ceval.c

/* Execute compiled code */

#define _PY_INTERPRETER

#include "Python.h"
#include "pycore_abstract.h"      // _PyIndex_Check()
#include "pycore_audit.h"         // _PySys_Audit()
#include "pycore_backoff.h"
#include "pycore_call.h"          // _PyObject_CallNoArgs()
#include "pycore_cell.h"          // PyCell_GetRef()
#include "pycore_ceval.h"         // SPECIAL___ENTER__
#include "pycore_code.h"
#include "pycore_dict.h"
#include "pycore_emscripten_signal.h"  // _Py_CHECK_EMSCRIPTEN_SIGNALS
#include "pycore_floatobject.h"   // _PyFloat_ExactDealloc()
#include "pycore_frame.h"
#include "pycore_function.h"
#include "pycore_genobject.h"     // _PyCoro_GetAwaitableIter()
#include "pycore_import.h"        // _PyImport_IsDefaultImportFunc()
#include "pycore_instruments.h"
#include "pycore_interpframe.h"   // _PyFrame_SetStackPointer()
#include "pycore_interpolation.h" // _PyInterpolation_Build()
#include "pycore_intrinsics.h"
#include "pycore_jit.h"
#include "pycore_list.h"          // _PyList_GetItemRef()
#include "pycore_long.h"          // _PyLong_GetZero()
#include "pycore_moduleobject.h"  // PyModuleObject
#include "pycore_object.h"        // _PyObject_GC_TRACK()
#include "pycore_opcode_metadata.h" // EXTRA_CASES
#include "pycore_opcode_utils.h"  // MAKE_FUNCTION_*
#include "pycore_optimizer.h"     // _PyUOpExecutor_Type
#include "pycore_pyatomic_ft_wrappers.h" // FT_ATOMIC_*
#include "pycore_pyerrors.h"      // _PyErr_GetRaisedException()
#include "pycore_pystate.h"       // _PyInterpreterState_GET()
#include "pycore_range.h"         // _PyRangeIterObject
#include "pycore_setobject.h"     // _PySet_Update()
#include "pycore_sliceobject.h"   // _PyBuildSlice_ConsumeRefs
#include "pycore_sysmodule.h"     // _PySys_GetOptionalAttrString()
#include "pycore_template.h"      // _PyTemplate_Build()
#include "pycore_traceback.h"     // _PyTraceBack_FromFrame
#include "pycore_tuple.h"         // _PyTuple_ITEMS()
#include "pycore_uop_ids.h"       // Uops

#include "dictobject.h"
#include "frameobject.h"          // _PyInterpreterFrame_GetLine
#include "opcode.h"
#include "pydtrace.h"
#include "setobject.h"
#include "pycore_stackref.h"

#include <stdbool.h>              // bool

#if !defined(Py_BUILD_CORE)
#  error "ceval.c must be build with Py_BUILD_CORE define for best performance"
#endif

#if !defined(Py_DEBUG) && !defined(Py_TRACE_REFS)
// GH-89279: The MSVC compiler does not inline these static inline functions
// in PGO build in _PyEval_EvalFrameDefault(), because this function is over
// the limit of PGO, and that limit cannot be configured.
// Define them as macros to make sure that they are always inlined by the
// preprocessor.

#undef Py_IS_TYPE
#define Py_IS_TYPE(ob, type) \
    (_PyObject_CAST(ob)->ob_type == (type))

#undef Py_XDECREF
#define Py_XDECREF(arg) \
    do { \
        PyObject *xop = _PyObject_CAST(arg); \
        if (xop != NULL) { \
            Py_DECREF(xop); \
        } \
    } while (0)

#ifndef Py_GIL_DISABLED

#undef Py_DECREF
#define Py_DECREF(arg) \
    do { \
        PyObject *op = _PyObject_CAST(arg); \
        if (_Py_IsImmortal(op)) { \
            _Py_DECREF_IMMORTAL_STAT_INC(); \
            break; \
        } \
        _Py_DECREF_STAT_INC(); \
        if (--op->ob_refcnt == 0) { \
            _PyReftracerTrack(op, PyRefTracer_DESTROY); \
            destructor dealloc = Py_TYPE(op)->tp_dealloc; \
            (*dealloc)(op); \
        } \
    } while (0)

#undef _Py_DECREF_SPECIALIZED
#define _Py_DECREF_SPECIALIZED(arg, dealloc) \
    do { \
        PyObject *op = _PyObject_CAST(arg); \
        if (_Py_IsImmortal(op)) { \
            _Py_DECREF_IMMORTAL_STAT_INC(); \
            break; \
        } \
        _Py_DECREF_STAT_INC(); \
        if (--op->ob_refcnt == 0) { \
            _PyReftracerTrack(op, PyRefTracer_DESTROY); \
            destructor d = (destructor)(dealloc); \
            d(op); \
        } \
    } while (0)

