Perhaps one of the most important structures of the Python object system is the
structure that defines a new type: the PyTypeObject structure. Type
objects can be handled using any of the PyObject_* or
PyType_* functions, but do not offer much that’s interesting to most
Python applications. These objects are fundamental to how objects behave, so
they are very important to the interpreter itself and to any extension module
that implements new types.
Type objects are fairly large compared to most of the standard types. The reason
for the size is that each type object stores a large number of values, mostly C
function pointers, each of which implements a small part of the type’s
functionality. The fields of the type object are examined in detail in this
section. The fields will be described in the order in which they occur in the
structure.
In addition to the following quick reference, the Examples
section provides at-a-glance insight into the meaning and use of
PyTypeObject.
The structure definition for PyTypeObject can be found in
Include/cpython/object.h. For convenience of reference, this repeats the
definition found there:
typedefstruct_typeobject{PyObject_VAR_HEADconstchar*tp_name;/* For printing, in format "<module>.<name>" */Py_ssize_ttp_basicsize,tp_itemsize;/* For allocation *//* Methods to implement standard operations */destructortp_dealloc;Py_ssize_ttp_vectorcall_offset;getattrfunctp_getattr;setattrfunctp_setattr;PyAsyncMethods*tp_as_async;/* formerly known as tp_compare (Python 2) or tp_reserved (Python 3) */reprfunctp_repr;/* Method suites for standard classes */PyNumberMethods*tp_as_number;PySequenceMethods*tp_as_sequence;PyMappingMethods*tp_as_mapping;/* More standard operations (here for binary compatibility) */hashfunctp_hash;ternaryfunctp_call;reprfunctp_str;getattrofunctp_getattro;setattrofunctp_setattro;/* Functions to access object as input/output buffer */PyBufferProcs*tp_as_buffer;/* Flags to define presence of optional/expanded features */unsignedlongtp_flags;constchar*tp_doc;/* Documentation string *//* Assigned meaning in release 2.0 *//* call function for all accessible objects */traverseproctp_traverse;/* delete references to contained objects */inquirytp_clear;/* Assigned meaning in release 2.1 *//* rich comparisons */richcmpfunctp_richcompare;/* weak reference enabler */Py_ssize_ttp_weaklistoffset;/* Iterators */getiterfunctp_iter;iternextfunctp_iternext;/* Attribute descriptor and subclassing stuff */PyMethodDef*tp_methods;PyMemberDef*tp_members;PyGetSetDef*tp_getset;// Strong reference on a heap type, borrowed reference on a static typePyTypeObject*tp_base;PyObject*tp_dict;descrgetfunctp_descr_get;descrsetfunctp_descr_set;Py_ssize_ttp_dictoffset;initproctp_init;allocfunctp_alloc;newfunctp_new;freefunctp_free;/* Low-level free-memory routine */inquirytp_is_gc;/* For PyObject_IS_GC */PyObject*tp_bases;PyObject*tp_mro;/* method resolution order */PyObject*tp_cache;/* no longer used */void*tp_subclasses;/* for static builtin types this is an index */PyObject*tp_weaklist;/* not used for static builtin types */destructortp_del;/* Type attribute cache version tag. Added in version 2.6. * If zero, the cache is invalid and must be initialized. */unsignedinttp_version_tag;destructortp_finalize;vectorcallfunctp_vectorcall;/* bitset of which type-watchers care about this type */unsignedchartp_watched;/* Number of tp_version_tag values used. * Set to _Py_ATTR_CACHE_UNUSED if the attribute cache is * disabled for this type (e.g. due to custom MRO entries). * Otherwise, limited to MAX_VERSIONS_PER_CLASS (defined elsewhere). */uint16_ttp_versions_used;}PyTypeObject;
The type object structure extends the PyVarObject structure. The
ob_size field is used for dynamic types (created by type_new(),
usually called from a class statement). Note that PyType_Type (the
metatype) initializes tp_itemsize, which means that its instances (i.e.
type objects) must have the ob_size field.
