gh-116381: Remove bad specializations, add fail stats (GH-116464)

* Remove bad specializations, add fail stats

6 files changed, 41 insertions, 194 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 3276a4a..0397d96 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -2238,11 +2238,8 @@ dummy_func(
         }
 
         family(CONTAINS_OP, INLINE_CACHE_ENTRIES_CONTAINS_OP) = {
-            CONTAINS_OP_LIST,
             CONTAINS_OP_SET,
-            CONTAINS_OP_TUPLE,
             CONTAINS_OP_DICT,
-            CONTAINS_OP_STR,
         };
 
         op(_CONTAINS_OP, (left, right -- b)) {
@@ -2266,46 +2263,25 @@ dummy_func(
 
         macro(CONTAINS_OP) = _SPECIALIZE_CONTAINS_OP + _CONTAINS_OP;
 
-        inst(CONTAINS_OP_LIST, (unused/1, left, right -- b)) {
-            DEOPT_IF(!PyList_CheckExact(right));
-            int res = _PyList_Contains(right, left);
-            DECREF_INPUTS();
-            ERROR_IF(res < 0, error);
-            b = (res ^ oparg) ? Py_True : Py_False;
-        }
-
         inst(CONTAINS_OP_SET, (unused/1, left, right -- b)) {
-            DEOPT_IF(!PySet_CheckExact(right));
+            DEOPT_IF(!(PySet_CheckExact(right) || PyFrozenSet_CheckExact(right)));
+            STAT_INC(CONTAINS_OP, hit);
+            // Note: both set and frozenset use the same seq_contains method!
             int res = _PySet_Contains((PySetObject *)right, left);
             DECREF_INPUTS();
             ERROR_IF(res < 0, error);
             b = (res ^ oparg) ? Py_True : Py_False;
         }
 
-        inst(CONTAINS_OP_TUPLE, (unused/1, left, right -- b)) {
-            DEOPT_IF(!PyTuple_CheckExact(right));
-            int res = _PyTuple_Contains((PyTupleObject *)right, left);
-            DECREF_INPUTS();
-            ERROR_IF(res < 0, error);
-            b = (res ^ oparg) ? Py_True : Py_False;
-        }
-
         inst(CONTAINS_OP_DICT, (unused/1, left, right -- b)) {
             DEOPT_IF(!PyDict_CheckExact(right));
+            STAT_INC(CONTAINS_OP, hit);
             int res = PyDict_Contains(right, left);
             DECREF_INPUTS();
             ERROR_IF(res < 0, error);
             b = (res ^ oparg) ? Py_True : Py_False;
         }
 
-        inst(CONTAINS_OP_STR, (unused/1, left, right -- b)) {
-            DEOPT_IF(!PyUnicode_CheckExact(right));
-            int res = PyUnicode_Contains(right, left);
-            DECREF_INPUTS();
-            ERROR_IF(res < 0, error);
-            b = (res ^ oparg) ? Py_True : Py_False;
-        }
-
         inst(CHECK_EG_MATCH, (exc_value, match_type -- rest, match)) {
             if (_PyEval_CheckExceptStarTypeValid(tstate, match_type) < 0) {
                 DECREF_INPUTS();
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 2e7b970..26ac159 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -2189,24 +2189,6 @@
             break;
         }
 
-        case _CONTAINS_OP_LIST: {
-            PyObject *right;
-            PyObject *left;
-            PyObject *b;
-            oparg = CURRENT_OPARG();
-            right = stack_pointer[-1];
-            left = stack_pointer[-2];
-            if (!PyList_CheckExact(right)) goto deoptimize;
-            int res = _PyList_Contains(right, left);
-            Py_DECREF(left);
-            Py_DECREF(right);
-            if (res < 0) goto pop_2_error_tier_two;
-            b = (res ^ oparg) ? Py_True : Py_False;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            break;
-        }
-
         case _CONTAINS_OP_SET: {
             PyObject *right;
             PyObject *left;
@@ -2214,7 +2196,9 @@
             oparg = CURRENT_OPARG();
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            if (!PySet_CheckExact(right)) goto deoptimize;
+            if (!(PySet_CheckExact(right) || PyFrozenSet_CheckExact(right))) goto deoptimize;
+            STAT_INC(CONTAINS_OP, hit);
+            // Note: both set and frozenset use the same seq_contains method!
             int res = _PySet_Contains((PySetObject *)right, left);
             Py_DECREF(left);
             Py_DECREF(right);
@@ -2225,24 +2209,6 @@
             break;
         }
 