#else // Py_GIL_DISABLED

#undef Py_DECREF
#define Py_DECREF(arg) \
    do { \
        PyObject *op = _PyObject_CAST(arg); \
        uint32_t local = _Py_atomic_load_uint32_relaxed(&op->ob_ref_local); \
        if (local == _Py_IMMORTAL_REFCNT_LOCAL) { \
            _Py_DECREF_IMMORTAL_STAT_INC(); \
            break; \
        } \
        _Py_DECREF_STAT_INC(); \
        if (_Py_IsOwnedByCurrentThread(op)) { \
            local--; \
            _Py_atomic_store_uint32_relaxed(&op->ob_ref_local, local); \
            if (local == 0) { \
                _Py_MergeZeroLocalRefcount(op); \
            } \
        } \
        else { \
            _Py_DecRefShared(op); \
        } \
    } while (0)

#undef _Py_DECREF_SPECIALIZED
#define _Py_DECREF_SPECIALIZED(arg, dealloc) Py_DECREF(arg)

#endif
#endif


static void
check_invalid_reentrancy(void)
{
#if defined(Py_DEBUG) && defined(Py_GIL_DISABLED)
    // In the free-threaded build, the interpreter must not be re-entered if
    // the world-is-stopped.  If so, that's a bug somewhere (quite likely in
    // the painfully complex typeobject code).
    PyInterpreterState *interp = _PyInterpreterState_GET();
    assert(!interp->stoptheworld.world_stopped);
#endif
}


#ifdef Py_DEBUG
static void
dump_item(_PyStackRef item)
{
    if (PyStackRef_IsNull(item)) {
        printf("<NULL>");
        return;
    }
    if (PyStackRef_IsTaggedInt(item)) {
        printf("%" PRId64, (int64_t)PyStackRef_UntagInt(item));
        return;
    }
    PyObject *obj = PyStackRef_AsPyObjectBorrow(item);
    if (obj == NULL) {
        printf("<nil>");
        return;
    }
    // Don't call __repr__(), it might recurse into the interpreter.
    printf("<%s at %p>", Py_TYPE(obj)->tp_name, (void *)obj);
}

static void
dump_stack(_PyInterpreterFrame *frame, _PyStackRef *stack_pointer)
{
    _PyFrame_SetStackPointer(frame, stack_pointer);
    _PyStackRef *locals_base = _PyFrame_GetLocalsArray(frame);
    _PyStackRef *stack_base = _PyFrame_Stackbase(frame);
    PyObject *exc = PyErr_GetRaisedException();
    printf("    locals=[");
    for (_PyStackRef *ptr = locals_base; ptr < stack_base; ptr++) {
        if (ptr != locals_base) {
            printf(", ");
        }
        dump_item(*ptr);
    }
    printf("]\n");
    if (stack_pointer < stack_base) {
        printf("    stack=%d\n", (int)(stack_pointer-stack_base));
    }
    else {
        printf("    stack=[");
        for (_PyStackRef *ptr = stack_base; ptr < stack_pointer; ptr++) {
            if (ptr != stack_base) {
                printf(", ");
            }
            dump_item(*ptr);
        }
        printf("]\n");
    }
    fflush(stdout);
    PyErr_SetRaisedException(exc);
    _PyFrame_GetStackPointer(frame);
}

static void
lltrace_instruction(_PyInterpreterFrame *frame,
                    _PyStackRef *stack_pointer,
                    _Py_CODEUNIT *next_instr,
                    int opcode,
                    int oparg)
{
    int offset = 0;
    if (frame->owner < FRAME_OWNED_BY_INTERPRETER) {
        dump_stack(frame, stack_pointer);
        offset = (int)(next_instr - _PyFrame_GetBytecode(frame));
    }
    const char *opname = _PyOpcode_OpName[opcode];
    assert(opname != NULL);
    if (OPCODE_HAS_ARG((int)_PyOpcode_Deopt[opcode])) {
        printf("%d: %s %d\n", offset * 2, opname, oparg);
    }
    else {
        printf("%d: %s\n", offset * 2, opname);
    }
    fflush(stdout);
}

static void
lltrace_resume_frame(_PyInterpreterFrame *frame)
{
    PyObject *fobj = PyStackRef_AsPyObjectBorrow(frame->f_funcobj);
    if (!PyStackRef_CodeCheck(frame->f_executable) ||
        fobj == NULL ||
        !PyFunction_Check(fobj)
    ) {
        printf("\nResuming frame.\n");
        return;
    }
    PyFunctionObject *f = (PyFunctionObject *)fobj;
    PyObject *exc = PyErr_GetRaisedException();
    PyObject *name = f->func_qualname;
    if (name == NULL) {
        name = f->func_name;
    }
    printf("\nResuming frame");
    if (name) {
        printf(" for ");
        if (PyObject_Print(name, stdout, 0) < 0) {
            PyErr_Clear();
        }
    }
    if (f->func_module) {
        printf(" in module ");
        if (PyObject_Print(f->func_module, stdout, 0) < 0) {
            PyErr_Clear();
        }
    }
    printf("\n");
    fflush(stdout);
    PyErr_SetRaisedException(exc);
}

