gh-149584: Reuse profiler result objects

Cache the last FrameInfo tuple per code object/instruction offset, reuse cached thread id objects, and append cached parent frames directly on full frame-cache hits. This cuts Python allocation churn in the steady-state profiler path.

Local benchmark: Tools/inspection/benchmark_external_inspection.py basic, 3s quiet mode, cache_frames/all_threads. Previous commit: 897,703 calls/sec, 806,084 samples/sec, 1.114 us/call. After this commit: 1,187,314 calls/sec, 1,047,698 samples/sec, 0.842 us/call. Incremental win: +289,611 calls/sec (+32.3%), -0.272 us/call.

Author: pablogsal

Modules/_remote_debugging/_remote_debugging.h

@@ -215,6 +215,8 @@ typedef struct {
     PyObject *file_name;
     int first_lineno;
     PyObject *linetable;  // bytes
+    PyObject *last_frame_info;
+    ptrdiff_t last_addrq;
     uintptr_t addr_code_adaptive;
 } CachedCodeMetadata;
 
@@ -227,6 +229,7 @@ typedef struct {
     uintptr_t thread_state_addr;
     uintptr_t addrs[FRAME_CACHE_MAX_FRAMES];
     Py_ssize_t num_addrs;
+    PyObject *thread_id_obj;                 // owned reference, NULL if empty
     PyObject *frame_list;                    // owned reference, NULL if empty
 } FrameCacheEntry;
 

Modules/_remote_debugging/code_objects.c

@@ -405,6 +405,8 @@ parse_code_object(RemoteUnwinderObject *unwinder,
         meta->func_name = func;
         meta->file_name = file;
         meta->linetable = linetable;
+        meta->last_frame_info = NULL;
+        meta->last_addrq = -1;
         meta->first_lineno = GET_MEMBER(int, code_object, unwinder->debug_offsets.code_object.firstlineno);
         meta->addr_code_adaptive = real_address + (uintptr_t)unwinder->debug_offsets.code_object.co_code_adaptive;
 
@@ -482,6 +484,12 @@ parse_code_object(RemoteUnwinderObject *unwinder,
     addrq = (uint16_t *)ip - (uint16_t *)meta->addr_code_adaptive;
 #endif
     ;  // Empty statement to avoid C23 extension warning
+
+    if (!unwinder->opcodes && meta->last_frame_info != NULL && meta->last_addrq == addrq) {
+        *result = Py_NewRef(meta->last_frame_info);
+        return 0;
+    }
+
     LocationInfo info = {0};
     bool ok = parse_linetable(addrq, PyBytes_AS_STRING(meta->linetable),
                               PyBytes_GET_SIZE(meta->linetable),
@@ -529,6 +537,11 @@ parse_code_object(RemoteUnwinderObject *unwinder,
         goto error;
     }
 
+    if (!unwinder->opcodes) {
+        Py_XSETREF(meta->last_frame_info, Py_NewRef(tuple));
+        meta->last_addrq = addrq;
+    }
+
     *result = tuple;
     return 0;
 

Modules/_remote_debugging/frame_cache.c

@@ -30,6 +30,7 @@ frame_cache_cleanup(RemoteUnwinderObject *unwinder)
         return;
     }
     for (int i = 0; i < FRAME_CACHE_MAX_THREADS; i++) {
+        Py_CLEAR(unwinder->frame_cache[i].thread_id_obj);
         Py_CLEAR(unwinder->frame_cache[i].frame_list);
     }
     PyMem_Free(unwinder->frame_cache);
@@ -142,6 +143,7 @@ frame_cache_invalidate_stale(RemoteUnwinderObject *unwinder, PyObject *result)
         }
         if (!found) {
             // Clear this entry
+            Py_CLEAR(unwinder->frame_cache[i].thread_id_obj);
             Py_CLEAR(unwinder->frame_cache[i].frame_list);
             unwinder->frame_cache[i].thread_id = 0;
             unwinder->frame_cache[i].thread_state_addr = 0;
@@ -275,6 +277,12 @@ frame_cache_store(
     }
     entry->thread_id = thread_id;
     entry->thread_state_addr = thread_state_addr;
+    if (entry->thread_id_obj == NULL) {
+        entry->thread_id_obj = PyLong_FromUnsignedLongLong(thread_id);
+        if (entry->thread_id_obj == NULL) {
+            return -1;
+        }
+    }
     memcpy(entry->addrs, addrs, num_addrs * sizeof(uintptr_t));
     entry->num_addrs = num_addrs;
     assert(entry->num_addrs == num_addrs);