-        case _CONTAINS_OP_TUPLE: {
-            PyObject *right;
-            PyObject *left;
-            PyObject *b;
-            oparg = CURRENT_OPARG();
-            right = stack_pointer[-1];
-            left = stack_pointer[-2];
-            if (!PyTuple_CheckExact(right)) goto deoptimize;
-            int res = _PyTuple_Contains((PyTupleObject *)right, left);
-            Py_DECREF(left);
-            Py_DECREF(right);
-            if (res < 0) goto pop_2_error_tier_two;
-            b = (res ^ oparg) ? Py_True : Py_False;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            break;
-        }
-
         case _CONTAINS_OP_DICT: {
             PyObject *right;
             PyObject *left;
@@ -2251,6 +2217,7 @@
             right = stack_pointer[-1];
             left = stack_pointer[-2];
             if (!PyDict_CheckExact(right)) goto deoptimize;
+            STAT_INC(CONTAINS_OP, hit);
             int res = PyDict_Contains(right, left);
             Py_DECREF(left);
             Py_DECREF(right);
@@ -2261,24 +2228,6 @@
             break;
         }
 
-        case _CONTAINS_OP_STR: {
-            PyObject *right;
-            PyObject *left;
-            PyObject *b;
-            oparg = CURRENT_OPARG();
-            right = stack_pointer[-1];
-            left = stack_pointer[-2];
-            if (!PyUnicode_CheckExact(right)) goto deoptimize;
-            int res = PyUnicode_Contains(right, left);
-            Py_DECREF(left);
-            Py_DECREF(right);
-            if (res < 0) goto pop_2_error_tier_two;
-            b = (res ^ oparg) ? Py_True : Py_False;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            break;
-        }
-
         case _CHECK_EG_MATCH: {
             PyObject *match_type;
             PyObject *exc_value;
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 54c4861..f5d125c 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -2176,6 +2176,7 @@
             right = stack_pointer[-1];
             left = stack_pointer[-2];
             DEOPT_IF(!PyDict_CheckExact(right), CONTAINS_OP);
+            STAT_INC(CONTAINS_OP, hit);
             int res = PyDict_Contains(right, left);
             Py_DECREF(left);
             Py_DECREF(right);
@@ -2186,28 +2187,6 @@
             DISPATCH();
         }
 
-        TARGET(CONTAINS_OP_LIST) {
-            frame->instr_ptr = next_instr;
-            next_instr += 2;
-            INSTRUCTION_STATS(CONTAINS_OP_LIST);
-            static_assert(INLINE_CACHE_ENTRIES_CONTAINS_OP == 1, "incorrect cache size");
-            PyObject *right;
-            PyObject *left;
-            PyObject *b;
-            /* Skip 1 cache entry */
-            right = stack_pointer[-1];
-            left = stack_pointer[-2];
-            DEOPT_IF(!PyList_CheckExact(right), CONTAINS_OP);
-            int res = _PyList_Contains(right, left);
-            Py_DECREF(left);
-            Py_DECREF(right);
-            if (res < 0) goto pop_2_error;
-            b = (res ^ oparg) ? Py_True : Py_False;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            DISPATCH();
-        }
-
         TARGET(CONTAINS_OP_SET) {
             frame->instr_ptr = next_instr;
             next_instr += 2;
@@ -2219,7 +2198,9 @@
             /* Skip 1 cache entry */
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            DEOPT_IF(!PySet_CheckExact(right), CONTAINS_OP);
+            DEOPT_IF(!(PySet_CheckExact(right) || PyFrozenSet_CheckExact(right)), CONTAINS_OP);
+            STAT_INC(CONTAINS_OP, hit);
+            // Note: both set and frozenset use the same seq_contains method!
             int res = _PySet_Contains((PySetObject *)right, left);
             Py_DECREF(left);
             Py_DECREF(right);
@@ -2230,50 +2211,6 @@
             DISPATCH();
         }
 
