patch raises 34 unit test "failures" which are improvements.
Two test files are concerned and have been modified accordingly:
* test_htmleditor_keyevent_handling
* test_richtext2.html
One test has been clarified (no real modification):
* test_bug414526.html
Of course, a specific unit test has been added, see `test_bug677752.html'.
** editor/libeditor/html/tests/test_htmleditor_keyevent_handling.html **
Outdenting now works properly, which results in 4 `FAIL'.
* 7372 ERROR TEST-UNEXPECTED-FAIL
| non-tabbable HTML editor: Shift+Tab after Tab on UL
- got "<ul><li id=\"target\">ul list item</li></ul>",
expected "<ul><ul><li id=\"target\">ul list item</li></ul></ul>"
* 7379 ERROR TEST-UNEXPECTED-FAIL
| non-tabbable HTML editor: Shift+Tab on UL
- got "ul list item",
expected "<ul><li id=\"target\">ul list item</li></ul>"
* 7415 ERROR TEST-UNEXPECTED-FAIL
| non-tabbable HTML editor: Shift+Tab after Tab on OL -
got "<ol><li id=\"target\">ol list item</li></ol>",
expected "<ol><ol><li id=\"target\">ol list item</li></ol></ol>"
* 7422 ERROR TEST-UNEXPECTED-FAIL
| non-tabbable HTML editor: Shfit+Tab on OL
- got "ol list item",
expected "<ol><li id=\"target\">ol list item</li></ol>"
** editor/libeditor/html/tests/browserscope/test_richtext2.html **
The 15 tests that now pass result in 15 `FAIL' and 15 `UNEXPECTED_PASS'.
Here's an overview of what we had before the patch:
* Section A - Apply Formatting Tests: +10 points
before patch: 21/31 (Selection: 9/31)
after patch: 28/31 (Selection: 12/31)
FB:BQ_TEXT-1_SI EXECUTION EXCEPTION
FB:BQ_TEXT-1_SI EXECUTION EXCEPTION
FB:BQ_BR.BR-1_SM EXECUTION EXCEPTION
FB:BQ_BR.BR-1_SM EXECUTION EXCEPTION
IND_TEXT-1_SI EXECUTION EXCEPTION
IND_TEXT-1_SI EXECUTION EXCEPTION
JC_TEXT-1_SC editing host is modified
JF_TEXT-1_SC editing host is modified
JL_TEXT-1_SC editing host is modified
JR_TEXT-1_SC editing host is modified
* Section AC - Apply Formatting Tests, using styleWithCSS: +5 points
before patch: 7/18 (Selection: 5/18)
after patch: 12/18 (Selection: 5/18)
IND_TEXT-1_SI editing host is modified
JC_TEXT-1_SC editing host is modified
JF_TEXT-1_SC editing host is modified
JL_TEXT-1_SC editing host is modified
JR_TEXT-1_SC editing host is modified
** editor/libeditor/html/tests/test_bug414526.html **
This test has been clarified to get more explicit report messages -- the test
themselves haven't been changed. A `todo_is' test has been added.
This test is the one that shows that `IsNodeInActiveEditor' can't be modified,
and that limiting the range promotion for block-level operations is preferrable.
Given the way GetPromotedPoint / IsNodeInActiveEditor are designed, when the
selection is in the first child element of the active editing host the
selection is extended outside of editor node.
As a result, `GetNodesFromSelection' (which is called at the beginning of
`WillAlign') returns two nodes:
1. an empty text node, which is enclosed in a <div style="text-align: ***">;
2. the first child element, which is destroyed when it's a <div>;
Proposed solution:
#1 can be avoided by checking that the node is editable in `WillAlign';
#2 can be avoided by not dropping <div> blocks in `RemoveAlignment'.
This is an opportunity to simplify `RemoveAlignment' a bit.
In design mode, pressing [Return] twice in a list (ol, ul, dl) splits the list
and inserts a paragraph. When the list is the active editing host, it should not
be split.
In some situations, pressing Enter in a contentEditable node fails:
* in an editable paragraph, list item or header, pressing Enter
creates another editable block;
* in an inline editable node, pressing Enter does nothing.
That's because nsHTMLEditRules::WillInsertBreak doesn't check for the active
editing host. This patch should fix it.
When a selection in an inline element is to be deleted, it is first expanded to
the parent block in order not to leave any empty inline element. This patch
ensures that the selection cannot be expanded outside of the contentEditable
element (active editing host).
In editable elements, create a paragraph instead of a <br> node
when [Return] is pressed:
* once in a header node (<h[1..6]>);
* twice in a list item node (<li>).
The main fix here involves the introduction of the nsHTMLEditor::IsNodeInActiveEditor
API. This API returns true if the current selection in the document falls
inside the active editable section. The patch adds a bunch of checks using
this API to various places in the code to make sure that editing operations do
not escape their editing host.
There are also some unrelated fixes to range promotion code which fix the
bugs that have existed in that code before, but because of the lack of this
check were not uncovered before.
This patch also removes the nsTextEditUtils::InBody API which only made sense
in designMode documents, and was insufficient for contenteditable elements.
This should prevent crashes, update view count and edit action nesting count mismatches, and more!
--HG--
extra : rebase_source : af1397b3c8de19d936ca70690a6f7ed284cc5d00