static int
maybe_lltrace_resume_frame(_PyInterpreterFrame *frame, PyObject *globals)
{
    if (globals == NULL) {
        return 0;
    }
    if (frame->owner >= FRAME_OWNED_BY_INTERPRETER) {
        return 0;
    }
    int r = PyDict_Contains(globals, &_Py_ID(__lltrace__));
    if (r < 0) {
        return -1;
    }
    int lltrace = r * 5;  // Levels 1-4 only trace uops
    if (!lltrace) {
        // Can also be controlled by environment variable
        char *python_lltrace = Py_GETENV("PYTHON_LLTRACE");
        if (python_lltrace != NULL && *python_lltrace >= '0') {
            lltrace = *python_lltrace - '0';  // TODO: Parse an int and all that
        }
    }
    if (lltrace >= 5) {
        lltrace_resume_frame(frame);
    }
    return lltrace;
}

#endif

static void monitor_reraise(PyThreadState *tstate,
                 _PyInterpreterFrame *frame,
                 _Py_CODEUNIT *instr);
static int monitor_stop_iteration(PyThreadState *tstate,
                 _PyInterpreterFrame *frame,
                 _Py_CODEUNIT *instr,
                 PyObject *value);
static void monitor_unwind(PyThreadState *tstate,
                 _PyInterpreterFrame *frame,
                 _Py_CODEUNIT *instr);
static int monitor_handled(PyThreadState *tstate,
                 _PyInterpreterFrame *frame,
                 _Py_CODEUNIT *instr, PyObject *exc);
static void monitor_throw(PyThreadState *tstate,
                 _PyInterpreterFrame *frame,
                 _Py_CODEUNIT *instr);

static int get_exception_handler(PyCodeObject *, int, int*, int*, int*);
static  _PyInterpreterFrame *
_PyEvalFramePushAndInit_Ex(PyThreadState *tstate, _PyStackRef func,
    PyObject *locals, Py_ssize_t nargs, PyObject *callargs, PyObject *kwargs, _PyInterpreterFrame *previous);

#ifdef HAVE_ERRNO_H
#include <errno.h>
#endif

int
Py_GetRecursionLimit(void)
{
    PyInterpreterState *interp = _PyInterpreterState_GET();
    return interp->ceval.recursion_limit;
}

void
Py_SetRecursionLimit(int new_limit)
{
    PyInterpreterState *interp = _PyInterpreterState_GET();
    _PyEval_StopTheWorld(interp);
    interp->ceval.recursion_limit = new_limit;
    _Py_FOR_EACH_TSTATE_BEGIN(interp, p) {
        int depth = p->py_recursion_limit - p->py_recursion_remaining;
        p->py_recursion_limit = new_limit;
        p->py_recursion_remaining = new_limit - depth;
    }
    _Py_FOR_EACH_TSTATE_END(interp);
    _PyEval_StartTheWorld(interp);
}

int
_Py_ReachedRecursionLimitWithMargin(PyThreadState *tstate, int margin_count)
{
    uintptr_t here_addr = _Py_get_machine_stack_pointer();
    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
    if (here_addr > _tstate->c_stack_soft_limit + margin_count * _PyOS_STACK_MARGIN_BYTES) {
        return 0;
    }
    if (_tstate->c_stack_hard_limit == 0) {
        _Py_InitializeRecursionLimits(tstate);
    }
    return here_addr <= _tstate->c_stack_soft_limit + margin_count * _PyOS_STACK_MARGIN_BYTES;
}

void
_Py_EnterRecursiveCallUnchecked(PyThreadState *tstate)
{
    uintptr_t here_addr = _Py_get_machine_stack_pointer();
    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
    if (here_addr < _tstate->c_stack_hard_limit) {
        Py_FatalError("Unchecked stack overflow.");
    }
}

#if defined(__s390x__)
#  define Py_C_STACK_SIZE 320000
#elif defined(_WIN32)
   // Don't define Py_C_STACK_SIZE, ask the O/S
#elif defined(__ANDROID__)
#  define Py_C_STACK_SIZE 1200000
#elif defined(__sparc__)
#  define Py_C_STACK_SIZE 1600000
#elif defined(__hppa__) || defined(__powerpc64__)
#  define Py_C_STACK_SIZE 2000000
#else
#  define Py_C_STACK_SIZE 4000000
#endif

#if defined(__EMSCRIPTEN__)

// Temporary workaround to make `pthread_getattr_np` work on Emscripten.
// Emscripten 4.0.6 will contain a fix:
// https://github.com/emscripten-core/emscripten/pull/23887