-        TARGET(CONTAINS_OP_STR) {
-            frame->instr_ptr = next_instr;
-            next_instr += 2;
-            INSTRUCTION_STATS(CONTAINS_OP_STR);
-            static_assert(INLINE_CACHE_ENTRIES_CONTAINS_OP == 1, "incorrect cache size");
-            PyObject *right;
-            PyObject *left;
-            PyObject *b;
-            /* Skip 1 cache entry */
-            right = stack_pointer[-1];
-            left = stack_pointer[-2];
-            DEOPT_IF(!PyUnicode_CheckExact(right), CONTAINS_OP);
-            int res = PyUnicode_Contains(right, left);
-            Py_DECREF(left);
-            Py_DECREF(right);
-            if (res < 0) goto pop_2_error;
-            b = (res ^ oparg) ? Py_True : Py_False;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            DISPATCH();
-        }
-
-        TARGET(CONTAINS_OP_TUPLE) {
-            frame->instr_ptr = next_instr;
-            next_instr += 2;
-            INSTRUCTION_STATS(CONTAINS_OP_TUPLE);
-            static_assert(INLINE_CACHE_ENTRIES_CONTAINS_OP == 1, "incorrect cache size");
-            PyObject *right;
-            PyObject *left;
-            PyObject *b;
-            /* Skip 1 cache entry */
-            right = stack_pointer[-1];
-            left = stack_pointer[-2];
-            DEOPT_IF(!PyTuple_CheckExact(right), CONTAINS_OP);
-            int res = _PyTuple_Contains((PyTupleObject *)right, left);
-            Py_DECREF(left);
-            Py_DECREF(right);
-            if (res < 0) goto pop_2_error;
-            b = (res ^ oparg) ? Py_True : Py_False;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            DISPATCH();
-        }
-
         TARGET(CONVERT_VALUE) {
             frame->instr_ptr = next_instr;
             next_instr += 1;
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index 6b3846d..4061ba3 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -183,10 +183,7 @@ static void *opcode_targets[256] = {
     &&TARGET_COMPARE_OP_INT,
     &&TARGET_COMPARE_OP_STR,
     &&TARGET_CONTAINS_OP_DICT,
-    &&TARGET_CONTAINS_OP_LIST,
     &&TARGET_CONTAINS_OP_SET,
-    &&TARGET_CONTAINS_OP_STR,
-    &&TARGET_CONTAINS_OP_TUPLE,
     &&TARGET_FOR_ITER_GEN,
     &&TARGET_FOR_ITER_LIST,
     &&TARGET_FOR_ITER_RANGE,
@@ -235,6 +232,9 @@ static void *opcode_targets[256] = {
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
+    &&_unknown_opcode,
+    &&_unknown_opcode,
+    &&_unknown_opcode,
     &&TARGET_INSTRUMENTED_RESUME,
     &&TARGET_INSTRUMENTED_END_FOR,
     &&TARGET_INSTRUMENTED_END_SEND,
diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index e438497..c88b251 100644
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -1218,15 +1218,6 @@
             break;
         }
 
-        case _CONTAINS_OP_LIST: {
-            _Py_UopsSymbol *b;
-            b = sym_new_unknown(ctx);
-            if (b == NULL) goto out_of_space;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            break;
-        }
-
         case _CONTAINS_OP_SET: {
             _Py_UopsSymbol *b;
             b = sym_new_unknown(ctx);
@@ -1236,15 +1227,6 @@
             break;
         }
 
