python-peps/pep-0573.rst

PEP: 573
Title: Module State Access from C Extension Methods
Version: $Revision$
Last-Modified: $Date$
Author: Petr Viktorin <encukou@gmail.com>,
        Nick Coghlan <ncoghlan@gmail.com>,
        Eric Snow <ericsnowcurrently@gmail.com>,
        Marcel Plch <gmarcel.plch@gmail.com>
BDFL-Delegate: Stefan Behnel
Discussions-To: import-sig@python.org
Status: Final
Type: Standards Track
Content-Type: text/x-rst
Created: 02-Jun-2016
Python-Version: 3.9
Post-History:


Abstract
========

This PEP proposes to add a way for CPython extension methods to access context,
such as the state of the modules they are defined in.

This will allow extension methods to use direct pointer dereferences
rather than PyState_FindModule for looking up module state, reducing or
eliminating the performance cost of using module-scoped state over process
global state.

This fixes one of the remaining roadblocks for adoption of :pep:`3121` (Extension
module initialization and finalization) and :pep:`489`
(Multi-phase extension module initialization).

While this PEP takes an additional step towards fully solving the problems that
:pep:`3121` and :pep:`489` started tackling, it does not attempt to resolve *all*
remaining concerns. In particular, access to the module state
from slot methods (``nb_add``, etc) is not solved.


Terminology
===========

Process-Global State
--------------------

C-level static variables. Since this is very low-level
memory storage, it must be managed carefully.

Per-module State
----------------

State local to a module object, allocated dynamically as part of a
module object's initialization. This isolates the state from other
instances of the module (including those in other subinterpreters).

Accessed by ``PyModule_GetState()``.


Static Type
-----------

A type object defined as a C-level static variable, i.e. a compiled-in type
object.

A static type needs to be shared between module instances and has no
information of what module it belongs to.
Static types do not have ``__dict__`` (although their instances might).


Heap Type
---------

A type object created at run time.


Defining Class
--------------

The defining class of a method (either bound or unbound) is the class on which
the method was defined.
A class that merely inherits the method from its base is not the defining class.

For example, ``int`` is the defining class of ``True.to_bytes``,
``True.__floor__`` and ``int.__repr__``.

In C, the defining class is the one defined with the corresponding
``tp_methods`` or "tp slots" [#tp-slots]_ entry.
For methods defined in Python, the defining class is saved in the
``__class__`` closure cell.


C-API
-----

The "Python/C API" as described in Python documentation.
CPython implements the C-API, but other implementations exist.


Rationale
=========

:pep:`489` introduced a new way to initialize extension modules, which brings
several advantages to extensions that implement it:

* The extension modules behave more like their Python counterparts.
* The extension modules can easily support loading into pre-existing
  module objects, which paves the way for extension module support for
  ``runpy`` or for systems that enable extension module reloading.
* Loading multiple modules from the same extension is possible, which
  makes it possible to test module isolation (a key feature for proper
  sub-interpreter support) from a single interpreter.

The biggest hurdle for adoption of :pep:`489` is allowing access to module state
from methods of extension types.
Currently, the way to access this state from extension methods is by looking up
the module via ``PyState_FindModule`` (in contrast to module level functions in
extension modules, which receive a module reference as an argument).
However, ``PyState_FindModule`` queries the thread-local state, making it
relatively costly compared to C level process global access and consequently
deterring module authors from using it.

Also, ``PyState_FindModule`` relies on the assumption that in each
subinterpreter, there is at most one module corresponding to
a given ``PyModuleDef``.  This assumption does not hold for modules that use
:pep:`489`'s multi-phase initialization, so ``PyState_FindModule`` is unavailable
for these modules.

A faster, safer way of accessing module-level state from extension methods
is needed.


Background
===========

The implementation of a Python method may need access to one or more of
the following pieces of information:

* The instance it is called on (``self``)
* The underlying function
* The *defining class*, i. e. the class the method was defined in
* The corresponding module
* The module state

In Python code, the Python-level equivalents may be retrieved as::

    import sys

    class Foo:
        def meth(self):
            instance = self
            module_globals = globals()
            module_object = sys.modules[__name__]  # (1)
            underlying_function = Foo.meth         # (1)
            defining_class = Foo                   # (1)
            defining_class = __class__             # (2)

.. note::

    The defining class is not ``type(self)``, since ``type(self)`` might
    be a subclass of ``Foo``.

The statements marked (1) implicitly rely on name-based lookup via the
function's ``__globals__``: either the ``Foo`` attribute to access the defining
class and Python function object, or ``__name__`` to find the module object in
``sys.modules``.

In Python code, this is feasible, as ``__globals__`` is set appropriately when
the function definition is executed, and even if the namespace has been
manipulated to return a different object, at worst an exception will be raised.

The ``__class__`` closure, (2), is a safer way to get the defining class, but it
still relies on ``__closure__`` being set appropriately.