#include "emscripten/stack.h"

#define pthread_attr_t workaround_pthread_attr_t
#define pthread_getattr_np workaround_pthread_getattr_np
#define pthread_attr_getguardsize workaround_pthread_attr_getguardsize
#define pthread_attr_getstack workaround_pthread_attr_getstack
#define pthread_attr_destroy workaround_pthread_attr_destroy

typedef struct {
    void *_a_stackaddr;
    size_t _a_stacksize, _a_guardsize;
} pthread_attr_t;

extern __attribute__((__visibility__("hidden"))) unsigned __default_guardsize;

// Modified version of pthread_getattr_np from the upstream PR.

int pthread_getattr_np(pthread_t thread, pthread_attr_t *attr) {
  attr->_a_stackaddr = (void*)emscripten_stack_get_base();
  attr->_a_stacksize = emscripten_stack_get_base() - emscripten_stack_get_end();
  attr->_a_guardsize = __default_guardsize;
  return 0;
}

// These three functions copied without any changes from Emscripten libc.

int pthread_attr_getguardsize(const pthread_attr_t *restrict a, size_t *restrict size)
{
	*size = a->_a_guardsize;
	return 0;
}

int pthread_attr_getstack(const pthread_attr_t *restrict a, void **restrict addr, size_t *restrict size)
{
/// XXX musl is not standard-conforming? It should not report EINVAL if _a_stackaddr is zero, and it should
///     report EINVAL if a is null: http://pubs.opengroup.org/onlinepubs/009695399/functions/pthread_attr_getstack.html
	if (!a) return EINVAL;
//	if (!a->_a_stackaddr)
//		return EINVAL;

	*size = a->_a_stacksize;
	*addr = (void *)(a->_a_stackaddr - *size);
	return 0;
}

int pthread_attr_destroy(pthread_attr_t *a)
{
	return 0;
}

#endif


void
_Py_InitializeRecursionLimits(PyThreadState *tstate)
{
    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
#ifdef WIN32
    ULONG_PTR low, high;
    GetCurrentThreadStackLimits(&low, &high);
    _tstate->c_stack_top = (uintptr_t)high;
    ULONG guarantee = 0;
    SetThreadStackGuarantee(&guarantee);
    _tstate->c_stack_hard_limit = ((uintptr_t)low) + guarantee + _PyOS_STACK_MARGIN_BYTES;
    _tstate->c_stack_soft_limit = _tstate->c_stack_hard_limit + _PyOS_STACK_MARGIN_BYTES;
#else
    uintptr_t here_addr = _Py_get_machine_stack_pointer();
#  if defined(HAVE_PTHREAD_GETATTR_NP) && !defined(_AIX) && !defined(__NetBSD__)
    size_t stack_size, guard_size;
    void *stack_addr;
    pthread_attr_t attr;
    int err = pthread_getattr_np(pthread_self(), &attr);
    if (err == 0) {
        err = pthread_attr_getguardsize(&attr, &guard_size);
        err |= pthread_attr_getstack(&attr, &stack_addr, &stack_size);
        err |= pthread_attr_destroy(&attr);
    }
    if (err == 0) {
        uintptr_t base = ((uintptr_t)stack_addr) + guard_size;
        _tstate->c_stack_top = base + stack_size;
#ifdef _Py_THREAD_SANITIZER
        // Thread sanitizer crashes if we use a bit more than half the stack.
        _tstate->c_stack_soft_limit = base + (stack_size / 2);
#else
        _tstate->c_stack_soft_limit = base + _PyOS_STACK_MARGIN_BYTES * 2;
#endif
        _tstate->c_stack_hard_limit = base + _PyOS_STACK_MARGIN_BYTES;
        assert(_tstate->c_stack_soft_limit < here_addr);
        assert(here_addr < _tstate->c_stack_top);
        return;
    }
#  endif
    _tstate->c_stack_top = _Py_SIZE_ROUND_UP(here_addr, 4096);
    _tstate->c_stack_soft_limit = _tstate->c_stack_top - Py_C_STACK_SIZE;
    _tstate->c_stack_hard_limit = _tstate->c_stack_top - (Py_C_STACK_SIZE + _PyOS_STACK_MARGIN_BYTES);
#endif
}