Modules/_remote_debugging/frames.c

@@ -543,35 +543,23 @@ try_full_cache_hit(
         return -1;
     }
 
-    Py_ssize_t cached_size = PyList_GET_SIZE(entry->frame_list);
-    PyObject *parent_slice = NULL;
-    if (cached_size > 1) {
-        parent_slice = PyList_GetSlice(entry->frame_list, 1, cached_size);
-        if (!parent_slice) {
-            Py_XDECREF(current_frame);
-            return -1;
-        }
-    }
-
     if (current_frame != NULL) {
         if (PyList_Append(ctx->frame_info, current_frame) < 0) {
             Py_DECREF(current_frame);
-            Py_XDECREF(parent_slice);
             return -1;
         }
         Py_DECREF(current_frame);
         STATS_ADD(unwinder, frames_read_from_memory, 1);
     }
 
-    if (parent_slice) {
-        Py_ssize_t cur_size = PyList_GET_SIZE(ctx->frame_info);
-        int result = PyList_SetSlice(ctx->frame_info, cur_size, cur_size, parent_slice);
-        Py_DECREF(parent_slice);
-        if (result < 0) {
+    Py_ssize_t cached_size = PyList_GET_SIZE(entry->frame_list);
+    for (Py_ssize_t i = 1; i < cached_size; i++) {
+        PyObject *cached_frame = PyList_GET_ITEM(entry->frame_list, i);
+        if (PyList_Append(ctx->frame_info, cached_frame) < 0) {
             return -1;
         }
-        STATS_ADD(unwinder, frames_read_from_cache, cached_size - 1);
     }
+    STATS_ADD(unwinder, frames_read_from_cache, cached_size > 1 ? cached_size - 1 : 0);
 
     STATS_INC(unwinder, frame_cache_hits);
     return 1;

Modules/_remote_debugging/module.c

@@ -166,6 +166,7 @@ cached_code_metadata_destroy(void *ptr)
     Py_DECREF(meta->func_name);
     Py_DECREF(meta->file_name);
     Py_DECREF(meta->linetable);
+    Py_XDECREF(meta->last_frame_info);
     PyMem_RawFree(meta);
 }
 

Modules/_remote_debugging/threads.c

@@ -554,10 +554,18 @@ unwind_stack_for_thread(
 
     *current_tstate = GET_MEMBER(uintptr_t, ts, unwinder->debug_offsets.thread_state.next);
 
-    thread_id = PyLong_FromLongLong(tid);
+    if (unwinder->cache_frames) {
+        FrameCacheEntry *entry = frame_cache_find(unwinder, (uint64_t)tid);
+        if (entry && entry->thread_id_obj) {
+            thread_id = Py_NewRef(entry->thread_id_obj);
+        }
+    }
     if (thread_id == NULL) {
-        set_exception_cause(unwinder, PyExc_RuntimeError, "Failed to create thread ID");
-        goto error;
+        thread_id = PyLong_FromLongLong(tid);
+        if (thread_id == NULL) {
+            set_exception_cause(unwinder, PyExc_RuntimeError, "Failed to create thread ID");
+            goto error;
+        }
     }
 
     RemoteDebuggingState *state = RemoteDebugging_GetStateFromObject((PyObject*)unwinder);