The type object’s reference count is initialized to 1 by the
PyObject_HEAD_INIT macro. Note that for statically allocated type
objects, the type’s instances (objects whose ob_type
points back to the type) do not count as references. But for
dynamically allocated type objects, the instances do
count as references.
This is the type’s type, in other words its metatype. It is initialized by the
argument to the PyObject_HEAD_INIT macro, and its value should normally be
&PyType_Type. However, for dynamically loadable extension modules that must
be usable on Windows (at least), the compiler complains that this is not a valid
initializer. Therefore, the convention is to pass NULL to the
PyObject_HEAD_INIT macro and to initialize this field explicitly at the
start of the module’s initialization function, before doing anything else. This
is typically done like this:
Foo_Type.ob_type=&PyType_Type;
This should be done before any instances of the type are created.
PyType_Ready() checks if ob_type is NULL, and if so,
initializes it to the ob_type field of the base class.
PyType_Ready() will not change this field if it is non-zero.
Each slot has a section describing inheritance. If PyType_Ready()
may set a value when the field is set to NULL then there will also be
a “Default” section. (Note that many fields set on PyBaseObject_Type
and PyType_Type effectively act as defaults.)
Pointer to a NUL-terminated string containing the name of the type. For types
that are accessible as module globals, the string should be the full module
name, followed by a dot, followed by the type name; for built-in types, it
should be just the type name. If the module is a submodule of a package, the
full package name is part of the full module name. For example, a type named
T defined in module M in subpackage Q in package P
should have the tp_name initializer "P.Q.M.T".
For dynamically allocated type objects,
this should just be the type name, and
the module name explicitly stored in the type dict as the value for key
'__module__'.
For statically allocated type objects,
the tp_name field should contain a dot.
Everything before the last dot is made accessible as the __module__
attribute, and everything after the last dot is made accessible as the
__name__ attribute.
If no dot is present, the entire tp_name field is made accessible as the
__name__ attribute, and the __module__ attribute is undefined
(unless explicitly set in the dictionary, as explained above). This means your
type will be impossible to pickle. Additionally, it will not be listed in
module documentations created with pydoc.
This field must not be NULL. It is the only required field
in PyTypeObject() (other than potentially
tp_itemsize).
There are two kinds of types: types with fixed-length instances have a zero
tp_itemsize field, types with variable-length instances have a non-zero
tp_itemsize field. For a type with fixed-length instances, all
instances have the same size, given in tp_basicsize.
(Exceptions to this rule can be made using
PyUnstable_Object_GC_NewWithExtraData().)
For a type with variable-length instances, the instances must have an
ob_size field, and the instance size is
tp_basicsize plus N times tp_itemsize,
where N is the “length” of the object.
Functions like PyObject_NewVar() will take the value of N as an
argument, and store in the instance’s ob_size field.
Note that the ob_size field may later be used for
other purposes. For example, int instances use the bits of
ob_size in an implementation-defined
way; the underlying storage and its size should be accessed using
PyLong_Export().
Also, the presence of an ob_size field in the
instance layout doesn’t mean that the instance structure is variable-length.
For example, the list type has fixed-length instances, yet those
instances have a ob_size field.
(As with int, avoid reading lists’ ob_size directly.
Call PyList_Size() instead.)
The tp_basicsize includes size needed for data of the type’s
tp_base, plus any extra data needed
by each instance.
The correct way to set tp_basicsize is to use the
sizeof operator on the struct used to declare the instance layout.
This struct must include the struct used to declare the base type.
In other words, tp_basicsize must be greater than or equal
to the base’s tp_basicsize.
tp_basicsize must be a multiple of _Alignof(PyObject).
When using sizeof on a struct that includes
PyObject_HEAD, as recommended, the compiler ensures this.
When not using a C struct, or when using compiler
extensions like __attribute__((packed)), it is up to you.
If the variable items require a particular alignment,
tp_basicsize and tp_itemsize must each be a
multiple of that alignment.
For example, if a type’s variable part stores a double, it is
your responsibility that both fields are a multiple of
_Alignof(double).
Inheritance:
These fields are inherited separately by subtypes.