/* The function _Py_EnterRecursiveCallTstate() only calls _Py_CheckRecursiveCall()
   if the recursion_depth reaches recursion_limit. */
int
_Py_CheckRecursiveCall(PyThreadState *tstate, const char *where)
{
    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
    uintptr_t here_addr = _Py_get_machine_stack_pointer();
    assert(_tstate->c_stack_soft_limit != 0);
    assert(_tstate->c_stack_hard_limit != 0);
    if (here_addr < _tstate->c_stack_hard_limit) {
        /* Overflowing while handling an overflow. Give up. */
        int kbytes_used = (int)(_tstate->c_stack_top - here_addr)/1024;
        char buffer[80];
        snprintf(buffer, 80, "Unrecoverable stack overflow (used %d kB)%s", kbytes_used, where);
        Py_FatalError(buffer);
    }
    if (tstate->recursion_headroom) {
        return 0;
    }
    else {
        int kbytes_used = (int)(_tstate->c_stack_top - here_addr)/1024;
        tstate->recursion_headroom++;
        _PyErr_Format(tstate, PyExc_RecursionError,
                    "Stack overflow (used %d kB)%s",
                    kbytes_used,
                    where);
        tstate->recursion_headroom--;
        return -1;
    }
}


const binaryfunc _PyEval_BinaryOps[] = {
    [NB_ADD] = PyNumber_Add,
    [NB_AND] = PyNumber_And,
    [NB_FLOOR_DIVIDE] = PyNumber_FloorDivide,
    [NB_LSHIFT] = PyNumber_Lshift,
    [NB_MATRIX_MULTIPLY] = PyNumber_MatrixMultiply,
    [NB_MULTIPLY] = PyNumber_Multiply,
    [NB_REMAINDER] = PyNumber_Remainder,
    [NB_OR] = PyNumber_Or,
    [NB_POWER] = _PyNumber_PowerNoMod,
    [NB_RSHIFT] = PyNumber_Rshift,
    [NB_SUBTRACT] = PyNumber_Subtract,
    [NB_TRUE_DIVIDE] = PyNumber_TrueDivide,
    [NB_XOR] = PyNumber_Xor,
    [NB_INPLACE_ADD] = PyNumber_InPlaceAdd,
    [NB_INPLACE_AND] = PyNumber_InPlaceAnd,
    [NB_INPLACE_FLOOR_DIVIDE] = PyNumber_InPlaceFloorDivide,
    [NB_INPLACE_LSHIFT] = PyNumber_InPlaceLshift,
    [NB_INPLACE_MATRIX_MULTIPLY] = PyNumber_InPlaceMatrixMultiply,
    [NB_INPLACE_MULTIPLY] = PyNumber_InPlaceMultiply,
    [NB_INPLACE_REMAINDER] = PyNumber_InPlaceRemainder,
    [NB_INPLACE_OR] = PyNumber_InPlaceOr,
    [NB_INPLACE_POWER] = _PyNumber_InPlacePowerNoMod,
    [NB_INPLACE_RSHIFT] = PyNumber_InPlaceRshift,
    [NB_INPLACE_SUBTRACT] = PyNumber_InPlaceSubtract,
    [NB_INPLACE_TRUE_DIVIDE] = PyNumber_InPlaceTrueDivide,
    [NB_INPLACE_XOR] = PyNumber_InPlaceXor,
    [NB_SUBSCR] = PyObject_GetItem,
};

const conversion_func _PyEval_ConversionFuncs[4] = {
    [FVC_STR] = PyObject_Str,
    [FVC_REPR] = PyObject_Repr,
    [FVC_ASCII] = PyObject_ASCII
};

const _Py_SpecialMethod _Py_SpecialMethods[] = {
    [SPECIAL___ENTER__] = {
        .name = &_Py_ID(__enter__),
        .error = (
            "'%T' object does not support the context manager protocol "
            "(missed __enter__ method)"
        ),
        .error_suggestion = (
            "'%T' object does not support the context manager protocol "
            "(missed __enter__ method) but it supports the asynchronous "
            "context manager protocol. Did you mean to use 'async with'?"
        )
    },
    [SPECIAL___EXIT__] = {
        .name = &_Py_ID(__exit__),
        .error = (
            "'%T' object does not support the context manager protocol "
            "(missed __exit__ method)"
        ),
        .error_suggestion = (
            "'%T' object does not support the context manager protocol "
            "(missed __exit__ method) but it supports the asynchronous "
            "context manager protocol. Did you mean to use 'async with'?"
        )
    },
    [SPECIAL___AENTER__] = {
        .name = &_Py_ID(__aenter__),
        .error = (
            "'%T' object does not support the asynchronous "
            "context manager protocol (missed __aenter__ method)"
        ),
        .error_suggestion = (
            "'%T' object does not support the asynchronous context manager "
            "protocol (missed __aenter__ method) but it supports the context "
            "manager protocol. Did you mean to use 'with'?"
        )
    },
    [SPECIAL___AEXIT__] = {
        .name = &_Py_ID(__aexit__),
        .error = (
            "'%T' object does not support the asynchronous "
            "context manager protocol (missed __aexit__ method)"
        ),
        .error_suggestion = (
            "'%T' object does not support the asynchronous context manager "
            "protocol (missed __aexit__ method) but it supports the context "
            "manager protocol. Did you mean to use 'with'?"
        )
    }
};

const size_t _Py_FunctionAttributeOffsets[] = {
    [MAKE_FUNCTION_CLOSURE] = offsetof(PyFunctionObject, func_closure),
    [MAKE_FUNCTION_ANNOTATIONS] = offsetof(PyFunctionObject, func_annotations),
    [MAKE_FUNCTION_KWDEFAULTS] = offsetof(PyFunctionObject, func_kwdefaults),
    [MAKE_FUNCTION_DEFAULTS] = offsetof(PyFunctionObject, func_defaults),
    [MAKE_FUNCTION_ANNOTATE] = offsetof(PyFunctionObject, func_annotate),
};

