nsTextEditorState::GetValue() uses mCachedValue only when the value isn't empty
string. However, with SetIsVoid() and IsVoid() of nsAString, nsTextEditorState
can cache empty value only with mCachedValue.
MozReview-Commit-ID: AmQDquEn9M8
Currently, nsTextEditorState caches the value of TextEditor only when it's for
a multi-line text control, i.e., <textarea>. However, using it for single-line
text control improves the score of attachment 8848015. Although this might
increase the cost of creating and registering mutation observer, but it may not
make damage to performance of loading pages since editor for each element won't
be initialized until each element gets focus.
MozReview-Commit-ID: DnjEJ5AUh3M
nsTextEditorState::GetValue() refers nsITextControlElement::IsPlainTextControl()
via nsTextEditorState::IsPlainTextEditor(). However, it always returns true and
virtual call with QI. So, we should get rid of these unnecessary methods.
MozReview-Commit-ID: 3gHdGrzlys4
When setting value of <input type="text">, nsTextEditorState removes all
ranges of normal selection first. Then, TextEditor sets the value. Finally,
TextEditor collapses the selection at the end of the text.
In bug 1386471, we got that there are some problems to remove the call of
Selection::RemoveAllRanges() in nsTextEditorState. Therefore, we need another
approach to improve Selection::Collapse().
The approach of this patch is, when removing all ranges from normal selection,
Selection can cache an nsRange instance if there is an instance which is not
referenced from other than the Selection (i.e., it'll be removed when
Selection::Clear() is called). Then, Selection::Collapse() can reuse it. With
this fix, Selection::Collapse() can reduce allocation cost and may reduce some
other cost like adding it to mutation observer.
However, keeping nsRange instance may cause increasing mutation observer's cost
since nsRange will be adjusted its start node/offset and end node/offset with
mutation observer to guarantee that the range is always valid. So, we can
cache such range only when the caller (or its callee) will set selection range
later. Therefore, this patch adds Selection::RemoveAllRangesTemporarily()
and make only nsTextEditorState::SetValue() and
ContentEventHandler::OnSelectionEvent() use it.
MozReview-Commit-ID: FjWrbz4S1ld
nsTextEditorState stores selection controller as
RefPtr<nsTextInputSelectionImpl> mSelCon. However, some methods still use
nsTextInputSelectionImpl::GetSelection(RawSelectionType, nsISelection**) which
is a virtual method overriding nsISelectionController.
So, instead, we should make it use
nsTextInputSelectionImpl::GetSelection(SelectionType).
MozReview-Commit-ID: Cvxa85LegsO
The failure mode in the attached crashtest is an inconsistency in the flattened
tree. Specifically, we null out mVideoControls in an nsVideoFrame, but defer
the UnbindFromTree call on that NAC element, which measn that its mParent still
points to the nsVideoFrame's mContent. Because all this stuff runs off of script
runners, and the anonymous content destroyer is not guaranteed to run before
other potential script runners, we end up running arbitrary script while the
tree mismatch exists. This script calls back into ProcessPendingRestyles, which
causes trouble.
We could build a separate deferral mechanism, but it's not clear that we actually
need to defer the unbind anymore. The deferred unbind was added in bug 489008,
which predated a lot of simplifications in layout/dom interaction.
MozReview-Commit-ID: 1JYAhiXKVJC
nsTextFrame stores text as single byte character array if all characters are
less than U+0100. Although, this saves footprint, but retrieving and modifying
text needs converting cost. Therefore, if it's created for a text node in
<input> or <textarea>, it should store text as char16_t array.
MozReview-Commit-ID: 9Z82rketT7g
Using concrete class rather than interface classes (nsI*Editor) will allow to reduce QI and some virtual calls. Therefore, Editor classes should be used as concrete class as far as possible.
Unfortunately, if classes referring editor are initialized via scriptable interface, we cannot do this because nsI*Editor is still not marked as builtinclass. Therefore, their editor may be implemented by JS. E.g., inline nsIInlineSpellChecker.init() and nsIDocShell.editor. Such remaining cases should be fixed after nsI*Editor classes are marked as builtinclass.
