The issue was that weren't using the same pivot and the same quaternion math for the rotation.
#jira UE-130496
#brooke.hubert
#preflight 61a83b83ad6629a51eac6537
#ROBOMERGE-AUTHOR: julien.stjean
#ROBOMERGE-SOURCE: CL 18349318 in //UE5/Release-5.0/... via CL 18349329
#ROBOMERGE-BOT: STARSHIP (Release-Engine-Staging -> Release-Engine-Test) (v895-18170469)
[CL 18350076 by julien stjean in ue5-release-engine-test branch]
This represents UE4/Main @17911760, Release-5.0 @17915875 and Dev-PerfTest @17914035
[CL 17918595 by aurel cordonnier in ue5-release-engine-test branch]
This represents UE4/Main @17774255, Release-5.0 @17791557 and Dev-PerfTest @17789485
[CL 17794212 by aurel cordonnier in ue5-release-engine-test branch]
Storing typed elements in UObjects can easily lead to reference leaks if the elements aren't cleared prior to the UObject being left pending GC. This actually made UTypedElementList tricky to use, as you had to remember to manually empty it when you'd finished with it to avoid reference leaks, and we've had several cases now where that was done incorrectly.
To address this issue, we've moved element lists back to being a normal C++ class, FTypedElementList. However, unlike the original version of FTypedElementList (which was itself a UStruct), this version is always heap-allocated and referenced via a TSharedPtr/TSharedRef.
This gives us a nice middle-ground of a well defined lifetime (ie, no lingering references prior to GC) while still being efficient to pass around, including for scripting APIs via FTypedElementListProxy (which just wraps the TSharedPtr in a UStruct).
The downside of this approach is that we need to wrap the FTypedElementList functions that we want to expose to the scripting API (see UTypedElementListLibrary), however that is a far more reasonable burden than requring every user of the typed element framework to know and understand that UTypedElementList had to be manually cleared to avoid potentially hard to find reference leaks (especially if via leaked via scripting APIs).
The core of this change is to TypedElementList.h/.cpp, with TypedElementListFwd.h existing to forward declare the pointer types, and TypedElementListProxy.h and TypedElementListLibrary.h existing to declare the proxy type and wrapped functions used for scripting APIs. TypedElementSelectionInterface.h (and its implementations) provide an example of using FTypedElementListProxy within a scripting API, and the rest of the change is mostly just fallout to transform const UTypedElementList* to FTypedElementListConstRef and UTypedElementList* to FTypedElementListRef.
#rb Brooke.Hubert
#preflight 60d2720c634cd100016c804b
#ROBOMERGE-SOURCE: CL 16776547 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v836-16769935)
[CL 16776582 by jamie dale in ue5-release-engine-test branch]
Previously we relied on element implementations to know what their parent or child elements might be, and to deal with them correctly in the following cases:
1) When "normalising" a selection to be safe for operations like a move or delete, eg) removing elements that are children of other selected elements.
2) When deleting an element, also ensuring that any implicitly destroyed child elements were deselected, eg) deleting an ISM component that has selected static mesh instances.
This approach hurts the modularity of the elements themselves, as it requires that the element implementations have intrinsic knowledge of any other child (or parent) element types that may exist, and to deal with them accordingly when needed. As a concrete example, an ISM component may have child elements in the form of static mesh instance elements, however the generic component element type does not (and should not) directly know that static mesh instance elements exist.
To address this issue, we've added a new interface, UTypedElementHierarchyInterface, which can be used to provide information about the logical parent<->child hierarchy information of elements. This is implemented as follows for our current element types:
- Actor:
- GetParentElement returns an invalid element handle.
- GetChildElements returns the element handles of any components on the actor.
- Component:
- GetParentElement returns its owner actor element handle.
- GetChildElements returns nothing by default, but UActorComponent::GetComponentChildElements may be overridden to control this behavior.
- eg) UInstancedStaticMeshComponent overrides it to return its static mesh instance element handles.
- SMInstance:
- GetParentElement returns its owner ISM component element handle.
- GetChildElements returns nothing.
Now the problems listed above can be solved by using this interface instead of relying on the element implementations:
1) Selection normalization is now handled by UTypedElementSelectionSet, replacing the previous duplicate implementations of UTypedElementCommonActions and UTypedElementViewportInteraction.
2) UTypedElementSelectionSet will now walk and also update the selection state of child elements, when asked to via the FTypedElementSelectionOptions.
Breaking Changes:
- UTypedElementCommonActions::GetElementsForAction and UTypedElementViewportInteraction::GetSelectedElementsToMove have been removed, in favor of using the selection normalization functions of UTypedElementSelectionSet.
- Note: You may still favor using FLevelEditorViewportClient::GetElementsToManipulate, as it will have removed normalized elements that also cannot be moved by a gizmo.
- UTypedElementCommonActions::DeleteElements and UTypedElementCommonActions::DuplicateElements have been removed, as these functions operated on an unnormalized selection and could be unsafe. UTypedElementCommonActions::DeleteNormalizedElements and UTypedElementCommonActions::DuplicateNormalizedElements should be used instead.
- AGroupActor::ForEachActorInGroup and AGroupActor::ForEachMovableActorInGroup now take a second AGroupActor* argument in their callback.
#rb Brooke.Hubert
#preflight 60ca33a678c3b00001e8f5df
#ROBOMERGE-SOURCE: CL 16705229 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v835-16672529)
[CL 16705233 by jamie dale in ue5-release-engine-test branch]
The editor viewport would previously skip the GizmoManipulationStopped call if nothing had actually moved, but this causes an imbalance in the APIs which call NotifyMovementStarted and NotifyMovementEnded.
GizmoManipulationStopped is now always called, but with an extra argument saying if anything actually moved.
#rb Brooke.Hubert
#preflight 60aeb2de1db8a70001cbe077
[CL 16537833 by Jamie Dale in ue5-main branch]