By contrast, extension methods are typically implemented as normal C functions.
This means that they only have access to their arguments and C level thread-local
and process-global states. Traditionally, many extension modules have stored
their shared state in C-level process globals, causing problems when:

* running multiple initialize/finalize cycles in the same process
* reloading modules (e.g. to test conditional imports)
* loading extension modules in subinterpreters

:pep:`3121` attempted to resolve this by offering the ``PyState_FindModule`` API,
but this still has significant problems when it comes to extension methods
(rather than module level functions):

* it is markedly slower than directly accessing C-level process-global state
* there is still some inherent reliance on process global state that means it
  still doesn't reliably handle module reloading

It's also the case that when looking up a C-level struct such as module state,
supplying an unexpected object layout can crash the interpreter, so it's
significantly more important to ensure that extension methods receive the kind
of object they expect.


Proposal
========

Currently, a bound extension method (``PyCFunction`` or
``PyCFunctionWithKeywords``) receives only ``self``, and (if applicable) the
supplied positional and keyword arguments.

While module-level extension functions already receive access to the defining
module object via their ``self`` argument, methods of extension types don't have
that luxury: they receive the bound instance via ``self``, and hence have no
direct access to the defining class or the module level state.

The additional module level context described above can be made available with
two changes.
Both additions are optional; extension authors need to opt in to start
using them:

* Add a pointer to the module to heap type objects.

* Pass the defining class to the underlying C function.

  In CPython, the defining class is readily available at the time the built-in
  method object (``PyCFunctionObject``) is created, so it can be stored
  in a new struct that extends ``PyCFunctionObject``.

The module state can then be retrieved from the module object via
``PyModule_GetState``.

Note that this proposal implies that any type whose methods need to access
`per-module state`_ must be a heap type, rather than a static type. This is
necessary to support loading multiple module objects from a single
extension: a static type, as a C-level global, has no information about
which module object it belongs to.


Slot methods
------------

The above changes don't cover slot methods, such as ``tp_iter`` or ``nb_add``.

The problem with slot methods is that their C API is fixed, so we can't
simply add a new argument to pass in the defining class.
Two possible solutions have been proposed to this problem:

* Look up the class through walking the MRO.
  This is potentially expensive, but will be usable if performance is not
  a problem (such as when raising a module-level exception).
* Storing a pointer to the defining class of each slot in a separate table,
  ``__typeslots__`` [#typeslots-mail]_.  This is technically feasible and fast,
  but quite invasive.

Modules affected by this concern also have the option of using
`thread-local state`_ or `PEP 567 context variables`_ as a caching mechanism, or
else defining their own reload-friendly lookup caching scheme.

Solving the issue generally is deferred to a future PEP.

.. _thread-local state: https://docs.python.org/3/c-api/init.html#thread-local-storage-support
.. _PEP 567 context variables: https://docs.python.org/3/c-api/contextvars.html


Specification
=============

Adding module references to heap types
--------------------------------------

A new factory method will be added to the C-API for creating modules::

    PyObject* PyType_FromModuleAndSpec(PyObject *module,
                                       PyType_Spec *spec,
                                       PyObject *bases)

This acts the same as ``PyType_FromSpecWithBases``, and additionally associates
the provided module object with the new type. (In CPython, this will set
``ht_module`` described below.)

Additionally, an accessor, ``PyObject * PyType_GetModule(PyTypeObject *)``
will be provided.
It will return the type's associated module if one is set,
otherwise it will set ``TypeError`` and return NULL.
When given a static type, it will always set ``TypeError`` and return NULL.

To implement this in CPython, the ``PyHeapTypeObject`` struct will get a
new member, ``PyObject *ht_module``, that will store a pointer to the
associated module.
It will be ``NULL`` by default and should not be modified after the type
object is created.

The ``ht_module`` member will not be inherited by subclasses; it needs to be
set using ``PyType_FromSpecWithBases`` for each individual type that needs it.

Usually, creating a class with ``ht_module`` set will create a reference
cycle involving the class and the module.
This is not a problem, as tearing down modules is not a performance-sensitive
operation, and module-level functions typically also create reference cycles.
The existing "set all module globals to None" code that breaks function cycles
through ``f_globals`` will also break the new cycles through ``ht_module``.


Passing the defining class to extension methods
-----------------------------------------------

A new signature flag, ``METH_METHOD``, will be added for use in
``PyMethodDef.ml_flags``. Conceptually, it adds ``defining_class``
to the function signature.
To make the initial implementation easier, the flag can only be used as
``(METH_FASTCALL | METH_KEYWORDS | METH_METHOD)``.
(It can't be used with other flags like ``METH_O`` or bare ``METH_FASTCALL``,
though it may be combined with ``METH_CLASS`` or ``METH_STATIC``).

