رفتن به محتوا
newblender

Transformation ledger

این محتوا هنوز به زبان شما در دسترس نیست.

Every candidate change that would turn Blender into newblender, with its evidence. The ledger is cross-stage: later stages add evidence to existing entries or open new ones.

  • Status values:
    • proposed: from discovery
    • probing: a runtime test is running
    • accepted / rejected: decided by the owner
  • Evidence levels (lab #9 ladder, simplified):
    • L1: documented and traced in source
    • L2: reproduced at runtime
    • L3: prototyped
    • L4: benchmarked against stock Blender
    • L5: proven in production use
ID Change Evidence Modules touched Level Status Dead fact
L-001 Replace modes with explicit editing sessions. Mesh is canonical; BMesh is opened, edited and flushed explicitly, with no implicit sync on a mode switch. 178/320 modelling ops gated by mode; mode_set dispatches by operator name; stale Mesh in edit mode; conversion cost never removed (S01 03 §5.1, §5.13; 04 Themes 1–2) editors/object/object_modes.cc, object_edit.cc, editors/mesh/editmesh_utils.cc, bmesh/intern/bmesh_mesh_convert.cc L1 proposed DF1
L-002 Selection-as-scope → explicit element sets. Every edit takes indices or an attribute mask; selection becomes an optional, plain attribute. 56 “remove” cards; loop_select exec already index-driven; .select_* already boolean attributes (S01 03 §2; 04 §3) editors/mesh/editmesh_select*.cc, bmop slot inputs L1 proposed DF1
L-003 Return results, not statuses. Every edit returns a diff (created / deleted / changed) plus validity checks. Operators end at DEG_id_tag_update + redraw; EDBM_op_finish only checks for errors (03 §5.12); 03c X2 “operators return only status” · P01: 4 of 21 region-borrowed operators returned FINISHED with real arguments and changed nothing measurable (probe) windowmanager operator return path, editmesh_utils.cc, bmop output slots L2 proposed DF4
L-004 Hidden context defaults → explicit parameters; machine-readable preconditions in place of polls that encode button location. Cursor, view axis, tool settings and local view read inside exec; heuristic wrong for 50%; poll ≠ need (03 §5.5–5.6) · P01: 0/23 region-gated operators pass poll headless, 21/23 pass once any (never-drawn) region is supplied; loop/ring/path pick then work from indices alone (probe) editors/* polls and execs, screen_ops.cc L2 proposed DF2
L-005 Expose the bmop slot schema as a generated, typed command vocabulary (83 bmops) for agent topology edits. Printf-style EDBM_op_init format strings parsed into typed slots (bmesh_opdefines.cc) (03 §5.7; 04 Theme 6) bmesh/intern/bmesh_opdefines.cc, bmesh_operators.cc L1 proposed DF2, DF3
L-006 Declarative-first artefact. The modifier / GN recipe is the primary output, stored and diffed as code. 91% of mesh objects carry modifiers; 68% of geometry ops have twins; nodes-as-code endorsed upstream (02; 03 §3; 03c S04) modifiers/, nodes/, nodebpy L1 proposed DF3
L-007 Converge duplicated cores on Mesh-native geometry::, following upstream (e.g. the two bevels). BM_mesh_bevel and geometry::mesh_bevel coexist; node tools round-trip BMesh↔Mesh (03 §5.9) bmesh/tools/, geometry/intern/ L1 proposed DF3
L-008 Differential transactions and branching from Euler-operator inverses, replacing snapshot undo; agent calls always transactional. Full BMesh→Mesh copy per undo step; Python calls push no undo by default (03 §5.4, §5.11; 04 Themes 3, 6) · P01: 5 sculpt gesture operators segfault in their undo push in a windowless session (probe) editors/mesh/editmesh_undo.cc, bmesh/intern/bmesh_core.cc L1 proposed DF1
L-009 Placement tools re-specified as data: knife → cut path or plane; retopo → target loops; brush → field or 3D path. No mouse shims. 27 gui-event ops, 14 with no twin; node tools already capture mouse/region as RNA props (03 §5.3, §5.14) editmesh_knife.cc, polybuild, paint_stroke.cc, node_group_operator.cc L1 proposed DF1, DF4
L-010 Machine QA gates replace eye checks: quad ratio, manifold, doubles, normals, applied scale, naming; plus an isomorphism compare. Studio checks are mostly visual; 94.5% quads shows the rules are real; mesh_comparison exists (01 §5; 02; 04 §3) new, on blenkernel mesh queries L1 proposed DF4
L-011 Headless boot profile: empty, windowless, deterministic; no default cube, and no CLI-only settings. 03c X2 (bpy module boots the default cube, CLI flags without API); transform needs only window+screen from startup.blend (03 §1.1) · P01: the same 5 crashes show edit code assuming a windowed session (probe) source/creator/, windowmanager/intern/wm_init_exit.cc L1 proposed DF1