// PEP 634: Structural Pattern Matching


// Return a tuple of values corresponding to keys, with error checks for
// duplicate/missing keys.
PyObject *
_PyEval_MatchKeys(PyThreadState *tstate, PyObject *map, PyObject *keys)
{
    assert(PyTuple_CheckExact(keys));
    Py_ssize_t nkeys = PyTuple_GET_SIZE(keys);
    if (!nkeys) {
        // No keys means no items.
        return PyTuple_New(0);
    }
    PyObject *seen = NULL;
    PyObject *dummy = NULL;
    PyObject *values = NULL;
    // We use the two argument form of map.get(key, default) for two reasons:
    // - Atomically check for a key and get its value without error handling.
    // - Don't cause key creation or resizing in dict subclasses like
    //   collections.defaultdict that define __missing__ (or similar).
    _PyCStackRef cref;
    _PyThreadState_PushCStackRef(tstate, &cref);
    int meth_found = _PyObject_GetMethodStackRef(tstate, map, &_Py_ID(get), &cref.ref);
    PyObject *get = PyStackRef_AsPyObjectBorrow(cref.ref);
    if (get == NULL) {
        goto fail;
    }
    seen = PySet_New(NULL);
    if (seen == NULL) {
        goto fail;
    }
    // dummy = object()
    dummy = _PyObject_CallNoArgs((PyObject *)&PyBaseObject_Type);
    if (dummy == NULL) {
        goto fail;
    }
    values = PyTuple_New(nkeys);
    if (values == NULL) {
        goto fail;
    }
    for (Py_ssize_t i = 0; i < nkeys; i++) {
        PyObject *key = PyTuple_GET_ITEM(keys, i);
        if (PySet_Contains(seen, key) || PySet_Add(seen, key)) {
            if (!_PyErr_Occurred(tstate)) {
                // Seen it before!
                _PyErr_Format(tstate, PyExc_ValueError,
                              "mapping pattern checks duplicate key (%R)", key);
            }
            goto fail;
        }
        PyObject *args[] = { map, key, dummy };
        PyObject *value = NULL;
        if (meth_found) {
            value = PyObject_Vectorcall(get, args, 3, NULL);
        }
        else {
            value = PyObject_Vectorcall(get, &args[1], 2, NULL);
        }
        if (value == NULL) {
            goto fail;
        }
        if (value == dummy) {
            // key not in map!
            Py_DECREF(value);
            Py_DECREF(values);
            // Return None:
            values = Py_NewRef(Py_None);
            goto done;
        }
        PyTuple_SET_ITEM(values, i, value);
    }
    // Success:
done:
    _PyThreadState_PopCStackRef(tstate, &cref);
    Py_DECREF(seen);
    Py_DECREF(dummy);
    return values;
fail:
    _PyThreadState_PopCStackRef(tstate, &cref);
    Py_XDECREF(seen);
    Py_XDECREF(dummy);
    Py_XDECREF(values);
    return NULL;
}

// Extract a named attribute from the subject, with additional bookkeeping to
// raise TypeErrors for repeated lookups. On failure, return NULL (with no
// error set). Use _PyErr_Occurred(tstate) to disambiguate.
static PyObject *
match_class_attr(PyThreadState *tstate, PyObject *subject, PyObject *type,
                 PyObject *name, PyObject *seen)
{
    assert(PyUnicode_CheckExact(name));
    assert(PySet_CheckExact(seen));
    if (PySet_Contains(seen, name) || PySet_Add(seen, name)) {
        if (!_PyErr_Occurred(tstate)) {
            // Seen it before!
            _PyErr_Format(tstate, PyExc_TypeError,
                          "%s() got multiple sub-patterns for attribute %R",
                          ((PyTypeObject*)type)->tp_name, name);
        }
        return NULL;
    }
    PyObject *attr;
    (void)PyObject_GetOptionalAttr(subject, name, &attr);
    return attr;
}

