gh-146527: Heap-allocate gc_stats to avoid bloating PyInterpreterState (#148057)

The gc_stats struct contains ring buffers of gc_generation_stats
entries (11 young + 3×2 old on default builds). Embedding it inline
in _gc_runtime_state, which is itself inline in PyInterpreterState,
pushed fields like _gil.locked and threads.head to offsets beyond
what out-of-process profilers and debuggers can reasonably read in
a single buffer (e.g. offset 9384 for _gil.locked vs an 8 KiB read
buffer).

Heap-allocate generation_stats via PyMem_RawCalloc in _PyGC_Init and
free it in _PyGC_Fini. This shrinks PyInterpreterState by ~1.6 KiB
and keeps the GIL, thread-list, and other frequently-inspected fields
at stable, low offsets.

Author: pablogsal

Include/internal/pycore_interp_structs.h

@@ -248,7 +248,7 @@ struct _gc_runtime_state {
     struct gc_generation old[2];
     /* a permanent generation which won't be collected */
     struct gc_generation permanent_generation;
-    struct gc_stats generation_stats;
+    struct gc_stats *generation_stats;
     /* true if we are currently running the collector */
     int collecting;
     // The frame that started the current collection. It might be NULL even when

Modules/gcmodule.c

@@ -347,9 +347,9 @@ gc_get_stats_impl(PyObject *module)
     /* To get consistent values despite allocations while constructing
        the result list, we use a snapshot of the running stats. */
     GCState *gcstate = get_gc_state();
-    stats[0] = gcstate->generation_stats.young.items[gcstate->generation_stats.young.index];
-    stats[1] = gcstate->generation_stats.old[0].items[gcstate->generation_stats.old[0].index];
-    stats[2] = gcstate->generation_stats.old[1].items[gcstate->generation_stats.old[1].index];
+    stats[0] = gcstate->generation_stats->young.items[gcstate->generation_stats->young.index];
+    stats[1] = gcstate->generation_stats->old[0].items[gcstate->generation_stats->old[0].index];
+    stats[2] = gcstate->generation_stats->old[1].items[gcstate->generation_stats->old[1].index];
 
     PyObject *result = PyList_New(0);
     if (result == NULL)

Python/gc.c

@@ -177,6 +177,11 @@ _PyGC_Init(PyInterpreterState *interp)
 {
     GCState *gcstate = &interp->gc;
 
+    gcstate->generation_stats = PyMem_RawCalloc(1, sizeof(struct gc_stats));
+    if (gcstate->generation_stats == NULL) {
+        return _PyStatus_NO_MEMORY();
+    }
+
     gcstate->garbage = PyList_New(0);
     if (gcstate->garbage == NULL) {
         return _PyStatus_NO_MEMORY();
@@ -1398,13 +1403,13 @@ static struct gc_generation_stats *
 gc_get_stats(GCState *gcstate, int gen)
 {
     if (gen == 0) {
-        struct gc_young_stats_buffer *buffer = &gcstate->generation_stats.young;
+        struct gc_young_stats_buffer *buffer = &gcstate->generation_stats->young;
         buffer->index = (buffer->index + 1) % GC_YOUNG_STATS_SIZE;
         struct gc_generation_stats *stats = &buffer->items[buffer->index];
         return stats;
     }
     else {
-        struct gc_old_stats_buffer *buffer = &gcstate->generation_stats.old[gen - 1];
+        struct gc_old_stats_buffer *buffer = &gcstate->generation_stats->old[gen - 1];
         buffer->index = (buffer->index + 1) % GC_OLD_STATS_SIZE;
         struct gc_generation_stats *stats = &buffer->items[buffer->index];
         return stats;
@@ -1415,12 +1420,12 @@ static struct gc_generation_stats *
 gc_get_prev_stats(GCState *gcstate, int gen)
 {
     if (gen == 0) {
-        struct gc_young_stats_buffer *buffer = &gcstate->generation_stats.young;
+        struct gc_young_stats_buffer *buffer = &gcstate->generation_stats->young;
         struct gc_generation_stats *stats = &buffer->items[buffer->index];
         return stats;
     }
     else {
-        struct gc_old_stats_buffer *buffer = &gcstate->generation_stats.old[gen - 1];
+        struct gc_old_stats_buffer *buffer = &gcstate->generation_stats->old[gen - 1];
         struct gc_generation_stats *stats = &buffer->items[buffer->index];
         return stats;
     }
@@ -2299,6 +2304,8 @@ _PyGC_Fini(PyInterpreterState *interp)
     GCState *gcstate = &interp->gc;
     Py_CLEAR(gcstate->garbage);
     Py_CLEAR(gcstate->callbacks);
+    PyMem_RawFree(gcstate->generation_stats);
+    gcstate->generation_stats = NULL;
 
     /* Prevent a subtle bug that affects sub-interpreters that use basic
      * single-phase init extensions (m_size == -1).  Those extensions cause objects

Python/gc_free_threading.c

@@ -1698,6 +1698,11 @@ _PyGC_Init(PyInterpreterState *interp)
 {
     GCState *gcstate = &interp->gc;
 
+    gcstate->generation_stats = PyMem_RawCalloc(1, sizeof(struct gc_stats));
+    if (gcstate->generation_stats == NULL) {
+        return _PyStatus_NO_MEMORY();
+    }
+
     gcstate->garbage = PyList_New(0);
     if (gcstate->garbage == NULL) {
         return _PyStatus_NO_MEMORY();
@@ -2387,12 +2392,12 @@ static struct gc_generation_stats *
 get_stats(GCState *gcstate, int gen)
 {
     if (gen == 0) {
-        struct gc_young_stats_buffer *buffer = &gcstate->generation_stats.young;
+        struct gc_young_stats_buffer *buffer = &gcstate->generation_stats->young;
         struct gc_generation_stats *stats = &buffer->items[buffer->index];
         return stats;
     }
     else {
-        struct gc_old_stats_buffer *buffer = &gcstate->generation_stats.old[gen - 1];
+        struct gc_old_stats_buffer *buffer = &gcstate->generation_stats->old[gen - 1];
         struct gc_generation_stats *stats = &buffer->items[buffer->index];
         return stats;
     }
@@ -2831,6 +2836,8 @@ _PyGC_Fini(PyInterpreterState *interp)
     GCState *gcstate = &interp->gc;
     Py_CLEAR(gcstate->garbage);
     Py_CLEAR(gcstate->callbacks);
+    PyMem_RawFree(gcstate->generation_stats);
+    gcstate->generation_stats = NULL;
 
     /* We expect that none of this interpreters objects are shared
        with other interpreters.