This is a simple mistake, it does not reset `mIsInDesignMode` when it emulates
a "blur".
Additionally, I realized that it does not check whether the focused element is
in the document or not and emulates "focus" of itself it the focused element is
in a different document. In such case, the editor shouldn't have focus, so,
this is a hidden bug, but "blur" notification may occur asynchronously.
Therefore, this could be a bug only in some edge cases.
Differential Revision: https://phabricator.services.mozilla.com/D216907
Since this issue is too obvious, I am not even sure I am doing the right
thing. This looks like it should be discovered years ago....
As far as I can tell from the spec, dragstart and dragend use the same node
as the target, but it's not always the same node in Firefox.
Differential Revision: https://phabricator.services.mozilla.com/D215827
`HTMLEditor` currently checks whether it's in the `designMode` or not with
checking the `Document` node is in the `designMode`. I.e., it checks the
real-time state of the `Document` node. However, if an instance was in the
`designMode`, it needs to finalize `Selection` even after the `Document` node
becomes not in the `designMode`. Additionally, elements in a shadow DOM in the
`designMode` can have `contenteditable` to make it editable. Then, the editable
elements should works as in the `contenteditable` mode rather than in the
`designMode`. Therefore, current `HTMLEditor::IsInDesignMode()` returns
wrong state when the finalization is called asynchronously or when an editing
host in a shadow DOM whose composed document is in the `designMode`.
This patch fixes known cases of these issues with improving the finalization
(and re-initialization with the new editing mode) and the caller side in
`Document`.
Differential Revision: https://phabricator.services.mozilla.com/D216022
`HTMLEditor` currently checks whether it's in the `designMode` or not with
checking the `Document` node is in the `designMode`. I.e., it checks the
real-time state of the `Document` node. However, if an instance was in the
`designMode`, it needs to finalize `Selection` even after the `Document` node
becomes not in the `designMode`. Additionally, elements in a shadow DOM in the
`designMode` can have `contenteditable` to make it editable. Then, the editable
elements should works as in the `contenteditable` mode rather than in the
`designMode`. Therefore, current `HTMLEditor::IsInDesignMode()` returns
wrong state when the finalization is called asynchronously or when an editing
host in a shadow DOM whose composed document is in the `designMode`.
This patch fixes known cases of these issues with improving the finalization
(and re-initialization with the new editing mode) and the caller side in
`Document`.
Differential Revision: https://phabricator.services.mozilla.com/D216022
MaybeEditorDeletedSourceNode was defined on nsIDragService but is more appropriately defined on nsIDragSession. This was not an issue previously since those were the same object.
Differential Revision: https://phabricator.services.mozilla.com/D211074
Since we now add the widget to the event in
dispatchDOMEventViaPresShellForTesting, WidgetGUIEvents that are sent in
mochitests via that method need to transform their screen coordinates by
nsIWidget::WidgetToScreenOffset, to mirror the transformation by
BrowserParent::TransformParentToChild that they don't get because they
skip the parent process.
This exposes a bunch of things that were done to work around this bug. They
are cleaned up here.
Differential Revision: https://phabricator.services.mozilla.com/D211068
Updates each client of the nsContentUtils method to get the right drag session -- the one for the widget that is currently the source or target of the drag session.
The change to nsDOMWindowUtils::DispatchDOMEventViaPresShellForTesting() supports the change to WidgetDragEvent::InitDropEffectForTests() and enabled a
large number of test fixes in the next patch.
Differential Revision: https://phabricator.services.mozilla.com/D211067
This patch enables the assertions when running:
* The `nsITextInputProcessor` tests
* The IME composition tests under `widget/`
* Under `editor/libeditor/tests`
* Under `editor/libeditor/crashtests`
* Under testing/web-platform/tests/editing`
* Under testing/web-platform/tests/input-events`
Differential Revision: https://phabricator.services.mozilla.com/D211835
This patch enables the assertions when running:
* The `nsITextInputProcessor` tests
* The IME composition tests under `widget/`
* Under `editor/libeditor/tests`
* Under `editor/libeditor/crashtests`
* Under testing/web-platform/tests/editing`
* Under testing/web-platform/tests/input-events`
Differential Revision: https://phabricator.services.mozilla.com/D211835
`input` event listeners may be async functions and may set the text control
element value after the editor ends dispatching an `input` event. Even in this
case, the `input` event listener should not be called recursively by committing
composition which is caused by setting the value because the value is
overwritten by the setting value which was intended by the web app.
