Belt: refresh comments that described earlier code

BeltBrim.cpp still described the brim as running inside the parallel
support step; it runs sequentially after it (generate_belt_brim).  The
GCodeWriter, calib.cpp and calib.hpp comments referred to an inheritance
layout and a dynamic_cast that no longer exist.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

src/libslic3r/BeltBrim.cpp

@@ -376,15 +376,12 @@ static void belt_brim_band_paths(const BeltBrimContext      &bc,
 // overhang outside the belt footprint and land in the brim ring, which the flattened
 // brim_object_gap - a belt-plane separation - does not cover.
 //
-// THREADING: this runs inside posSupportMaterial, which Print::process() executes for all
-// objects in a tbb::parallel_for (Print.cpp).  Object slices are finished by then and safe
-// to read across objects, but SUPPORT layers are not: another object's thread may be
-// inside clear_support_layers() - which deletes the SupportLayer pointers - right now, so
-// touching a foreign object's support_layers() here is a use-after-free.  Only this
-// object's own supports are consulted; they are complete, because make_belt_brim() runs at
-// the tail of this object's own generate_support_material().  The cost is that the brim
-// does not dodge a *different* object's support at the same Z, which needs the objects to
-// overlap in the belt direction in the first place.
+// SEQUENCING: this reads the layers and support layers of every object on the plate, so
+// it must not overlap with another object's support step, which rebuilds them.
+// Print::process() therefore generates the belt brims one object after the other once the
+// parallel support step is over (PrintObject::generate_belt_brim()), and an object that
+// arrives on or leaves the plate invalidates every other object's support step
+// (PrintApply.cpp) so the brims are clipped against what is there now.
 // `region_bbox` bounds the brim; anything outside it cannot clip a brim line, so whole
 // objects are skipped without materialising their polygons.  On a typical plate the
 // objects do not overlap and every foreign object drops out here, which matters because

src/libslic3r/GCodeWriter.hpp

@@ -236,8 +236,8 @@ protected:
     Vec3d apply_axis_remap(const Vec3d &pos) const;
 
     // Motion uses the global/base process variant until a filament becomes active.
-    // Protected so subclasses index the per-extruder speed options (travel_speed,
-    // travel_speed_z, initial_layer_travel_speed) exactly as the base writer does.
+    // Indexes the per-extruder speed options (travel_speed, travel_speed_z,
+    // initial_layer_travel_speed).
     size_t     m_cached_extruder_idx;
 
 private:

src/libslic3r/calib.cpp

@@ -896,8 +896,7 @@ void CalibPressureAdvancePattern::_refresh_writer(bool is_bbl_machine, const Mod
         }
         m_writer = std::move(belt_writer);
     } else if (m_writer && dynamic_cast<const BeltKinematics *>(&m_writer->kinematics()) != nullptr) {
-        // Previously configured for a belt printer; drop back to a plain writer,
-        // exactly as the old dynamic_cast<BeltGCodeWriter*> check did.
+        // Previously configured for a belt printer; drop back to a plain writer.
         m_writer = std::make_shared<GCodeWriter>();
     }
 

src/libslic3r/calib.hpp

@@ -370,9 +370,9 @@ private:
 
     const Calib_Params &m_params;
 
-    // Polymorphic so belt printers get belt kinematics in world-coordinates
-    // mode (_refresh_writer); shared_ptr keeps the class copyable — the writer
-    // is rebuilt by refresh_setup() before every use anyway.
+    // Belt printers get belt kinematics installed on it (_refresh_writer);
+    // shared_ptr keeps the class copyable — the writer is rebuilt by
+    // refresh_setup() before every use anyway.
     std::shared_ptr<GCodeWriter> m_writer{std::make_shared<GCodeWriter>()};
     Vec3d              m_starting_point;
     bool               m_is_start_point_fixed = false;