PEP: 670 Title: Convert macros to functions in the Python C API Author: Erlend Egeberg Aasland , Victor Stinner Status: Draft Type: Standards Track Content-Type: text/x-rst Created: 19-Oct-2021 Python-Version: 3.11 Abstract ======== Convert macros to static inline functions or regular functions to avoid macro pitfalls. Convert macros and static inline functions to regular functions to make them usable by Python extensions which cannot use macros or static inline functions, like extensions written in a programming languages other than C or C++. Function arguments of pointer types are still cast and return types are not changed to prevent emitting new compiler warnings. Macros which can be used as l-value in an assignment are not converted to functions to avoid introducing incompatible changes. Rationale ========= The use of macros may have unintended adverse effects that are hard to avoid, even for experienced C developers. Some issues have been known for years, while others have been discovered recently in Python. Working around macro pitfalls makes the macro coder harder to read and to maintain. Converting macros to functions has multiple advantages: * By design, functions don't have macro pitfalls. * Arguments type and return type are well defined. * Debuggers and profilers can retrieve the name of inlined functions. * Debuggers can put breakpoints on inlined functions. * Variables have a well defined scope. * Code is usually easier to read and to maintain than similar macro code. Functions don't need the following workarounds for macro pitfalls: * Add parentheses around arguments. * Use line continuation characters if the function is written on multiple lines. * Add commas to execute multiple expressions. * Use ``do { ... } while (0)`` to write multiple statements. Converting macros and static inline functions to regular functions makes these regular functions accessible to projects which use Python but cannot use macros and static inline functions. Macro Pitfalls ============== The `GCC documentation `_ lists several common macro pitfalls: - Misnesting - Operator precedence problems - Swallowing the semicolon - Duplication of side effects - Self-referential macros - Argument prescan - Newlines in arguments Performance and inlining ======================== Static inline functions is a feature added to the C99 standard. Modern C compilers have efficient heuristics to decide if a function should be inlined or not. When a C compiler decides to not inline, there is likely a good reason. For example, inlining would reuse a register which requires to save/restore the register value on the stack and so increases the stack memory usage, or be less efficient. Debug build ----------- Benchmarks must not be run on a Python debug build, only on release build. Moreover, using LTO and PGO optimizations is recommended for best performances and reliable benchmarks. PGO helps the compiler to decide if function should be inlined or not. ``./configure --with-pydebug`` uses the ``-Og`` compiler option if it's supported by the compiler (GCC and LLVM Clang support it): optimize debugging experience. Otherwise, the ``-O0`` compiler option is used: disable most optimizations. With GCC 11, ``gcc -Og`` can inline static inline functions, whereas ``gcc -O0`` does not inline static inline functions. Examples: * Call ``Py_INCREF()`` in ``PyBool_FromLong()``: * ``gcc -Og``: inlined * ``gcc -O0``: not inlined, call ``Py_INCREF()`` function * Call ``_PyErr_Occurred()`` in ``_Py_CheckFunctionResult()``: * ``gcc -Og``: inlined * ``gcc -O0``: not inlined, call ``_PyErr_Occurred()`` function On Windows, when Python is built in debug mode by Visual Studio, static inline functions are not inlined. Force inlining -------------- The ``Py_ALWAYS_INLINE`` macro can be used to force inlining. This macro uses ``__attribute__((always_inline))`` with GCC and Clang, and ``__forceinline`` with MSC. Previous attempts to use ``Py_ALWAYS_INLINE`` didn't show any benefit, and were abandoned. See for example: `bpo-45094 `_: "Consider using ``__forceinline`` and ``__attribute__((always_inline))`` on static inline functions (``Py_INCREF``, ``Py_TYPE``) for debug build". When the ``Py_INCREF()`` macro was converted to a static inline functions in 2018 (`commit `__), it was decided not to force inlining. The machine code was analyzed with multiple C compilers and compiler options: ``Py_INCREF()`` was always inlined without having to force inlining. The only case where it was not inlined was the debug build. See discussion in the `bpo-35059 `_: "Convert ``Py_INCREF()`` and ``PyObject_INIT()`` to inlined functions". Disable inlining ---------------- On the other side, the ``Py_NO_INLINE`` macro can be used to disable inlining. It can be used to reduce the stack memory usage, or to prevent inlining on LTO+PGO builds, which are generally more aggressive to inline code: see `bpo-33720 `_. The ``Py_NO_INLINE`` macro uses ``__attribute__ ((noinline))`` with GCC and Clang, and ``__declspec(noinline)`` with MSC. Specification ============= Convert macros to static inline functions ----------------------------------------- Most macros should be converted to static inline functions to prevent `macro pitfalls`_. The following macros should not be converted: * Empty macros. Example: ``#define Py_HAVE_CONDVAR``. * Macros only defining a number, even if a constant with a well defined type can better. Example: ``#define METH_VARARGS 0x0001``. * Compatibility layer for different C compilers, C language extensions, or recent C features. Example: ``#define Py_ALWAYS_INLINE __attribute__((always_inline))``. * Macros that need C preprocessor features, like stringification and concatenation. Example: ``Py_STRINGIFY()``. * Macros which can be used as l-value in an assignment. This change is an incompatible change and is out of the scope of this PEP. Example: ``PyBytes_AS_STRING()``. * Macros having different return types depending on the code path. Convert static inline functions to regular functions ---------------------------------------------------- The performance impact of converting static inline functions to regular functions should be measured with benchmarks. If there is a significant slowdown, there should be a good reason to do the conversion. One reason can be hiding implementation details. To avoid any risk of performance slowdown on Python built without LTO, it is possible to keep a private static inline function in the internal C API and use it in Python, but expose a regular function in the public C API. Using static inline functions in the internal C API is fine: the internal C API exposes implementation details by design and should not be used outside Python. Cast pointer arguments ---------------------- Existing cast ''''''''''''' Currently, most macros accepting pointers cast pointer arguments to their expected types. For example, in Python 3.6, the ``Py_TYPE()`` macro casts its argument to ``PyObject*``:: #define Py_TYPE(ob) (((PyObject*)(ob))->ob_type) The ``Py_TYPE()`` macro accepts the ``PyObject*`` type, but also any pointer types, such as ``PyLongObject*`` and ``PyDictObject*``. Add a new macro to keep the cast '''''''''''''''''''''''''''''''' When a macro is converted to a function and the macro casts at least one of its arguments, a new macro is added to keep the cast. The new macro and the function have the same name. Example with the ``Py_TYPE()`` macro converted to a static inline function:: static inline PyTypeObject* Py_TYPE(PyObject *ob) { return ob->ob_type; } #define Py_TYPE(ob) Py_TYPE((PyObject*)(ob)) The cast is kept for all pointer types, not only ``PyObject*``. Removing a cast to ``void*`` would emit a new warning if the function is called with a variable of ``const void*`` type. For example, the ``PyUnicode_WRITE()`` macro casts its *data* argument to ``void*``, and so accepts ``const void*`` type, even if it writes into *data*. Avoid the cast in the limited C API version 3.11 '''''''''''''''''''''''''''''''''''''''''''''''' The cast is removed from the limited C API version 3.11 and newer: the caller must pass the expected type, or perform the cast. An example with the ``Py_TYPE()`` function:: static inline PyTypeObject* Py_TYPE(PyObject *ob) { return ob->ob_type; } #if !defined(Py_LIMITED_API) || Py_LIMITED_API+0 < 0x030b0000 # define Py_TYPE(ob) Py_TYPE((PyObject*)(ob)) #endif Return type is not changed -------------------------- When a macro is converted to a function, its return type must not change to prevent emitting new compiler warnings. For example, Python 3.7 changed ``PyUnicode_AsUTF8()`` return type from ``char*`` to ``const char*`` (`commit `__). The change emitted new compiler warnings when building C extensions expecting ``char*``. This PEP doesn't change the return type to prevent this issue. Backwards Compatibility ======================= The PEP is designed to avoid C API incompatible changes. Only C extensions explicitly targeting the limited C API version 3.11 must now pass the expected types to functions: pointer arguments are no longer cast to the expected types. Function arguments of pointer types are still cast and return types are not changed to prevent emitting new compiler warnings. Macros which can be used as l-value in an assignment are not modified by this PEP to avoid