Official Codebase of 3DCodeVerse
Python
scene assets: one inliner and one GLB stamp — three copies deleted
gaoypeng4f65b13157a4
diff --git a/harness/codeverse/tracks/scene_assets.py b/harness/codeverse/tracks/scene_assets.pyindex ad88118d..e6ad4214 100644--- a/harness/codeverse/tracks/scene_assets.py+++ b/harness/codeverse/tracks/scene_assets.py@@ -43,6 +43,7 @@ from codeverse.tracks.planner import plan as run_plannerfrom codeverse.tracks.prompting import (base_prompt_context,constraints_text,+ current_files,expected_files,language_system_prompt,reference_images,@@ -327,7 +328,7 @@ def _generate_asset(ctx: RunContext, asset: AssetPlan, rel: str, *, language: Lalabel = f"asset_{to_snake(asset.name)}" + ("_retry" if attempt else "")prompt = _asset_prompt(ctx, asset, rel, language=language, files=files)if feedback:- prompt = prompt + "\n\n" + feedback + _current_files_block(ctx.ws, files)+ prompt = prompt + "\n\n" + feedback + "## Current file(s) (rewrite COMPLETELY)\n" + _inline(ctx, files)task = GenerationTask(label=label, prompt=prompt,system=_asset_system(ctx, language),files_hint=files, round=attempt, kind="asset", temperature=0.5, timeout_s=timeout_s,@@ -340,14 +341,11 @@ def _generate_asset(ctx: RunContext, asset: AssetPlan, rel: str, *, language: Lamax_turns=ctx.policy.agent_max_turns, wrapup_turns=ctx.policy.agent_wrapup_turns)-def _current_files_block(ws: Workspace, files: list[str], limit: int = 24_000) -> str:- """The file(s) the ONE feedback repair rewrites, inlined — a hero has parts, not one file."""- out = ["## Current file(s) (rewrite COMPLETELY)"]- for f in files:- body = _read(ws, f, limit)- if body:- out.append(f"--- {f} ---\n```\n{body}\n```")- return "\n".join(out) + "\n"+def _inline(ctx: RunContext, files: list[str]) -> str:+ """The file(s) a single-shot rewrite must return, inlined the way every other prompt+ inlines them (``prompting.current_files``: trimmed to one budget, harness-owned files+ skipped) — a hero has parts, not one file."""+ return "".join(f"--- {rel} ---\n```\n{body}\n```\n" for rel, body in current_files(ctx, files, 24_000).items())# ----------------------------------------------------------------------------- blender asset (hero)@@ -428,8 +426,8 @@ def build_blender_asset(ctx: RunContext, asset: AssetPlan, *, judge: bool) -> Assub_ws.commit("asset built")rel = asset_file(asset)dest = _copy_glb(ctx.ws, Path(chk.glb), rel)- result = AssetResult(name=asset.name, kind=asset.kind, ok=True, path=rel, size_m=chk.size_m, strategy=strategy, notes=notes,- clips=glb_clip_count(dest))+ result = _stamp_glb(AssetResult(name=asset.name, kind=asset.kind, ok=True, path=rel, size_m=chk.size_m, strategy=strategy,+ notes=notes), ctx, dest)if _judge_wanted(ctx, asset, chk, judge=judge):def _render(out_dir: Path):return ctx.services.render_object(dest, out_dir, views=OBJECT_VIEWS_QUICK, width=512, height=512)@@ -441,8 +439,7 @@ def build_blender_asset(ctx: RunContext, asset: AssetPlan, *, judge: bool) -> Asreturn False_copy_glb(ctx.ws, Path(oc.build.glb_path), rel)sub_ws.commit("asset fix")- result.size_m = _measure_size(ctx, dest) or result.size_m- result.clips = glb_clip_count(dest)+ _stamp_glb(result, ctx, dest)return Truedef _snapshot() -> Callable[[], None]:@@ -456,8 +453,7 @@ def build_blender_asset(ctx: RunContext, asset: AssetPlan, *, judge: bool) -> Asif body:(sub_ws.root / f).write_text(body)sub_ws.commit("asset fix reverted (judged worse)")- result.size_m = _measure_size(ctx, dest) or result.size_m- result.clips = glb_clip_count(dest)+ _stamp_glb(result, ctx, dest)return _restore@@ -487,9 +483,7 @@ def hero_plan(sub: RunContext, asset: AssetPlan) -> StaticPlan:raiselog.warning("hero planner failed for %s, using the one-part sheet: %s", asset.name, e)sub.events.emit("asset.plan_failed", asset=asset.name, error=f"{type(e).__name__}: {e}"[:300])- plan = asset_plan(asset)- sub.ws.write_json(sub.ws.plan_path, plan)- return plan+ return asset_plan(asset)sub.events.emit("asset.planned", asset=asset.name, n_parts=len(plan.parts))return plan@@ -632,7 +626,7 @@ def _judge_and_fix(ctx: RunContext, asset: AssetPlan, result: AssetResult, rendereturn resultinstructions = [f"- {i.target}: {i.instruction}" for i in verdict.improvement_plan[:4]]gen = single_shot_ctx(ctx) or ctx- current = "\n\n".join(f"--- {f} ---\n{_read(ctx.ws, f)}" for f in files if _read(ctx.ws, f)) if gen.single_shot else ""+ current = _inline(gen, files) if gen.single_shot else ""prompt = render("tracks/scene_asset.j2", **base_prompt_context(gen, asset_name=asset.name, asset_kind=asset.kind, asset_description=asset.description,asset_size=asset.approx_size_m, asset_file=files[0], asset_files=files, asset_language=language.value,@@ -704,6 +698,13 @@ def _asset_prompt(ctx: RunContext, asset: AssetPlan, rel: str, *, language: Langreturn prompt+def _stamp_glb(result: AssetResult, ctx: RunContext, glb: Path) -> AssetResult:+ """What the shipped GLB says about itself: its measured size and the clips it carries."""+ result.size_m = _measure_size(ctx, glb) or result.size_m+ result.clips = glb_clip_count(glb)+ return result++def glb_clip_count(glb: Path) -> int:"""How many animation clips a GLB carries (its JSON chunk; no loader needed)."""try: