refactor GC scanning code to reflect jl_binding_t are now first class (…

…#51035)

Removes some redundant code.

Credits to Eduardo for pointing this out in the GC meeting.

(cherry picked from commit 5bc558c)

src/gc.c

@@ -2278,19 +2278,10 @@ STATIC_INLINE void gc_mark_excstack(jl_ptls_t ptls, jl_excstack_t *excstack, siz
 }
 
 // Mark module binding
-STATIC_INLINE void gc_mark_module_binding(jl_ptls_t ptls, jl_module_t *mb_parent, jl_binding_t **mb_begin,
- jl_binding_t **mb_end, uintptr_t nptr,
+STATIC_INLINE void gc_mark_module_binding(jl_ptls_t ptls, jl_module_t *mb_parent, uintptr_t nptr,
  uint8_t bits) JL_NOTSAFEPOINT
 {
  jl_gc_markqueue_t *mq = &ptls->mark_queue;
- for (; mb_begin < mb_end; mb_begin++) {
- jl_binding_t *b = *mb_begin;
- if (b == (jl_binding_t *)jl_nothing)
- continue;
- verify_parent1("module", mb_parent, mb_begin, "binding_buff");
- gc_assert_parent_validity((jl_value_t *)mb_parent, (jl_value_t *)b);
- gc_try_claim_and_push(mq, b, &nptr);
- }
  jl_value_t *bindings = (jl_value_t *)jl_atomic_load_relaxed(&mb_parent->bindings);
  gc_assert_parent_validity((jl_value_t *)mb_parent, bindings);
  gc_try_claim_and_push(mq, bindings, &nptr);
@@ -2422,13 +2413,8 @@ FORCE_INLINE void gc_mark_outrefs(jl_ptls_t ptls, jl_gc_markqueue_t *mq, void *_
  else if (foreign_alloc)
  objprofile_count(jl_module_type, bits == GC_OLD_MARKED, sizeof(jl_module_t));
  jl_module_t *mb_parent = (jl_module_t *)new_obj;
- jl_svec_t *bindings = jl_atomic_load_relaxed(&mb_parent->bindings);
- jl_binding_t **table = (jl_binding_t**)jl_svec_data(bindings);
- size_t bsize = jl_svec_len(bindings);
- uintptr_t nptr = ((bsize + mb_parent->usings.len + 1) << 2) | (bits & GC_OLD);
- jl_binding_t **mb_begin = table + 1;
- jl_binding_t **mb_end = table + bsize;
- gc_mark_module_binding(ptls, mb_parent, mb_begin, mb_end, nptr, bits);
+ uintptr_t nptr = ((mb_parent->usings.len + 1) << 2) | (bits & GC_OLD);
+ gc_mark_module_binding(ptls, mb_parent, nptr, bits);
  }
  else if (vtag == jl_task_tag << 4) {
  if (update_meta)