Unfortunately, we cannot check whether the value setter is an async `input`
event listener. Therefore, we need to suppress `input` events even after the
editor ends dispatching `input` event.
On the other hand, we should not suppress `input` event if the value is set
by a `compositionupdate` event listener which runs before the editor starts
handling the latest composition change because once we do that, the app won't
receive `input` event for the composition. Therefore, we should not suppress
the editor starts handling the composition change (including during
`beforeinput` event dispatching, but the `beforeinput` is not cancelable, so,
the result will be odd, therefore, we have no tests for the cases).
Therefore, this patch adds a new method to `TextComposition` to make
`TextControlState::SetValue` can check whether the editor is handling the
latest composition change or after that. So, during composition, setting
value should cause `input` events between `TextComposition` starts dispatching a
`compositionupdate` event and `EditorBase` starts handling `eCompositionChange`
event which is dispatched after `compositionupdate`.
Differential Revision: https://phabricator.services.mozilla.com/D214676
As far as testing on Chrome, they don't dispatch any events during setting
text control value (including `compositionupdate` and `compositionend`).
However, we dispatch `compositionupdate`, `compositionend`, `beforeinput` and
`input` synchronously. Therefore, `input` event listener may run again with
the old value. This may make web apps confused because they may not expect
the nested `input` event listener call which is not caused by user intention
nor a `Document.execCommand` call. Therefore, we should stop dispatching
the nested `beforeinput` and `input` events which are caused by committing
composition for setting value since the value will be updated soon, so,
the input caused by committing composition does not occur actually. However,
`compositionend` event should be fired even in the nested event loop because
the web apps may manage whether they has a composition with a pair of
`compositionstart` and `compositionend` event listeners even though that won't
work on Chrome.
Additionally, the nested `compositionupdate` and `compositionend` event may
cause a `Document.execCommand` call. However, that must be unexpected behavior
for web apps and anyway the value will be overwritten to the new value.
Therefore, let's make `Document::ExecCommand` do nothing in the situation.
Differential Revision: https://phabricator.services.mozilla.com/D213756
While handling an "insertParagraph" command, it's closed at least in the editing
host. Therefore, it's file to use editing host outside `<body>` because it
won't make the tree messy outside the editing host.
Differential Revision: https://phabricator.services.mozilla.com/D213376
nsAutoTArray was renamed to AutoTArray.
nsDataHashtable and nsJSThingHashtable don't exist any more.
nsTHashMap and nsTHashSet now exist.
nsDeque is properly typed now.
Pair is now CompactPair.
Differential Revision: https://phabricator.services.mozilla.com/D213624
Currently, the `<br>` element type -- whether normal `<br>` element or padding
`<br>` element for empty editor or last line -- is managed by the flags of
`nsINode`. Therefore, changing the flag does not cause mutation, so
`IMEContentObserver` cannot observe the type changes. However,
`ContentEventHandler` treats the padding `<br>` elements as invisible.
Therefore, when a `<br>` element becomes a padding one, `IMEContentObserver`
needs to notify IME of atext removed notification, and also when a `<br>`
element becomes a normal one (i.e., visible), `IMEContentObserver` needs to
notify IME of a text added notification.
Therefore, this patch makes `EditorBase` disconnect the `<br>` element
temporarily to make `IMEContentObserver` observable the type change.
Depends on D211698
Differential Revision: https://phabricator.services.mozilla.com/D211699
When 2nd or later line, the method inserts one-line text to start of a `Text`
node following `<br>` which is inserted by the method. Then, splits the `Text`
node to insert another `<br>`. This creates a lot of unnecessary
`SplitNodeTransaction`s and that causes a lot of copying memory operation to
set the data of the right `Text` node.
This patch makes the method creates a `Text` node when inserting a middle line
of inserting text. Therefore, `SplitNodeTransaction` is created at most one
(to split a `Text` node if the caller wants to insert a text middle of it).