// On success (match), return a tuple of extracted attributes. On failure (no
// match), return NULL. Use _PyErr_Occurred(tstate) to disambiguate.
PyObject*
_PyEval_MatchClass(PyThreadState *tstate, PyObject *subject, PyObject *type,
                   Py_ssize_t nargs, PyObject *kwargs)
{
    if (!PyType_Check(type)) {
        const char *e = "called match pattern must be a class";
        _PyErr_Format(tstate, PyExc_TypeError, e);
        return NULL;
    }
    assert(PyTuple_CheckExact(kwargs));
    // First, an isinstance check:
    if (PyObject_IsInstance(subject, type) <= 0) {
        return NULL;
    }
    // So far so good:
    PyObject *seen = PySet_New(NULL);
    if (seen == NULL) {
        return NULL;
    }
    PyObject *attrs = PyList_New(0);
    if (attrs == NULL) {
        Py_DECREF(seen);
        return NULL;
    }
    // NOTE: From this point on, goto fail on failure:
    PyObject *match_args = NULL;
    // First, the positional subpatterns:
    if (nargs) {
        int match_self = 0;
        if (PyObject_GetOptionalAttr(type, &_Py_ID(__match_args__), &match_args) < 0) {
            goto fail;
        }
        if (match_args) {
            if (!PyTuple_CheckExact(match_args)) {
                const char *e = "%s.__match_args__ must be a tuple (got %s)";
                _PyErr_Format(tstate, PyExc_TypeError, e,
                              ((PyTypeObject *)type)->tp_name,
                              Py_TYPE(match_args)->tp_name);
                goto fail;
            }
        }
        else {
            // _Py_TPFLAGS_MATCH_SELF is only acknowledged if the type does not
            // define __match_args__. This is natural behavior for subclasses:
            // it's as if __match_args__ is some "magic" value that is lost as
            // soon as they redefine it.
            match_args = PyTuple_New(0);
            match_self = PyType_HasFeature((PyTypeObject*)type,
                                            _Py_TPFLAGS_MATCH_SELF);
        }
        assert(PyTuple_CheckExact(match_args));
        Py_ssize_t allowed = match_self ? 1 : PyTuple_GET_SIZE(match_args);
        if (allowed < nargs) {
            const char *plural = (allowed == 1) ? "" : "s";
            _PyErr_Format(tstate, PyExc_TypeError,
                          "%s() accepts %d positional sub-pattern%s (%d given)",
                          ((PyTypeObject*)type)->tp_name,
                          allowed, plural, nargs);
            goto fail;
        }
        if (match_self) {
            // Easy. Copy the subject itself, and move on to kwargs.
            if (PyList_Append(attrs, subject) < 0) {
                goto fail;
            }
        }
        else {
            for (Py_ssize_t i = 0; i < nargs; i++) {
                PyObject *name = PyTuple_GET_ITEM(match_args, i);
                if (!PyUnicode_CheckExact(name)) {
                    _PyErr_Format(tstate, PyExc_TypeError,
                                  "__match_args__ elements must be strings "
                                  "(got %s)", Py_TYPE(name)->tp_name);
                    goto fail;
                }
                PyObject *attr = match_class_attr(tstate, subject, type, name,
                                                  seen);
                if (attr == NULL) {
                    goto fail;
                }
                if (PyList_Append(attrs, attr) < 0) {
                    Py_DECREF(attr);
                    goto fail;
                }
                Py_DECREF(attr);
            }
        }
        Py_CLEAR(match_args);
    }
    // Finally, the keyword subpatterns:
    for (Py_ssize_t i = 0; i < PyTuple_GET_SIZE(kwargs); i++) {
        PyObject *name = PyTuple_GET_ITEM(kwargs, i);
        PyObject *attr = match_class_attr(tstate, subject, type, name, seen);
        if (attr == NULL) {
            goto fail;
        }
        if (PyList_Append(attrs, attr) < 0) {
            Py_DECREF(attr);
            goto fail;
        }
        Py_DECREF(attr);
    }
    Py_SETREF(attrs, PyList_AsTuple(attrs));
    Py_DECREF(seen);
    return attrs;
fail:
    // We really don't care whether an error was raised or not... that's our
    // caller's problem. All we know is that the match failed.
    Py_XDECREF(match_args);
    Py_DECREF(seen);
    Py_DECREF(attrs);
    return NULL;
}


static int do_raise(PyThreadState *tstate, PyObject *exc, PyObject *cause);

PyObject *
PyEval_EvalCode(PyObject *co, PyObject *globals, PyObject *locals)
{
    PyThreadState *tstate = _PyThreadState_GET();
    if (locals == NULL) {
        locals = globals;
    }
    PyObject *builtins = _PyDict_LoadBuiltinsFromGlobals(globals);
    if (builtins == NULL) {
        return NULL;
    }
    PyFrameConstructor desc = {
        .fc_globals = globals,
        .fc_builtins = builtins,
        .fc_name = ((PyCodeObject *)co)->co_name,
        .fc_qualname = ((PyCodeObject *)co)->co_name,
        .fc_code = co,
        .fc_defaults = NULL,
        .fc_kwdefaults = NULL,
        .fc_closure = NULL
    };
    PyFunctionObject *func = _PyFunction_FromConstructor(&desc);
    _Py_DECREF_BUILTINS(builtins);
    if (func == NULL) {
        return NULL;
    }
    EVAL_CALL_STAT_INC(EVAL_CALL_LEGACY);
    PyObject *res = _PyEval_Vector(tstate, func, locals, NULL, 0, NULL);
    Py_DECREF(func);
    return res;
}


