gh-142531: Fix free-threaded GC performance regression

If there are many untracked tuples, the GC will run too often, resulting
in poor performance.  The fix is to include untracked tuples in the
"long lived" object count. The number of frozen objects is also now
included since the free-threaded GC must scan those too.

Author: nascheme

Misc/NEWS.d/next/Core_and_Builtins/2025-12-10-23-03-10.gh-issue-142531.NUEa1T.rst

@@ -0,0 +1,5 @@
+Fix a free-threaded GC performance regression.  If there are many untracked
+tuples, the GC will run too often, resulting in poor performance.  The fix
+is to include untracked tuples in the "long lived" object count. The number
+of frozen objects is also now included since the free-threaded GC must
+scan those too.

Python/gc_free_threading.c

@@ -375,6 +375,19 @@ op_from_block(void *block, void *arg, bool include_frozen)
     return op;
 }
 
+// As above but returns untracked and frozen objects as well.
+static PyObject *
+op_from_block_all_gc(void *block, void *arg)
+{
+    struct visitor_args *a = arg;
+    if (block == NULL) {
+        return NULL;
+    }
+    PyObject *op = (PyObject *)((char*)block + a->offset);
+    assert(PyObject_IS_GC(op));
+    return op;
+}
+
 static int
 gc_visit_heaps_lock_held(PyInterpreterState *interp, mi_block_visit_fun *visitor,
                          struct visitor_args *arg)
@@ -1186,12 +1199,20 @@ static bool
 scan_heap_visitor(const mi_heap_t *heap, const mi_heap_area_t *area,
                   void *block, size_t block_size, void *args)
 {
-    PyObject *op = op_from_block(block, args, false);
+    PyObject *op = op_from_block_all_gc(block, args);
     if (op == NULL) {
         return true;
     }
-
     struct collection_state *state = (struct collection_state *)args;
+    // The free-threaded GC cost is proportional to the number of objects in
+    // the mimalloc GC heap and so we should include the counts for untracked
+    // and frozen objects as well.  This is especially important if many
+    // tuples have been untracked.
+    state->long_lived_total++;
+    if (!_PyObject_GC_IS_TRACKED(op) || gc_is_frozen(op)) {
+        return true;
+    }
+
     if (gc_is_unreachable(op)) {
         // Disable deferred refcounting for unreachable objects so that they
         // are collected immediately after finalization.
@@ -1222,7 +1243,6 @@ scan_heap_visitor(const mi_heap_t *heap, const mi_heap_area_t *area,
     // object is reachable, restore `ob_tid`; we're done with these objects
     gc_restore_tid(op);
     gc_clear_alive(op);
-    state->long_lived_total++;
     return true;
 }