(That is, if the field is set to zero, PyType_Ready() will copy
the value from the base type, indicating that the instances do not
need additional storage.)
If the base type has a non-zero tp_itemsize, it is generally not safe to set
tp_itemsize to a different non-zero value in a subtype (though this
depends on the implementation of the base type).
A pointer to the instance destructor function. The function signature is:
voidtp_dealloc(PyObject*self);
The destructor function should remove all references which the instance owns
(e.g., call Py_CLEAR()), free all memory buffers owned by the
instance, and call the type’s tp_free function to
free the object itself.
If you may call functions that may set the error indicator, you must use
PyErr_GetRaisedException() and PyErr_SetRaisedException()
to ensure you don’t clobber a preexisting error indicator (the deallocation
could have occurred while processing a different error):
The dealloc handler itself must not raise an exception; if it hits an error
case it should call PyErr_FormatUnraisable() to log (and clear) an
unraisable exception.
No guarantees are made about when an object is destroyed, except:
Python will destroy an object immediately or some time after the final
reference to the object is deleted, unless its finalizer
(tp_finalize) subsequently resurrects the
object.
An object will not be destroyed while it is being automatically finalized
(tp_finalize) or automatically cleared
(tp_clear).
CPython currently destroys an object immediately from Py_DECREF()
when the new reference count is zero, but this may change in a future
version.
It is recommended to call PyObject_CallFinalizerFromDealloc() at the
beginning of tp_dealloc to guarantee that the object is always
finalized before destruction.
If the type supports garbage collection (the Py_TPFLAGS_HAVE_GC
flag is set), the destructor should call PyObject_GC_UnTrack()
before clearing any member fields.
It is permissible to call tp_clear from
tp_dealloc to reduce code duplication and to guarantee that the
object is always cleared before destruction. Beware that
tp_clear might have already been called.
If the type is heap allocated (Py_TPFLAGS_HEAPTYPE), the
deallocator should release the owned reference to its type object (via
Py_DECREF()) after calling the type deallocator. See the example
code below.:
tp_dealloc must leave the exception status unchanged. If it
needs to call something that might raise an exception, the exception state
must be backed up first and restored later (after logging any exceptions
with PyErr_WriteUnraisable()).
Example:
staticvoidfoo_dealloc(PyObject*self){PyObject*exc=PyErr_GetRaisedException();if(PyObject_CallFinalizerFromDealloc(self)<0){// self was resurrected.gotodone;}PyTypeObject*tp=Py_TYPE(self);if(tp->tp_flags&Py_TPFLAGS_HAVE_GC){PyObject_GC_UnTrack(self);}// Optional, but convenient to avoid code duplication.if(tp->tp_clear&&tp->tp_clear(self)<0){PyErr_WriteUnraisable(self);}// Any additional destruction goes here.tp->tp_free(self);self=NULL;// In case PyErr_WriteUnraisable() is called below.if(tp->tp_flags&Py_TPFLAGS_HEAPTYPE){Py_CLEAR(tp);}done:// Optional, if something was called that might have raised an// exception.if(PyErr_Occurred()){PyErr_WriteUnraisable(self);}PyErr_SetRaisedException(exc);}
tp_dealloc may be called from
any Python thread, not just the thread which created the object (if the
object becomes part of a refcount cycle, that cycle might be collected by
a garbage collection on any thread). This is not a problem for Python
API calls, since the thread on which tp_dealloc is called
with an attached thread state. However, if the object being
destroyed in turn destroys objects from some other C library, care
should be taken to ensure that destroying those objects on the thread
which called tp_dealloc will not violate any assumptions of
the library.
Inheritance:
This field is inherited by subtypes.
See also
Object Life Cycle for details about how this slot relates to other slots.
An optional offset to a per-instance function that implements calling
the object using the vectorcall protocol,
a more efficient alternative
of the simpler tp_call.
This field is only used if the flag Py_TPFLAGS_HAVE_VECTORCALL
is set. If so, this must be a positive integer containing the offset in the
instance of a vectorcallfunc pointer.