incompatible changes. Rejected Ideas ============== Keep macros, but fix some macro issues -------------------------------------- Macros are always "inlined" with any C compiler. The duplication of side effects can be worked around in the caller of the macro. People using macros should be considered "consenting adults". People who feel unsafe with macros should simply not use them. These ideas are rejected because macros _are_ error prone, and it is too easy to miss a macro pitfall when writing and reviewing macro code. Moreover, macros are harder to read and maintain than functions. Examples of Macro Pitfalls ========================== Duplication of side effects --------------------------- Macros:: #define PySet_Check(ob) \ (Py_IS_TYPE(ob, &PySet_Type) \ || PyType_IsSubtype(Py_TYPE(ob), &PySet_Type)) #define Py_IS_NAN(X) ((X) != (X)) If the *op* or the *X* argument has a side effect, the side effect is duplicated: it executed twice by ``PySet_Check()`` and ``Py_IS_NAN()``. For example, the ``pos++`` argument in the ``PyUnicode_WRITE(kind, data, pos++, ch)`` code has a side effect. This code is safe because the ``PyUnicode_WRITE()`` macro only uses its 3rd argument once and so does not duplicate ``pos++`` side effect. Misnesting ---------- Example of the `bpo-43181: Python macros don't shield arguments `_. The ``PyObject_TypeCheck()`` macro before it has been fixed:: #define PyObject_TypeCheck(ob, tp) \ (Py_IS_TYPE(ob, tp) || PyType_IsSubtype(Py_TYPE(ob), (tp))) C++ usage example:: PyObject_TypeCheck(ob, U(f(c))) The preprocessor first expands it:: (Py_IS_TYPE(ob, f(c)) || ...) C++ ``"<"`` and ``">"`` characters are not treated as brackets by the preprocessor, so the ``Py_IS_TYPE()`` macro is invoked with 3 arguments: * ``ob`` * ``f(c)`` The compilation fails with an error on ``Py_IS_TYPE()`` which only takes 2 arguments. The bug is that the *op* and *tp* arguments of ``PyObject_TypeCheck()`` must be put between parentheses: replace ``Py_IS_TYPE(ob, tp)`` with ``Py_IS_TYPE((ob), (tp))``. In regular C code, these parentheses are redundant, can be seen as a bug, and so are often forgotten when writing macros. To avoid Macro Pitfalls, the ``PyObject_TypeCheck()`` macro has been converted to a static inline function: `commit `__. Examples of hard to read macros =============================== PyObject_INIT() --------------- Example showing the usage of commas in a macro which has a return value. Python 3.7 macro:: #define PyObject_INIT(op, typeobj) \ ( Py_TYPE(op) = (typeobj), _Py_NewReference((PyObject *)(op)), (op) ) Python 3.8 function (simplified code):: static inline PyObject* _PyObject_INIT(PyObject *op, PyTypeObject *typeobj) { Py_TYPE(op) = typeobj; _Py_NewReference(op); return op; } #define PyObject_INIT(op, typeobj) \ _PyObject_INIT(_PyObject_CAST(op), (typeobj)) * The function doesn't need the line continuation character ``"\"``. * It has an explicit ``"return op;"`` rather than the surprising ``", (op)"`` syntax at the end of the macro. * It uses short statements on multiple lines, rather than being written as a single long line. * Inside the function, the *op* argument has the well defined type ``PyObject*`` and so doesn't need casts like ``(PyObject *)(op)``. * Arguments don't need to be put inside parentheses: use ``typeobj``, rather than ``(typeobj)``. _Py_NewReference() ------------------ Example showing the usage of an ``#ifdef`` inside a macro. Python 3.7 macro (simplified code):: #ifdef COUNT_ALLOCS # define _Py_INC_TPALLOCS(OP) inc_count(Py_TYPE(OP)) # define _Py_COUNT_ALLOCS_COMMA , #else # define _Py_INC_TPALLOCS(OP) # define _Py_COUNT_ALLOCS_COMMA #endif /* COUNT_ALLOCS */ #define _Py_NewReference(op) ( \ _Py_INC_TPALLOCS(op) _Py_COUNT_ALLOCS_COMMA \ Py_REFCNT(op) = 1) Python 3.8 function (simplified code):: static inline void _Py_NewReference(PyObject *op) { _Py_INC_TPALLOCS(op); Py_REFCNT(op) = 1; } PyUnicode_READ_CHAR() --------------------- This macro reuses arguments, and possibly calls ``PyUnicode_KIND`` multiple times:: #define PyUnicode_READ_CHAR(unicode, index) \ (assert(PyUnicode_Check(unicode)), \ assert(PyUnicode_IS_READY(unicode)), \ (Py_UCS4) \ (PyUnicode_KIND((unicode)) == PyUnicode_1BYTE_KIND ? \ ((const Py_UCS1 *)(PyUnicode_DATA((unicode))))[(index)] : \ (PyUnicode_KIND((unicode)) == PyUnicode_2BYTE_KIND ? \ ((const Py_UCS2 *)(PyUnicode_DATA((unicode))))[(index)] : \ ((const Py_UCS4 *)(PyUnicode_DATA((unicode))))[(index)] \ ) \ )) Possible implementation as a static inlined function:: static inline Py_UCS4 PyUnicode_READ_CHAR(PyObject *unicode, Py_ssize_t index) { assert(PyUnicode_Check(unicode)); assert(PyUnicode_IS_READY(unicode)); switch (PyUnicode_KIND(unicode)) { case PyUnicode_1BYTE_KIND: return (Py_UCS4)((const Py_UCS1 *)(PyUnicode_DATA(unicode)))[index]; case PyUnicode_2BYTE_KIND: return (Py_UCS4)((const Py_UCS2 *)(PyUnicode_DATA(unicode)))[index]; case PyUnicode_4BYTE_KIND: default: return (Py_UCS4)((const Py_UCS4 *)(PyUnicode_DATA(unicode)))[index]; } } Macros converted to functions since Python 3.8 ============================================== List of macros already converted to functions between Python 3.8 and Python 3.11 showing that these conversions didn't not impact the Python performance and didn't break the backward compatibility, even if some converted macros are very commonly used by C extensions like ``Py_INCREF()``. Macros converted to static inline functions ------------------------------------------- Python 3.8: * ``Py_DECREF()`` * ``Py_INCREF()`` * ``Py_XDECREF()`` * ``Py_XINCREF()`` * ``PyObject_INIT()`` * ``PyObject_INIT_VAR()`` * ``_PyObject_GC_UNTRACK()`` * ``_Py_Dealloc()`` Macros converted to regular functions ------------------------------------- Python 3.9: * ``PyIndex_Check()`` * ``PyObject_CheckBuffer()`` * ``PyObject_GET_WEAKREFS_LISTPTR()`` * ``PyObject_IS_GC()`` * ``PyObject_NEW()``: alias to ``PyObject_New()`` * ``PyObject_NEW_VAR()``: alias to ``PyObjectVar_New()`` To avoid any risk of performance slowdown on Python built without LTO, private static inline functions have been added to the internal C API: * ``_PyIndex_Check()`` * ``_PyObject_IS_GC()`` * ``_PyType_HasFeature()`` * ``_PyType_IS_GC()`` Static inline functions converted to regular functions ------------------------------------------------------- Python 3.11: * ``PyObject_CallOneArg()`` * ``PyObject_Vectorcall()`` * ``PyVectorcall_Function()`` * ``_PyObject_FastCall()`` To avoid any risk of performance slowdown on Python built without LTO, a private static inline function has been added to the internal C API: * ``_PyVectorcall_FunctionInline()`` Incompatible changes -------------------- While other converted macros didn't break the backward compatibility, there is an exception. The 3 macros ``Py_REFCNT()``, ``Py_TYPE()`` and ``Py_SIZE()`` have been converted to static inline functions in Python 3.10 and 3.11 to disallow using them as l-value in assignment. It is an incompatible change made on purpose: see `bpo-39573 `_ for the rationale. This PEP does not convert macros which can be used as l-value to avoid introducing incompatible changes. Benchmark comparing macros and static inline functions ====================================================== Benchmark run on Fedora 35 (Linux) with GCC 11 on a laptop with 8 logical CPUs (4 physical CPU cores). The `PR 29728 `_ replaces existing the following static inline functions with macros: * ``PyObject_TypeCheck()`` * ``PyType_Check()``, ``PyType_CheckExact()`` * ``PyType_HasFeature()`` * ``PyVectorcall_NARGS()`` * ``Py_DECREF()``, ``Py_XDECREF()`` * ``Py_INCREF()``, ``Py_XINCREF()`` * ``Py_IS_TYPE()`` * ``Py_NewRef()`` * ``Py_REFCNT()``, ``Py_TYPE()``, ``Py_SIZE()`` When static inline functions are inlined: Release build ------------------------------------------------------- Benchmark of the ``./python -m test -j5`` command on Python built in release mode with ``gcc -O3``, LTO and PGO: * Macros (PR 29728): 361 sec +- 1 sec * Static inline functions (reference): 361 sec +- 1 sec There is **no significant performance difference** between macros and static inline functions when static inline functions **are inlined**. When static inline functions are not inlined: Debug build and -O0 ----------------------------------------------------------------- Benchmark of the ``./python -m test -j10`` command on Python built in debug mode with ``gcc -O0`` (explicitly disable compiler optimizations): * Macros (PR 29728): 345 sec ± 5 sec * Static inline functions (reference): 360 sec ± 6 sec Replacing macros with static inline functions makes Python **1.04x slower** when the compiler **does not inline** static inline functions. Post History ============ * python-dev: `PEP 670: Convert macros to functions in the Python C API `_ (October 2021) References ========== * `bpo-45490 `_: [meta][C API] Avoid C macro pitfalls and usage of static inline functions (October 2021). * `What to do with unsafe macros `_ (March 2021). * `bpo-43502 `_: [C-API] Convert obvious unsafe macros to static inline functions (March 2021). Version History =============== * Version 2: No longer remove return values; remove argument casting from the limited C API. * Version 1: First public version Copyright ========= This document is placed in the public domain or under the CC0-1.0-Universal license, whichever is more permissive.