fix(h3): abortar si la malla sigue por encima del tope de cuelgue

[cursor]

El hang-fix de Hunyuan3D (#14) añadió guard_mesh_complexity para no texturizar/exportar mallas de cientos de miles de caras. Esa red de seguridad delegaba en reduce_faces(), que traga cualquier excepción (OOM, topología incompatible, FaceReducer de v2 sobre una malla v2.1) y devuelve la malla original.

Impacto
Quality/PBR (octree 384+) que genera >200k caras: si FaceReducer falla, el worker registra Mesh simplification skipped y sigue a Paint/export con la malla densa. Eso pincha CPU/RAM y bloquea el driver GPU — exactamente la clase de cuelgue que #14 intentaba cerrar.

Cómo reproducir

1. Generar un modelo Quality/PBR con octree alto.
2. reduce_faces lanza (OOM o postprocesador incompatible).
3. El worker no aborta; textura/exporta la malla completa y el host se congela.

Qué cambia

• Tras simplificar, si la malla sigue por encima del tope de cuelgue (200k textura / 400k export), el job falla en vez de continuar.
• El tope de usuario (reduce_face / target_face_num) sigue siendo best-effort: solo se aborta cuando se supera el límite que congela el host.

Validación
pytest tests/test_model3d_hang_guards.py — 8 passed.

  

---

Note

Medium Risk
Changes failure behavior on the critical path before GPU-heavy texturing/export; jobs that previously hung the host will now fail fast, which is intentional but affects high-octree quality runs when simplification fails.

Overview
Hunyuan3D mesh guarding no longer continues into texturing/export when face reduction cannot bring the mesh under the host hang limits (200k for texture, 400k for export).

guard_mesh_complexity now distinguishes the target face count (user reduce_face / target_face_num, capped by the hang limit) from the hang_limit itself. After calling best-effort reduce_faces (which may return the original mesh on error), it raises RuntimeError if the face count is still above hang_limit, with guidance to lower octree resolution instead of freezing the machine.

A new unit test asserts this fail-closed path when reduce_faces returns an unchanged oversized mesh.

^{Reviewed by Cursor Bugbot for commit f4acade. Configure here.}

[github-actions]

PR Review — Loreframe Studio

Risk: low
Scope: 2 file(s); +32/-5; backend services

Automated review from scripts/analyze_pr.py. This is a heuristic pass (no LLM) so humans still own the merge decision.

Findings

• No heuristic issues. Still run the CI checklist below.

Changed files

• modified: app/services/hunyuan3d/worker.py, tests/test_model3d_hang_guards.py

CONTRIBUTING checklist

• python scripts/verify_clean_repo.py
• python -m compileall -q app/services app/launch.py scripts
• cd ui && npm run build if the UI changed
• No weights, CivitAI sidecars, or generated guides
• Stays local-first (no required accounts / telemetry)

Posted by the repo PR review workflow. Re-runs on each push to the PR.

[cursor]

Risk: medium. No apruebo: el check Cursor Bugbot quedó en skipping y no hay una señal Bugbot completa. Asigné revisor para revisión humana.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[cursor]

Risk: medium. Left a non-blocking comment: Cursor Bugbot was present and finished as skipped, so this run does not approve. Human review is needed; a reviewer was assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}