The vectorcallfunc pointer may be NULL, in which case the instance behaves
as if Py_TPFLAGS_HAVE_VECTORCALL was not set: calling the instance
falls back to tp_call.
Any class that sets Py_TPFLAGS_HAVE_VECTORCALL must also set
tp_call and make sure its behaviour is consistent
with the vectorcallfunc function.
This can be done by setting tp_call to PyVectorcall_Call().
Changed in version 3.8: Before version 3.8, this slot was named tp_print.
In Python 2.x, it was used for printing to a file.
In Python 3.0 to 3.7, it was unused.
Changed in version 3.12: Before version 3.12, it was not recommended for
mutable heap types to implement the vectorcall
protocol.
When a user sets __call__ in Python code, only tp_call is
updated, likely making it inconsistent with the vectorcall function.
Since 3.12, setting __call__ will disable vectorcall optimization
by clearing the Py_TPFLAGS_HAVE_VECTORCALL flag.
An optional pointer to the get-attribute-string function.
This field is deprecated. When it is defined, it should point to a function
that acts the same as the tp_getattro function, but taking a C string
instead of a Python string object to give the attribute name.
This field is inherited by subtypes together with tp_getattro: a subtype
inherits both tp_getattr and tp_getattro from its base type when
the subtype’s tp_getattr and tp_getattro are both NULL.
An optional pointer to the function for setting and deleting attributes.
This field is deprecated. When it is defined, it should point to a function
that acts the same as the tp_setattro function, but taking a C string
instead of a Python string object to give the attribute name.
This field is inherited by subtypes together with tp_setattro: a subtype
inherits both tp_setattr and tp_setattro from its base type when
the subtype’s tp_setattr and tp_setattro are both NULL.
The function must return a string or a Unicode object. Ideally,
this function should return a string that, when passed to
eval(), given a suitable environment, returns an object with the
same value. If this is not feasible, it should return a string starting with
'<' and ending with '>' from which both the type and the value of the
object can be deduced.
Inheritance:
This field is inherited by subtypes.
Default:
When this field is not set, a string of the form <%sobjectat%p> is
returned, where %s is replaced by the type name, and %p by the object’s
memory address.
Pointer to an additional structure that contains fields relevant only to
objects which implement the number protocol. These fields are documented in
Number Object Structures.
Inheritance:
The tp_as_number field is not inherited, but the contained fields are
inherited individually.
Pointer to an additional structure that contains fields relevant only to
objects which implement the sequence protocol. These fields are documented
in Sequence Object Structures.
Inheritance:
The tp_as_sequence field is not inherited, but the contained fields
are inherited individually.
Pointer to an additional structure that contains fields relevant only to
objects which implement the mapping protocol. These fields are documented in
Mapping Object Structures.
Inheritance:
The tp_as_mapping field is not inherited, but the contained fields
are inherited individually.
The value -1 should not be returned as a
normal return value; when an error occurs during the computation of the hash
value, the function should set an exception and return -1.
This field can be set explicitly to PyObject_HashNotImplemented() to
block inheritance of the hash method from a parent type. This is interpreted
as the equivalent of __hash__=None at the Python level, causing
isinstance(o,collections.Hashable) to correctly return False. Note
that the converse is also true - setting __hash__=None on a class at
the Python level will result in the tp_hash slot being set to
PyObject_HashNotImplemented().
This field is inherited by subtypes together with
tp_richcompare: a subtype inherits both of
tp_richcompare and tp_hash, when the subtype’s
tp_richcompare and tp_hash are both NULL.
An optional pointer to a function that implements calling the object. This
should be NULL if the object is not callable. The signature is the same as
for PyObject_Call():
An optional pointer to a function that implements the built-in operation
str(). (Note that str is a type now, and str() calls the
constructor for that type. This constructor calls PyObject_Str() to do
the actual work, and PyObject_Str() will call this handler.)
The function must return a string or a Unicode object. It should be a “friendly” string
representation of the object, as this is the representation that will be used,
among other things, by the print() function.
Inheritance:
This field is inherited by subtypes.
