LIMITATIONS:
1) Re-instancing will only update UClass instance data.
2) Adding and removing properties should only be done towards the end of a class or structure and can not be followed by complex data types.
3) Adding and removing properties from a base class should not be done if a derived class contains complex data types.
KNOWN ISSUES:
1) Changes to enumerations and structures will not be reflected in existing blueprints. However, adding new nodes to the blueprint will show the updated enumeration or structure.
2) If a class contains an enumeration or structure as a member, the class will not be re-instanced if enumeration or structure is changed.
CHANGES:
1) LiveCodingServer
1a) Modified to always execute certain static instances during load.
1b) Modified to exclude the _Statics static structures to avoid patching to old copies.
2) Added support for LiveCoding reinstancing
2a) Refactored deferred registration system for UClass, UEnum, and UScriptStruct to use a common system that works for normal game, hot reload and live coding.
2b) Type specific version check data is possible (i.e. enum doesn't have a size)
2c) Single registration static for UClass
2d) Single registration class for all types that is just a blind forward to API.
2e) Static and dynamic registrations use different API entry points to avoid having overloaded argument lists that just apply to one or the other.
2f) Shims for older API
3) New common "Reload" system to avoid using HotReload code.
3a) Support common delegates regardless of who is reloading/reinstancing.
3b) Re-instancing code moved from HotReload to Kismet2 (where the bulk of the re-instance code already existed).
3c) Modified PyWrapper to use new helper class instead of depending on HotRelaod
3d) Added WITH_RELOAD which is defined if HotReload or LiveCoding is enabled.
3e) Modifed existing code to use new #define and delegates.
Robert did the review on the changes covered by Part 2. Remaining changes are all straightforward.
#rb robert.manuszewski
#jira UE-74493
[CL 15736777 by Tim Smith in ue5-main branch]
- To maintain backwards compatibility an overload for non TCHAR raw string types has also been added which will be converted to the correct format via a temp FString first.
- Fixed cases of Fstring /= char* found in AndroidPlatformFile.cpp and PythonScriptPluginPreload.cpp
- Added unit tests for the /= operator when passing in TCHAR*, ANSICHAR*, FString, FStringView and TStringBuilder in various forms.
- Changed GetNumfor TStringBuilderBase to return int32 (which is what TStringBuilderBase::Len() returns anyway) rather than SIZE_T to prevent compiler warnings about casting types.
#rb Steve.Robb
[CL 15674959 by paul chipchase in ue5-main branch]
This can save 1-2s during Python initialization, by removing repeated calls to FileExists for everything on sys.path.
It does limit Unreal Python modules to being stored within the root of a known Content/Python directory, however this limitation is in-line with all existing modules that ship with the Engine, and allows us to avoid the cost of recursively scanning for modules.
#rb Rex.Hill
#ROBOMERGE-SOURCE: CL 15594550 in //UE5/Release-5.0-EarlyAccess/...
#ROBOMERGE-BOT: STARSHIP (Release-5.0-EarlyAccess -> Main) (v777-15581079)
[CL 15594583 by jamie dale in ue5-main branch]
Slate can shutdown prior to the Python plugin, which means it can clear an array of delegates (which may contain captured Python objects) without taking the Python GIL. TPyAutoGILPtr has been added as a way to capture Python instances that will safely take the GIL during copy, move, or destruction operations.
#rb Patrick.Laflamme
#ROBOMERGE-SOURCE: CL 15366274 in //UE5/Release-5.0-EarlyAccess/...
#ROBOMERGE-BOT: STARSHIP (Release-5.0-EarlyAccess -> Main) (v771-15082668)
[CL 15368939 by jamie dale in ue5-main branch]
- Hooked up console edit box
- Hooked up active tool messages to modes (the mode specific messages need a lot of improvement
- Added status bars to standalone asset editors
#rb louise.rasmussen
[CL 13264057 by Matt Kuhlenschmidt in ue5-main branch]
The internal MakeArray node would default to adding 1 pin, even if the input array of pin names was empty
#jira UE-91948
[FYI] Joji.Tsuruga
#rb none
#ROBOMERGE-SOURCE: CL 12786873 in //UE4/Release-4.25/... via CL 12786876 via CL 12786885
#ROBOMERGE-BOT: RELEASE (Release-Engine-Staging -> Main) (v681-12776863)
[CL 12786906 by jamie dale in Main branch]
We now keep the GIL locked for the duration that the PySelf pointer is being used
#jira
#rb none
#rnx
#ROBOMERGE-SOURCE: CL 12786098 in //UE4/Release-4.25/... via CL 12786105 via CL 12786111
#ROBOMERGE-BOT: RELEASE (Release-Engine-Staging -> Main) (v681-12776863)
[CL 12786839 by jamie dale in Main branch]
#rb jamie.dale
#ROBOMERGE-OWNER: michael.noland
#ROBOMERGE-AUTHOR: michael.noland
#ROBOMERGE-SOURCE: CL 12330683 via CL 12330685 via CL 12330756 via CL 12331679
#ROBOMERGE-BOT: (v670-12295787)
[CL 12331680 by michael noland in Main branch]
This isn't supported by the reflection system and is already disallowed, but the error message was confusing. It now gives an error message more like UHT, and suggests using a struct as an alternative to direct nesting.
#jira
#rb Lauren.Barnes
#ROBOMERGE-SOURCE: CL 12253865 in //UE4/Release-4.25/... via CL 12253868
#ROBOMERGE-BOT: RELEASE (Release-4.25Plus -> Main) (v668-12245121)
[CL 12296563 by jamie dale in Main branch]
