UndoStack will now merge adjacent commands *if they want to be merged*
instead of bundling everything you push onto it until you tell it
to "finalize the combo."
This uses less memory and gives applications full control over how
their undo stacks end up. :^)
When pushing a new command on an undo stack, we will now attempt to
merge it into the stack's current command.
Merging is implemented by overriding the "merge_with(Command const&)"
virtual on GUI::Command. :^)
Renamed the virtual from "on_edit_action" to "will_execute" so it
doesn't clash with our convention for Function hook names.
Also tighten the parameter type to GUI::TextDocumentUndoCommand
since that's the only kind of command it will receive.
This was quite unreliable before. Changes to the undo stack's modified
state are now reflected in the document's modified state, and the
GUI::TextEditor widget has its undo/redo actions updated automatically.
UndoStack is still a bit hard to understand due to the lazy coalescing
of commands, and that's something we should improve upon (e.g with more
explicit, incremental command merging.) But for now, this is a nice
improvement and undo/redo finally behaves in a way that feels natural.
Have TextDocument listen for state changes on the internal undo stack,
and forward those to all clients via a new virtual function.
This simplifies updating the can_undo / can_redo states of TextEditor.
The undo stack was very difficult to understand as it grew by adding
new undo commands to the front of the internal vector. This meant we
had to keep updating indices as the stack grew and shrank.
This patch makes the internal vector grow by appending instead.
Since the `redo` action never goes back to `index: 0`,
we have to mark the clean index as being the current
non-empty index for the undo/redo navigation to work properly.
The problem is that if we never `undo`, the stack index stays at zero,
which is the empty container waiting for commands. In that situation,
if we save the document, it registers the clean index as being 1
(the non-empty index) but because the stack index has never left zero,
the document was being reported as modified, being out of sync with
the window modified state.
Since we keep a stack of command combos, let's call entries on the
stack "Combo" instead of "UndoCommandsContainer".
And since it has a vector of commands, let's call it "commands"
instead of "m_undo_vector".
This patch removes an incorrect way for TextDocument::text_in_range
to return early when the first line of the selection was empty. This
fixes an issue in TextEditor where the status bar showed that 0
characters are selected when the selection started on an empty line.
This also moves Widget::load_from_json into Core::Object as a virtual
function in order to allow loading non-widget objects in GML (e.g.
BoxLayout).
Co-authored-by: Gunnar Beutner <gbeutner@serenityos.org>
This widget provides a scrollable view onto another (child) widget.
If the child is larger than the parent, scrollbars are provided for
panning around the child.
Recent changes in the button painting code made this unnecessary. For
the case of value() == max(), the scrubber button would overlap the
increment button.
Fixes#6838.
Most of the IPC that happens between clients and WindowServer when
creating and configuring windows can be asynchronous. This further
reduces the amount of ping-ponging played during application startup.
Creating a menu/menubar needs to be synchronous because we need the
ID from the response, but adding stuff *to* menus (and adding menus
to menubars, and menubars to windows) can all be asynchronous.
This dramatically reduces the amount of IPC ping-pong played by
each GUI application during startup.
I measured how long it takes TextEditor to enter the main event loop
and it's over 10% faster here. (Down from ~86ms to ~74ms)
This changes client methods so that they return the IPC response's
return value directly - instead of the response struct - for IPC
methods which only have a single return value.
This patches fixes a crash of the Userland/TextEditor where it would
crash when deleting a range spanning two lines. This was because the
TextEditor would delete the range and modify the cursor position
before clearing the selection. This would trigger a status bar update
with the invalid selection.
This allows the painter to be scaled separately in both directions, and
not just in integer intervals. This is crucial for proper SVG viewBox
support.
Most bitmap-related things verify the scale to be one as of now.
This commit unifies methods and method/param names between the above
classes, as well as adds [[nodiscard]] and ALWAYS_INLINE where
appropriate. It also renamed the various move_by methods to
translate_by, as that more closely matches the transformation
terminology.
Until now, this has been hackishly tracked by the TextEditor app's
main widget. Let's do it in GUI::TextDocument instead, so that anyone
who uses this class can know whether it's modified or not.
This state lives in WindowServer and has no local copy in the client
process for now. This may turn out to be a performance issue, and if
it does we can easily cache it.
When a Button has a menu, the AbstractButton behaviour will now not
be used in the mousemove_event. This was already the case for
mousedown_event.
Why only sometimes?
Normally the presence of the menu prevents mousemove_events from being
delivered to the button. But the menu doesn't spawn immediately. So
sometimes mousemove events got through to the AbstractButton after the
menu was told to spawn but before it appeared. This caused the
m_being_pressed field of AbstractButton to be set to true. But there
was never a mouseup_event because the menu got those instead.