Default:
When this field is not set, PyObject_Repr() is called to return a string
representation.
This field is inherited by subtypes together with tp_getattr: a subtype
inherits both tp_getattr and tp_getattro from its base type when
the subtype’s tp_getattr and tp_getattro are both NULL.
In addition, setting value to NULL to delete an attribute must be
supported. It is usually convenient to set this field to
PyObject_GenericSetAttr(), which implements the normal
way of setting object attributes.
This field is inherited by subtypes together with tp_setattr: a subtype
inherits both tp_setattr and tp_setattro from its base type when
the subtype’s tp_setattr and tp_setattro are both NULL.
Pointer to an additional structure that contains fields relevant only to objects
which implement the buffer interface. These fields are documented in
Buffer Object Structures.
Inheritance:
The tp_as_buffer field is not inherited,
but the contained fields are inherited individually.
This field is a bit mask of various flags. Some flags indicate variant
semantics for certain situations; others are used to indicate that certain
fields in the type object (or in the extension structures referenced via
tp_as_number, tp_as_sequence, tp_as_mapping, and
tp_as_buffer) that were historically not always present are valid; if
such a flag bit is clear, the type fields it guards must not be accessed and
must be considered to have a zero or NULL value instead.
Inheritance:
Inheritance of this field is complicated. Most flag bits are inherited
individually, i.e. if the base type has a flag bit set, the subtype inherits
this flag bit. The flag bits that pertain to extension structures are strictly
inherited if the extension structure is inherited, i.e. the base type’s value of
the flag bit is copied into the subtype together with a pointer to the extension
structure. The Py_TPFLAGS_HAVE_GC flag bit is inherited together with
the tp_traverse and tp_clear fields, i.e. if the
Py_TPFLAGS_HAVE_GC flag bit is clear in the subtype and the
tp_traverse and tp_clear fields in the subtype exist and have
NULL values.
The following bit masks are currently defined; these can be ORed together using
the | operator to form the value of the tp_flags field. The macro
PyType_HasFeature() takes a type and a flags value, tp and f, and
checks whether tp->tp_flags&f is non-zero.
This bit is set when the type object itself is allocated on the heap, for
example, types created dynamically using PyType_FromSpec(). In this
case, the ob_type field of its instances is considered a reference to
the type, and the type object is INCREF’ed when a new instance is created, and
DECREF’ed when an instance is destroyed (this does not apply to instances of
subtypes; only the type referenced by the instance’s ob_type gets INCREF’ed or
DECREF’ed). Heap types should also support garbage collection
as they can form a reference cycle with their own module object.
This bit is set when the type can be used as the base type of another type. If
this bit is clear, the type cannot be subtyped (similar to a “final” class in
Java).
The Py_TPFLAGS_HAVE_GC flag bit is inherited
together with the tp_traverse and tp_clear
fields, i.e. if the Py_TPFLAGS_HAVE_GC flag bit is
clear in the subtype and the tp_traverse and
tp_clear fields in the subtype exist and have NULL
values.
This is a bitmask of all the bits that pertain to the existence of certain
fields in the type object and its extension structures. Currently, it includes
the following bits: Py_TPFLAGS_HAVE_STACKLESS_EXTENSION.
This bit indicates that objects behave like unbound methods.
If this flag is set for type(meth), then:
meth.__get__(obj,cls)(*args,**kwds) (with obj not None)
must be equivalent to meth(obj,*args,**kwds).
meth.__get__(None,cls)(*args,**kwds)
must be equivalent to meth(*args,**kwds).
This flag enables an optimization for typical method calls like
obj.meth(): it avoids creating a temporary “bound method” object for
obj.meth.
Added in version 3.8.
Inheritance:
This flag is never inherited by types without the
Py_TPFLAGS_IMMUTABLETYPE flag set. For extension types, it is
inherited whenever tp_descr_get is inherited.
This macro value relies on the implementation of the VM, so its value is not
stable and may change in a future version. Prefer using individual
flags instead.
Only usable with variable-size types, i.e. ones with non-zero
tp_itemsize.