/* Interpreter main loop */

PyObject *
PyEval_EvalFrame(PyFrameObject *f)
{
    /* Function kept for backward compatibility */
    PyThreadState *tstate = _PyThreadState_GET();
    return _PyEval_EvalFrame(tstate, f->f_frame, 0);
}

PyObject *
PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
{
    PyThreadState *tstate = _PyThreadState_GET();
    return _PyEval_EvalFrame(tstate, f->f_frame, throwflag);
}

#include "ceval_macros.h"

int _Py_CheckRecursiveCallPy(
    PyThreadState *tstate)
{
    if (tstate->recursion_headroom) {
        if (tstate->py_recursion_remaining < -50) {
            /* Overflowing while handling an overflow. Give up. */
            Py_FatalError("Cannot recover from Python stack overflow.");
        }
    }
    else {
        if (tstate->py_recursion_remaining <= 0) {
            tstate->recursion_headroom++;
            _PyErr_Format(tstate, PyExc_RecursionError,
                        "maximum recursion depth exceeded");
            tstate->recursion_headroom--;
            return -1;
        }
    }
    return 0;
}

static const _Py_CODEUNIT _Py_INTERPRETER_TRAMPOLINE_INSTRUCTIONS[] = {
    /* Put a NOP at the start, so that the IP points into
    * the code, rather than before it */
    { .op.code = NOP, .op.arg = 0 },
    { .op.code = INTERPRETER_EXIT, .op.arg = 0 },  /* reached on return */
    { .op.code = NOP, .op.arg = 0 },
    { .op.code = INTERPRETER_EXIT, .op.arg = 0 },  /* reached on yield */
    { .op.code = RESUME, .op.arg = RESUME_OPARG_DEPTH1_MASK | RESUME_AT_FUNC_START }
};

#ifdef Py_DEBUG
extern void _PyUOpPrint(const _PyUOpInstruction *uop);
#endif


/* Disable unused label warnings.  They are handy for debugging, even
   if computed gotos aren't used. */

/* TBD - what about other compilers? */
#if defined(__GNUC__) || defined(__clang__)
#  pragma GCC diagnostic push
#  pragma GCC diagnostic ignored "-Wunused-label"
#elif defined(_MSC_VER) /* MS_WINDOWS */
#  pragma warning(push)
#  pragma warning(disable:4102)
#endif


PyObject **
_PyObjectArray_FromStackRefArray(_PyStackRef *input, Py_ssize_t nargs, PyObject **scratch)
{
    PyObject **result;
    if (nargs > MAX_STACKREF_SCRATCH) {
        // +1 in case PY_VECTORCALL_ARGUMENTS_OFFSET is set.
        result = PyMem_Malloc((nargs + 1) * sizeof(PyObject *));
        if (result == NULL) {
            return NULL;
        }
        result++;
    }
    else {
        result = scratch;
    }
    for (int i = 0; i < nargs; i++) {
        result[i] = PyStackRef_AsPyObjectBorrow(input[i]);
    }
    return result;
}

void
_PyObjectArray_Free(PyObject **array, PyObject **scratch)
{
    if (array != scratch) {
        PyMem_Free(array);
    }
}


/* _PyEval_EvalFrameDefault is too large to optimize for speed with PGO on MSVC.
 */
#if (defined(_MSC_VER) && \
     (_MSC_VER < 1943) && \
     defined(_Py_USING_PGO))
#define DO_NOT_OPTIMIZE_INTERP_LOOP
#endif

#ifdef DO_NOT_OPTIMIZE_INTERP_LOOP
#  pragma optimize("t", off)
/* This setting is reversed below following _PyEval_EvalFrameDefault */
#endif

#if Py_TAIL_CALL_INTERP
#include "opcode_targets.h"
#include "generated_cases.c.h"
#endif

#if (defined(__GNUC__) && __GNUC__ >= 10 && !defined(__clang__)) && defined(__x86_64__)
/*
 * gh-129987: The SLP autovectorizer can cause poor code generation for
 * opcode dispatch in some GCC versions (observed in GCCs 12 through 15,
 * probably caused by https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115777),
 * negating any benefit we get from vectorization elsewhere in the
 * interpreter loop. Disabling it significantly affected older GCC versions
 * (prior to GCC 9, 40% performance drop), so we have to selectively disable
 * it.
 */
#define DONT_SLP_VECTORIZE __attribute__((optimize ("no-tree-slp-vectorize")))