This reverts commit 4b817a565c.
Fixes#64988.
Fixes#64997.
This caused several regressions (#64988, #64997,
https://github.com/godotengine/godot/issues/64997#issuecomment-1229970605)
which point at a flaw in the current logic:
- `Control::NOTIFICATION_ENTER_TREE` triggers a *deferred* notification with
`NOTIFCATION_THEME_CHANGED` as introduced in #62845.
- Some classes use their `THEME_CHANGED` to cache theme items in
member variables (e.g. `style_normal`, etc.), and use those member
variables in `ENTER_TREE`, `READY`, `DRAW`, etc. Since the `THEME_CHANGE`
notification is now deferred, they end up accessing invalid state and this
can lead to not applying theme properly (e.g. for EditorHelp) or crashing
(e.g. for EditorLog or CodeEdit).
So we need to go back to the drawing board and see if `THEME_CHANGED` can be
called earlier so that the previous logic still works?
Or can we refactor all engine code to make sure that:
- `ENTER_TREE` and similar do not depend on theme properties cached in member
variables.
- Or `THEME_CHANGE` does trigger a general UI update to make sure that any
bad theme handling in `ENTER_TREE` and co. gets fixed when `THEME_CHANGE`
does arrive for the first time. But that means having a temporary invalid
(and possibly still crashing) state, and doing some computations twice
which might be heavy (e.g. `EditorHelp::_update_doc()`).
When a Graph contains cycles, e.g. 1->2->3->1 _layering would end
up in an infinite loop since IS_SUBSET would never be true.
By keeping check of the size of z, which contains the already layered
nodes, one can detect a freeze (since it should change after
current_layer increases. If it doesn't "u" didn't change and q and u
will never be equal resulting in a freeze/infinite while loop).
If a freeze happens warn the user and put all the nodes part of the
cycle (leftover in p) to the first layer which will end up in them
being listed top to bottom.
This is a virtual method that can be used to add additional error
condition checks while the connection is still being dragged. If true is
returned, the connection is valid. If false is returned, the connection
is invalid and thus not possible (ie. it will not snap). The virtual
method is exposed with an underscore to scripts.
* Map is unnecessary and inefficient in almost every case.
* Replaced by the new HashMap.
* Renamed Map to RBMap and Set to RBSet for cases that still make sense
(order matters) but use is discouraged.
There were very few cases where replacing by HashMap was undesired because
keeping the key order was intended.
I tried to keep those (as RBMap) as much as possible, but might have missed
some. Review appreciated!
Adds a new, cleaned up, HashMap implementation.
* Uses Robin Hood Hashing (https://en.wikipedia.org/wiki/Hash_table#Robin_Hood_hashing).
* Keeps elements in a double linked list for simpler, ordered, iteration.
* Allows keeping iterators for later use in removal (Unlike Map<>, it does not do much
for performance vs keeping the key, but helps replace old code).
* Uses a more modern C++ iterator API, deprecates the old one.
* Supports custom allocator (in case there is a wish to use a paged one).
This class aims to unify all the associative template usage and replace it by this one:
* Map<> (whereas key order does not matter, which is 99% of cases)
* HashMap<>
* OrderedHashMap<>
* OAHashMap<>
This commit adds two signals:
* connection_drag_begun, which is emitted when a connection is started
to be created by the user and
* `connection_drag_ended`, which is emitted when no longer a connection
is created.
Additionally `force_connection_drag_end()` adds the possibility to end
the connection dragging. If called from user code, no other connection
request signals are invoked. This is useful to add `GraphNode`s via
shortcuts while the user is dragging a connection to directly connect
the newly added node.