Indicates that the variable-sized portion of an instance of this type is
at the end of the instance’s memory area, at an offset of
Py_TYPE(obj)->tp_basicsize (which may be different in each
subclass).
When setting this flag, be sure that all superclasses either
use this memory layout, or are not variable-sized.
Python does not check this.
Functions such as PyLong_Check() will call PyType_FastSubclass()
with one of these flags to quickly determine if a type is a subclass
of a built-in type; such specific checks are faster than a generic
check, like PyObject_IsInstance(). Custom types that inherit
from built-ins should have their tp_flags
set appropriately, or the code that interacts with such types
will behave differently depending on what kind of check is used.
This bit is set when the tp_finalize slot is present in the
type structure.
Added in version 3.4.
Deprecated since version 3.8: This flag isn’t necessary anymore, as the interpreter assumes the
tp_finalize slot is always present in the
type structure.
Disallow creating instances of the type: set
tp_new to NULL and don’t create the __new__
key in the type dictionary.
The flag must be set before creating the type, not after. For example, it
must be set before PyType_Ready() is called on the type.
The flag is set automatically on static types if
tp_base is NULL or &PyBaseObject_Type and
tp_new is NULL.
Inheritance:
This flag is not inherited.
However, subclasses will not be instantiable unless they provide a
non-NULL tp_new (which is only possible
via the C API).
Note
To disallow instantiating a class directly but allow instantiating
its subclasses (e.g. for an abstract base class),
do not use this flag.
Instead, make tp_new only succeed for
subclasses.
This bit indicates that instances of the class may match mapping patterns
when used as the subject of a match block. It is automatically
set when registering or subclassing collections.abc.Mapping, and
unset when registering collections.abc.Sequence.
This bit indicates that instances of the class may match sequence patterns
when used as the subject of a match block. It is automatically
set when registering or subclassing collections.abc.Sequence, and
unset when registering collections.abc.Mapping.
This bit indicates that instances of this type will have an “inline values”
array (containing the object’s attributes) placed directly after the end
of the object.
An optional pointer to a NUL-terminated C string giving the docstring for this
type object. This is exposed as the __doc__ attribute on the
type and instances of the type.
This field is inherited by subtypes together with tp_clear and the
Py_TPFLAGS_HAVE_GC flag bit: the flag bit, tp_traverse, and
tp_clear are all inherited from the base type if they are all zero in
the subtype.
An optional pointer to a clear function. The signature is:
inttp_clear(PyObject*);
The purpose of this function is to break reference cycles that are causing a
cyclic isolate so that the objects can be safely destroyed. A
cleared object is a partially destroyed object; the object is not obligated
to satisfy design invariants held during normal use.
tp_clear does not need to delete references to objects that
can’t participate in reference cycles, such as Python strings or Python
integers. However, it may be convenient to clear all references, and write
the type’s tp_dealloc function to invoke
tp_clear to avoid code duplication. (Beware that
tp_clear might have already been called. Prefer calling
idempotent functions like Py_CLEAR().)
Any non-trivial cleanup should be performed in
tp_finalize instead of tp_clear.
Note
If tp_clear fails to break a reference cycle then the
objects in the cyclic isolate may remain indefinitely
uncollectable (“leak”). See gc.garbage.
Note
Referents (direct and indirect) might have already been cleared; they are
not guaranteed to be in a consistent state.
Note
The tp_clear function can be called from any
thread.
Note
An object is not guaranteed to be automatically cleared before its
destructor (tp_dealloc) is called.
This function differs from the destructor
(tp_dealloc) in the following ways:
The purpose of clearing an object is to remove references to other objects
that might participate in a reference cycle. The purpose of the
destructor, on the other hand, is a superset: it must release all
resources it owns, including references to objects that cannot participate
in a reference cycle (e.g., integers) as well as the object’s own memory
(by calling tp_free).
When tp_clear is called, other objects might still hold
references to the object being cleared. Because of this,
tp_clear must not deallocate the object’s own memory
(tp_free). The destructor, on the other hand,
is only called when no (strong) references exist, and as such, must
safely destroy the object itself by deallocating it.
tp_clear might never be automatically called. An object’s
destructor, on the other hand, will be automatically called some time
after the object becomes unreachable (i.e., either there are no references
to the object or the object is a member of a cyclic isolate).