C functions for methods defined using this flag combination will be called
using a new C signature called ``PyCMethod``::

    PyObject *PyCMethod(PyObject *self,
                        PyTypeObject *defining_class,
                        PyObject *const *args,
                        size_t nargsf,
                        PyObject *kwnames)

Additional combinations like ``(METH_VARARGS | METH_METHOD)`` may be added
in the future (or even in the initial implementation of this PEP).
However, ``METH_METHOD`` should always be an *additional* flag, i.e., the
defining class should only be passed in if needed.

In CPython, a new structure extending ``PyCFunctionObject`` will be added
to hold the extra information::

    typedef struct {
        PyCFunctionObject func;
        PyTypeObject *mm_class; /* Passed as 'defining_class' arg to the C func */
    } PyCMethodObject;

The ``PyCFunction`` implementation will pass ``mm_class`` into a
``PyCMethod`` C function when it finds the ``METH_METHOD`` flag being set.
A new macro ``PyCFunction_GET_CLASS(cls)`` will be added for easier access
to ``mm_class``.

C methods may continue to use the other ``METH_*`` signatures if they do
not require access to their defining class/module.
If ``METH_METHOD`` is not set, casting to ``PyCMethodObject`` is invalid.


Argument Clinic
---------------

To support passing the defining class to methods using Argument Clinic,
a new converter called ``defining_class`` will be added to CPython's Argument
Clinic tool.

Each method may only have one argument using this converter, and it must
appear after ``self``, or, if ``self`` is not used, as the first argument.
The argument will be of type ``PyTypeObject *``.

When used, Argument Clinic will select
``METH_FASTCALL | METH_KEYWORDS | METH_METHOD`` as the calling convention.
The argument will not appear in ``__text_signature__``.

The new converter will initially not be compatible with ``__init__`` and
``__new__`` methods, which cannot use the ``METH_METHOD`` convention.


Helpers
-------

Getting to `per-module state`_ from a heap type is a very common task. To make
this easier, a helper will be added::

    void *PyType_GetModuleState(PyObject *type)

This function takes a heap type and on success, it returns pointer to the state
of the module that the heap type belongs to.

On failure, two scenarios may occur. When a non-type object, or a type without a
module is passed in, ``TypeError`` is set and ``NULL`` returned. If the module
is found, the pointer to the state, which may be ``NULL``, is returned without
setting any exception.


Modules Converted in the Initial Implementation
-----------------------------------------------

To validate the approach, the ``_elementtree`` module will be modified during
the initial implementation.


Summary of API Changes and Additions
====================================

The following will be added to Python C-API:

   * ``PyType_FromModuleAndSpec`` function
   * ``PyType_GetModule`` function
   * ``PyType_GetModuleState`` function
   * ``METH_METHOD`` call flag
   * ``PyCMethod`` function signature

The following additions will be added as CPython implementation details,
and won't be documented:

   * ``PyCFunction_GET_CLASS`` macro
   * ``PyCMethodObject`` struct
   * ``ht_module`` member of ``_heaptypeobject``
   * ``defining_class`` converter in Argument Clinic


Backwards Compatibility
=======================

One new pointer is added to all heap types.
All other changes are adding new functions and structures,
or changes to private implementation details.

Implementation
==============

An initial implementation is available in a Github repository [#gh-repo]_;
a patchset is at [#gh-patch]_.


Possible Future Extensions
==========================

Slot methods
------------

A way of passing defining class (or module state) to slot methods may be
added in the future.

A previous version of this PEP proposed a helper function that would determine
a defining class by searching the MRO for a class that defines a slot to a
particular function. However, this approach would fail if a class is mutated
(which is, for heap types, possible from Python code).
Solving this problem is left to future discussions.


Easy creation of types with module references
---------------------------------------------

It would be possible to add a :pep:`489` execution slot type to make
creating heap types significantly easier than calling
``PyType_FromModuleAndSpec``.
This is left to a future PEP.

It may be good to add a good way to create static exception types from the
limited API. Such exception types could be shared between subinterpreters,
but instantiated without needing specific module state.
This is also left to possible future discussions.


Optimization
------------

As proposed here, methods defined with the ``METH_METHOD`` flag only support
one specific signature.

If it turns out that other signatures are needed for performance reasons,
they may be added.


References
==========

.. [#tp-slots] https://docs.python.org/3/c-api/typeobj.html#tp-slots

.. [#typeslots-mail] [Import-SIG] On singleton modules, heap types, and subinterpreters
   (https://mail.python.org/pipermail/import-sig/2015-July/001035.html)

.. [#gh-repo]
   https://github.com/Dormouse759/cpython/tree/pep-c-rebase_newer

.. [#gh-patch]
   https://github.com/Dormouse759/cpython/compare/master...Dormouse759:pep-c-rebase_newer


Copyright
=========

This document is placed in the public domain or under the
CC0-1.0-Universal license, whichever is more permissive.