Fix CAMERA_CULL AND-on-same-object regression vs Cycles - #1
Fix CAMERA_CULL AND-on-same-object regression vs Cycles#1liminallyspaced wants to merge 1 commit into
Conversation
Cycles object_cull.cpp ANDs when both object cull flags are set on the same name (keeps nearby off-frustum for reflections). Dual flags on the CAMERA_CULL scatter list would regress the Classroom cull slice of the 41% plate. Split into disjoint sets: payload objects get use_camera_cull only; distance_objects (far tiny via min_camera_distance) get use_distance_cull only. Scene may enable both flags; missing distance RNA stays camera-only. Manifest stays 0.3.5; no store % change.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit:2320888cd9
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if jrnl.set_prop(obj, "cycles.use_camera_cull", True, SPEED_TAG): | ||
| # Never set use_distance_cull on camera-only names (AND regression). |
There was a problem hiding this comment.
When an object already has cycles.use_distance_cull=True (for example from a saved/manual distance-cull setup), this branch still writes cycles.use_camera_cull=True after the scene distance flag has been enabled. That leaves both object flags on a camera-only name, so Cycles uses AND semantics and nearby off-frustum scatter will not be camera-culled—the regression the new partition is meant to avoid. Please clear or skip the opposite per-object flag before tagging camera-only objects, and mirror that protection for distance-only objects.
Useful? React with 👍 / 👎.
| scatter remains eligible. No measured %; store plates stay Classroom 41% / | ||
| loft 52%. |
There was a problem hiding this comment.
This new release note prints the 41%/52% benchmark shorthand without the baseline-to-fast timings and measurement context, and the same shorthand is added in the research docs. The repository guidance permits benchmark numbers only as measured pairs, so rendered or copied notes no longer carry the required proof context; either omit these percentages here or spell out the full measured pairs.
AGENTS.md reference: AGENTS.md:L52-L58
Useful? React with 👍 / 👎.
Summary
object.cycles.use_camera_cullandobject.cycles.use_distance_cullon the same CAMERA_CULL scatter names and claimed OR semantics.intern/cycles/blender/object_cull.cppANDs when both object flags are on (keeps nearby off-frustum for reflections). Dual flags on the Classroom scatter list would regress the cull slice of the 41% plate.CAMERA_CULL): scene enables simplify + camera + distance cull;payload["objects"]get camera-only; disjointpayload["distance_objects"](far tiny viamin_camera_distance≥ margin) get distance-only. Missing distance RNA → camera path alone.Test plan
python3 tests/test_speed_solver.py— ALL TESTS PASSEDuse_distance_cull; distance list never getsuse_camera_cull; scene both flags on; missing RNA OKblender_manifest.tomlstill0.3.5(not bumped)Authoritative report:
docs/research/PLUGWALK-2026-08-30-14.md§1.