Depends on D211697
Differential Revision: https://phabricator.services.mozilla.com/D211698
Currently, the editor of Gecko always unwraps first line of the right child
block after deleting selected range when the range starts in a parent block
and ends in a child block. This behavior is almost same as the other browsers,
but the other browsers deletes only preceding lines of the right child block
(i.e., without unwrapping the first line of the right child block) if the range
starts from start of a preceding line, for example, when deleting
`<div>abc<br>[def<p>g]hi<br>jkl`, Gecko moves "hi" to the parent `<div>`,
but the other browsers keeps it in the child `<p>`.
For emulating this special handling, we need to touch 2 paths.
One is `Backspace` when selection is collapsed at start of the child block. In
this case, only when the preceding line is empty, i.e., there are 2 line breaks
(either `<br>` or `\n` in `white-space: pre-*`), the following break should
be deleted, but the child block should not be touched.
The other is, deleting when selection is not collapsed or `Delete` when
selection is collapsed at immediately before the child block. In the latter
case, `HTMLEditor::HandleDeleteSelection` extends `Selection` using
`nsFrameSelection`. Then, handle it with same path as deleting non-collapsed
range.
The former is handled with `HandleDeleteLineBreak` and
`ComputeRangeToDeleteLineBreak`. The latter is handled with
`HandleDeleteNonCollapsedRange` and `ComputeRangeToDeleteNonCollapsedRange`.
The new handlers use the `ComputeRangeToDelete*`. Therefore, `beforeinput`
reports exactly same range from `getTargetRanges`. However, existing paths
do not use same approach and this patch makes `HandleDeleteNonCollapsedRange`
fall it back to `HandleDeleteNonCollapsedRange`. Therefore, some `if` checks
in `HandleDeleteNonCollapsedRange` are ugly, but I have no better idea to
implement this smarter.
Differential Revision: https://phabricator.services.mozilla.com/D207690
Currently, the editor of Gecko always unwraps first line of the right child
block after deleting selected range when the range starts in a parent block
and ends in a child block. This behavior is almost same as the other browsers,
but the other browsers deletes only preceding lines of the right child block
(i.e., without unwrapping the first line of the right child block) if the range
starts from start of a preceding line, for example, when deleting
`<div>abc<br>[def<p>g]hi<br>jkl`, Gecko moves "hi" to the parent `<div>`,
but the other browsers keeps it in the child `<p>`.
For emulating this special handling, we need to touch 2 paths.
One is `Backspace` when selection is collapsed at start of the child block. In
this case, only when the preceding line is empty, i.e., there are 2 line breaks
(either `<br>` or `\n` in `white-space: pre-*`), the following break should
be deleted, but the child block should not be touched.
The other is, deleting when selection is not collapsed or `Delete` when
selection is collapsed at immediately before the child block. In the latter
case, `HTMLEditor::HandleDeleteSelection` extends `Selection` using
`nsFrameSelection`. Then, handle it with same path as deleting non-collapsed
range.
The former is handled with `HandleDeleteLineBreak` and
`ComputeRangeToDeleteLineBreak`. The latter is handled with
`HandleDeleteNonCollapsedRange` and `ComputeRangeToDeleteNonCollapsedRange`.
The new handlers use the `ComputeRangeToDelete*`. Therefore, `beforeinput`
reports exactly same range from `getTargetRanges`. However, existing paths
do not use same approach and this patch makes `HandleDeleteNonCollapsedRange`
fall it back to `HandleDeleteNonCollapsedRange`. Therefore, some `if` checks
in `HandleDeleteNonCollapsedRange` are ugly, but I have no better idea to
implement this smarter.
Differential Revision: https://phabricator.services.mozilla.com/D207690
`nsDOMAttributeMap::GetAttribute` creates `dom::Attr` so that it's not cheap.
Instead it should access `Element::mAttrs` with the accessors or
`BorrowedAttrInfo`.
Differential Revision: https://phabricator.services.mozilla.com/D208087
It assumes that the range is always starts and ends in different node. This
is true for now, but this will be called with a text node to delete only
preformatted line break. Note that the only caller of it does not need the
text node(s) if it becomes empty. Therefore, this patch makes it remove the
text node in such case.
Note that the test changed in
`input-events-get-target-ranges-deleting-in-list-items.tentative.html` was
wrong and only Firefox passed it because the range description was
`(#text "", 0) - (#text "", 10)` since the text nodes are removed after
deleting the text data of them. Now, they become
`(#text "list-item1", 0) - (#text "list-item2", 10)`.