Note that this patch also creates nsIdentifierMapEntry.h which is separated from nsDocument.h because ShadowRoot.h needs the class but exposing nsDocument.h to the global and includes it causes bustage on Linux and Android. Therefore, for fixing the include hell, this patch touches them and ContentChild.cpp.
MozReview-Commit-ID: i6fLWw6Qeo
When editor has focus, input.value setter will call UpdateOverlayTextVisibility via nsTextInputListener::EditAction -> nsTextControlFrame::SetValueChanged at first. But SetValue will call UpdateOverlayTextVisibility again via ValueWasChanged.
So it is unnecessary to call UpdateOverlayTextVisibility on nsTextEditorState::SetValue when we have the editor.
MozReview-Commit-ID: Hw3Bh64Euo6
We get previous input.value twice in HTMLInputElement::SetValue and nsTextEditorState::SetValue when setting input.value. Since nsTextEditorState::GetValue uses DocumentEncoder, it is expensive. So we should use old value as parameter of nsTextEditorState::SetValue if possible.
MozReview-Commit-ID: A1UPfETTVCn
nsIPlaintestEditor.setText still use BeginPlaceHolderTransaction and EndPlaceHolderTransaction. But since input.value setter doesn't create undo transaction, it is unnecessary to save/restore selection via AutoPlaceHolderBatch. So before calling setText, we should reset selection to reduce saving and restoring selection.
Save/Restore selection is ~7% of input.value setter.
MozReview-Commit-ID: 6yBKCtRmkQt
Although we use StartBatchChanges and EndBatchChanges in nsTextEditorState::SetValue, we have a path that EndBatchChanges isn't called. So we should use RAII class to call EndBatchChanges correctly.
MozReview-Commit-ID: 6bjtTT9wItA
See bug 1346723. I would like to turn off undo when not user interaction. Except to Gecko, other browsers don't create undo history by input.value setter.
MozReview-Commit-ID: 9P1eOKTXCXN
When not using eSetValue_BySetUserInput, we should use SetText transaction instead for fast path. For backward compatibility, when input.value setter is by user interaction, I keep original way. Because the original way doesn't replace all text when some string matches.
MozReview-Commit-ID: IDm7Y1NBmaK
Currently, edit commands for native key bindings are stored in widget. This is
stateful and really complicated in content process because it needs to cache
them.
We can make this simpler if we make WidgetKeyboardEvent store edit commands for
the key combination. Then, child process can handle it even if it's delayed
event or it's a nested event.
This patch adds arrays to WidgetKeyboardEvent to store edit commands which are
initialized with nsIWidget::ExecuteNativeKeyBinding() and adds
WidgetKeyboardEvent::ExecuteEditCommands() to execute stored edit commands as
same as nsIWidget::ExecutenativeKeyBinding().
MozReview-Commit-ID: BGRvBrLz5lp
input.value setter removes editor flags and max-length to set value. To clean up code, I would like to use RAII class to set and restore it.
Actually, when the frame is being destroyed by InsertText, it isn't restored. But it might be safe since the flags is set again on nsTextEditorState::PrepareEditor by focus.
MozReview-Commit-ID: J0OYYluWD8z
Actually, we use GetValue to check whether value is empty or not for placeholder. But since GetValue uses TextEditor::OutputToString when on editor, it is expensive. Since editor has DocumentIsEmpty method, we should use it for this case.
MozReview-Commit-ID: rQX8yjnWQz
maxlength will be set by nsTextControlFrame::AttributeChanged via RestyleManager. If element is display:none, RestyleManager won't call Frame's AttributeChanged. So we should always initialize maxlength when setting focus.
Also, wrap attribute for textarea element will be updated by HTMLTextAreaElement even if display:none. So this issue doesn't occur. maxlength might have to be updated by HTMLInputElement. But it is unnecessary to update editor's maxlength on display:none since this is used on focused editor.
MozReview-Commit-ID: JHODOBTv62v
Otherwise we end up calling into editor code and doing a bunch of work even though the value hasn't actually changed, when a value with \r in it is set repeatedly.