-        case _CONTAINS_OP_TUPLE: {
-            _Py_UopsSymbol *b;
-            b = sym_new_unknown(ctx);
-            if (b == NULL) goto out_of_space;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            break;
-        }
-
         case _CONTAINS_OP_DICT: {
             _Py_UopsSymbol *b;
             b = sym_new_unknown(ctx);
@@ -1254,15 +1236,6 @@
             break;
         }
 
-        case _CONTAINS_OP_STR: {
-            _Py_UopsSymbol *b;
-            b = sym_new_unknown(ctx);
-            if (b == NULL) goto out_of_space;
-            stack_pointer[-2] = b;
-            stack_pointer += -1;
-            break;
-        }
-
         case _CHECK_EG_MATCH: {
             _Py_UopsSymbol *rest;
             _Py_UopsSymbol *match;
diff --git a/Python/specialize.c b/Python/specialize.c
index 5d339b8..b1f9eb7 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -600,6 +600,12 @@ _PyCode_Quicken(PyCodeObject *code)
 #define SPEC_FAIL_TO_BOOL_SET         17
 #define SPEC_FAIL_TO_BOOL_TUPLE       18
 
+// CONTAINS_OP
+#define SPEC_FAIL_CONTAINS_OP_STR        9
+#define SPEC_FAIL_CONTAINS_OP_TUPLE      10
+#define SPEC_FAIL_CONTAINS_OP_LIST       11
+#define SPEC_FAIL_CONTAINS_OP_USER_CLASS 12
+
 static int function_kind(PyCodeObject *code);
 static bool function_check_args(PyObject *o, int expected_argcount, int opcode);
 static uint32_t function_get_version(PyObject *o, int opcode);
@@ -2562,34 +2568,40 @@ success:
     cache->counter = adaptive_counter_cooldown();
 }
 
-void
-_Py_Specialize_ContainsOp(PyObject *value, _Py_CODEUNIT *instr)
-{
-    assert(ENABLE_SPECIALIZATION);
-    assert(_PyOpcode_Caches[CONTAINS_OP] == INLINE_CACHE_ENTRIES_COMPARE_OP);
-    _PyContainsOpCache *cache = (_PyContainsOpCache  *)(instr + 1);
+#ifdef Py_STATS
+static int containsop_fail_kind(PyObject *value) {
     if (PyUnicode_CheckExact(value)) {
-        instr->op.code = CONTAINS_OP_STR;
-        goto success;
+        return SPEC_FAIL_CONTAINS_OP_STR;
     }
     if (PyList_CheckExact(value)) {
-        instr->op.code = CONTAINS_OP_LIST;
-        goto success;
+        return SPEC_FAIL_CONTAINS_OP_LIST;
     }
     if (PyTuple_CheckExact(value)) {
-        instr->op.code = CONTAINS_OP_TUPLE;
-        goto success;
+        return SPEC_FAIL_CONTAINS_OP_TUPLE;
+    }
+    if (PyType_Check(value)) {
+        return SPEC_FAIL_CONTAINS_OP_USER_CLASS;
     }
+    return SPEC_FAIL_OTHER;
+}
+#endif   // Py_STATS
+
+void
+_Py_Specialize_ContainsOp(PyObject *value, _Py_CODEUNIT *instr)
+{
+    assert(ENABLE_SPECIALIZATION);
+    assert(_PyOpcode_Caches[CONTAINS_OP] == INLINE_CACHE_ENTRIES_COMPARE_OP);
+    _PyContainsOpCache *cache = (_PyContainsOpCache  *)(instr + 1);
     if (PyDict_CheckExact(value)) {
         instr->op.code = CONTAINS_OP_DICT;
         goto success;
     }
-    if (PySet_CheckExact(value)) {
+    if (PySet_CheckExact(value) || PyFrozenSet_CheckExact(value)) {
         instr->op.code = CONTAINS_OP_SET;
         goto success;
     }
 
-
+    SPECIALIZATION_FAIL(CONTAINS_OP, containsop_fail_kind(value));
     STAT_INC(CONTAINS_OP, failure);
     instr->op.code = CONTAINS_OP;
     cache->counter = adaptive_counter_backoff(cache->counter);