Depends on D207688
Differential Revision: https://phabricator.services.mozilla.com/D207689
It uses `AsElement()` which always casts itself to `Element*`, however, the
instance may be non-element node if there is no empty parent of `aEmptyContent`.
Fortunately, all callers of this method uses the result as `nsIContent*` to
call `DeleteNodeWithTransaction()`. Therefore, we don't have crash bugs caused
by this.
Depends on D207687
Differential Revision: https://phabricator.services.mozilla.com/D207688
`Point_Deprecated()` is really error-prone since its result meaning is different
whether the scan direction is backward or forward. Therefore, if a caller wants
a point in a text node and the direction is only one of them, we can change it
to use `WSScanResult::PointAtReachedContent()` or
`WSScanResult::PointAfterReachedContent()`.
Depends on D207686
Differential Revision: https://phabricator.services.mozilla.com/D207687
When it's called, it just returns at the reached content node. However, this
does not make sense when it reached a character in the text node.
Depends on D207685
Differential Revision: https://phabricator.services.mozilla.com/D207686
They point the found character point if scanning forward. However, they point
the next character point if scanning backward. Therefore, I don't have any
good idea to name them. Therefore, I rename them to `*_Deprecated()` and the
callers should use better name method later.
Depends on D207684
Differential Revision: https://phabricator.services.mozilla.com/D207685
When `WSScanResult::ReachedCurrentBlockBoundary()` returns `true`, it may have
reached inline editing host rather than editable block. Therefore, the
method name is definitely error-prone. This patch adds new state and all
users of `WSType::CurrentBlockBoundary` keeps checking the new type too for
keeping current behavior.
I think that we should make `WSRunScanner` aware of inline editing host next
to block boundary before removing odd check of
`ReachedInlineEditingHostBoundary()`, `StartsFromInlineEditingHostBoundary()`
and `EndsByInlineEditingHostBoundary()` to avoid unexpected regressions in the
wild.
Depends on D207683
Differential Revision: https://phabricator.services.mozilla.com/D207684
Different from the comment around setter of `WSType::CurrentBlockBoundary`,
neither `HTMLEditUtils::GetNextLeafContentOrNextBlockElement` nor
`HTMLEditUtils::GetPreviousLeafContentOrPreviousBlockElement` returns
`nullptr` when it reaches non-editable element, and the setters always set
`mContent` to `aEditableBlockParentOrTopmostEditableInlineElement`. Therefore,
`mContent` is always an editable element when `ReachedCurrentBlockBoundary()`
returns `true`.
Depends on D207682
Differential Revision: https://phabricator.services.mozilla.com/D207683
Currently, it checks whether `mContent` is `nullptr` or not even if the scanner
reached something. However, this makes the users of this object check whether
it has reasonable content or not and that makes the users messy. Therefore,
the scanning method should guarantee that it's always error if it does not reach
any content.
Depends on D207681
Differential Revision: https://phabricator.services.mozilla.com/D207682
This case should never happen because nobody can check proper editable state
in uncomposed tree unless the uncomposed tree has its own editing host.
Therefore, this patch adds assertions into the scanning methods too.
Differential Revision: https://phabricator.services.mozilla.com/D207681
`while (RefPtr<Attr> attr = attributes->Item(0))` causes a warning in
`nsDOMAttributeMap::IndexedGetter()` because of out of bounds. Additionally,
we should not make a loop with live DOM tree information if the loop updates
the DOM tree. Therefore, this patch makes it collect all attributes first
before touching the DOM tree.
Depends on D207240
Differential Revision: https://phabricator.services.mozilla.com/D207241
Sorry for the massive patch but I found it hard to split without
introducing a bunch of copies around...
This mostly makes necko and DOM agree on which strings to use, which
should result on less copies and conversions.
Differential Revision: https://phabricator.services.mozilla.com/D205601
Although from the maintenance cost point of view, we should not duplicate
`EditorBase::DeleteRangesWithTransaction`, but let's add a wrapper for it
for making the callers simpler.
Differential Revision: https://phabricator.services.mozilla.com/D203854
It and its helper class, `AutoInclusiveAncestorBlockElementsJoiner` work with
multiple ranges, but they handle special cases only with the first range.
That means that deleting content will be different if same structure is selected
with multiple ranges and that must be not expected by the users.
Differential Revision: https://phabricator.services.mozilla.com/D203853
