PCG Surface Sampler generates 0 points on runtime-created geometry (5.7.3)
First let me do this as an introduction to me. I’m closing in on 70 fast and have been doing D&D in one form or another for since my early 20’s. Over time discussion and books anime all started pointing too whether we could do a full dive gaming experience. Now we have AI’s that can get us their part way. I have found that the best guessing puts it 20 years out… why too late for me to experience. So for the Past year i have been working on having the AI do the coding for me and i do the directing. There is a lot i don’t know and will never know but ive made some head way. first by doing a modular D&D package. it took time and i learned a lot from which AI was good at what. Copilot i found was good for window dressing and general question but coding was a huge error fest. Grok and SuperGrok are so social media that it was like talking to an influencer. UGH. finally i found Claude coding to be my main creator now. Im using Opus 5 High, and on the pro plan. Im retired and unable to afford the $100 a month for max. You also have to be careful on wording or those tokens disappear fast. I finally had to setup a -RAG and have all the doc’s i could find placed in it and scoured the internet for anything on unreal 5 that would help. Im only doing free cause of the money constraints. I have got to a point in the city building that seems to be just redundant and would like the community to look at this and see if you can help me guide Claude in the right direction. Prompts would help.
Adding numbers to what looks like a known regression. Mythralen builds its entire
world at runtime in C++ - no Landscape, no placed actors, terrain and city are meshes
created each session.
SYMPTOM
World Ray Hit Query → Surface Sampler → Static Mesh Spawner, component hosted on a
runtime-spawned actor:
SurfaceSampler_3 Generated 0 points in 122,500 cells
SurfaceSampler_13 Generated 0 points in 653,800 cells
776,300 cells, zero hits, no error and no warning. Only visible under
-LogCmds=“LogPCG Verbose”. Bounds are valid and logged: 280,000 x 280,000 x 40,000 uu.
THE CONTRADICTION
Same point, same frame, same channel, z=21,200 straight down:
LineTraceSingleByChannel HIT MythralenCityLevelBuilder_0 at z=2948
LineTraceSingleByObjectType HIT MythralenCityLevelBuilder_0 at z=2948
PCG (SweepMultiByObjectType) 0 points
BOTH HOSTS FAIL AT RUNTIME, DIFFERENTLY
component on a spawned actor bounds correct, samplers run, ray query returns nothing
spawned APCGVolume bounds measure 0 x 0 x 0 uu, samplers never run at all
The volume case has a clear cause: a volume takes its extent from its BRUSH, and
brushes are editor-only, so a runtime-spawned volume has no bounds. Attaching a mesh
fixes GetActorBounds but not what PCG reads. So the “use a volume instead” workaround
from the threads below isn’t available to a runtime-built world.
ELIMINATED BY MEASUREMENT
bounds tag filter (control: NONE) trace_complex ray_origin
point cap Self actor selector generation timing override_default_params
PCG cache (pcg.Cache.Runtime.Enabled 0 - no change)
object type components were WorldDynamic; now explicitly ECC_WorldStatic, which made
the object-type trace above start hitting. PCG still returns 0.
QUESTION
Is World Ray Hit Query + Surface Sampler supported against geometry created at runtime
with an actor-hosted PCG component? These look like the same issue and none has an answer:
/t/some-get-data-node-for-pcg-graphs-does-not-work-within-actors/2714078
/t/pcg-issue-runtime-only-works-with-landscape-input-not-actor/1231716
one of which reports it working in 5.2 and breaking in 5.3.
FOR CONTEXT
The hand-written C++ scatter it would replace works fine on the same surfaces:
8,365 trees, 77.7 fps, 2,139 draws, 1.61M tris, 3060 Ti.