Some properties are overwritten by the old CDO on reinstancing about usharp HOT 2 CLOSED

@Algorithman has done some great work looking into this issue. It isn't an issue in UEngine::CopyPropertiesForUnrelatedObjects as described in the original post, but instead an issue in the hotreload default value propagation in UpdateDefaultProperties (#80 also stems from this issue). See the PR he has made for UE4 here https://github.com/EpicGames/UnrealEngine/pull/5830.

There are also issues with reinstancing subobjects in ReplaceObjectHelper (KismetReinstanceUtilities.cpp) which result in non structurally modified subobjects to become corrupt after hotreload (this impacts the C++ hotreload too). This can be tested by defining a component in C++, adding it a blueprint, modifying an exposed property on that component in C++, then hotreloading. It looks like an RF_ArchetypeObject is marked as "pending kill" and the replacement object doesn't work as expected in its place (possibly an object reference isn't being updated?). It should be noted that if you make a structural change, then modify the value in subsequent compiles, it doesn't have the same issue, but there are asserts which fire complaining about archetype object issues (bArchetypeReinstanced). These RF_ArchetypeObject tagged objects are suffixed with _GEN_VARIABLE, none of the other objects create the same issue.

Creating a correct fix for the subobject issue in ReplaceObjectHelper is a bit out of the scope of my knowledge area for how all of this is set up. There are some hacks which should work which involve skipping the RF_ArchetypeObject tagged objects whilst reinstancing, but this will most likely cause issues in other places. I could change USharp to detect a CDO change and force a full recompile if so (this would result in that assert being fired on non-structural changes, but it should work. I'm unsure of the side effects).

Some more information on the subobject issue:

After you have made the structural change, the _GEN_VARIABLE (RF_ArchetypeObject) objects no longer show up in the call to ReplaceObjectHelper. This is possibly related to why the bArchetypeReinstanced assert fires. It also explains why there the object doesn't become corrupt on subsequent compiles (because it's no longer being marked as pending kill).

from usharp.

Fixed in b79d8d4

The ReplaceObjectHelper issue as described above requires engine source modification, see https://github.com/pixeltris/UnrealEngine/tree/hotreload.

from usharp.