No guarantees are made about when, if, or how often Python automatically
clears an object, except:
Python will not automatically clear an object if it is reachable, i.e.,
there is a reference to it and it is not a member of a cyclic
isolate.
Python will not automatically clear an object if it has not been
automatically finalized (see tp_finalize). (If
the finalizer resurrected the object, the object may or may not be
automatically finalized again before it is cleared.)
If an object is a member of a cyclic isolate, Python will not
automatically clear it if any member of the cyclic isolate has not yet
been automatically finalized (tp_finalize).
Python will not destroy an object until after any automatic calls to its
tp_clear function have returned. This ensures that the act
of breaking a reference cycle does not invalidate the self pointer
while tp_clear is still executing.
Python will not automatically call tp_clear multiple times
concurrently.
CPython currently only automatically clears objects as needed to break
reference cycles in a cyclic isolate, but future versions might
clear objects regularly before their destruction.
Taken together, all tp_clear functions in the
system must combine to break all reference cycles. This is subtle, and if
in any doubt supply a tp_clear function. For
example, the tuple type does not implement a
tp_clear function, because it’s possible to prove
that no reference cycle can be composed entirely of tuples. Therefore the
tp_clear functions of other types are responsible
for breaking any cycle containing a tuple. This isn’t immediately obvious,
and there’s rarely a good reason to avoid implementing
tp_clear.
Implementations of tp_clear should drop the instance’s references to
those of its members that may be Python objects, and set its pointers to those
members to NULL, as in the following example:
The Py_CLEAR() macro should be used, because clearing references is
delicate: the reference to the contained object must not be released
(via Py_DECREF()) until
after the pointer to the contained object is set to NULL. This is because
releasing the reference may cause the contained object to become trash,
triggering a chain of reclamation activity that may include invoking arbitrary
Python code (due to finalizers, or weakref callbacks, associated with the
contained object). If it’s possible for such code to reference self again,
it’s important that the pointer to the contained object be NULL at that time,
so that self knows the contained object can no longer be used. The
Py_CLEAR() macro performs the operations in a safe order.
This field is inherited by subtypes together with tp_traverse and the
Py_TPFLAGS_HAVE_GC flag bit: the flag bit, tp_traverse, and
tp_clear are all inherited from the base type if they are all zero in
the subtype.
See also
Object Life Cycle for details about how this slot relates to other slots.
The first parameter is guaranteed to be an instance of the type
that is defined by PyTypeObject.
The function should return the result of the comparison (usually Py_True
or Py_False). If the comparison is undefined, it must return
Py_NotImplemented, if another error occurred it must return NULL and
set an exception condition.
Return Py_True or Py_False from the function, depending on the
result of a comparison.
VAL_A and VAL_B must be orderable by C comparison operators (for example,
they may be C ints or floats). The third argument specifies the requested
operation, as for PyObject_RichCompare().
This field is inherited by subtypes together with tp_hash:
a subtype inherits tp_richcompare and tp_hash when
the subtype’s tp_richcompare and tp_hash are both
NULL.
Default:
PyBaseObject_Type provides a tp_richcompare
implementation, which may be inherited. However, if only
tp_hash is defined, not even the inherited function is used
and instances of the type will not be able to participate in any
comparisons.
While this field is still supported, Py_TPFLAGS_MANAGED_WEAKREF
should be used instead, if at all possible.
If the instances of this type are weakly referenceable, this field is greater
than zero and contains the offset in the instance structure of the weak
reference list head (ignoring the GC header, if present); this offset is used by
PyObject_ClearWeakRefs() and the PyWeakref_* functions. The
instance structure needs to include a field of type PyObject* which is
initialized to NULL.
Do not confuse this field with tp_weaklist; that is the list head for
weak references to the type object itself.
This field is inherited by subtypes, but see the rules listed below. A subtype
may override this offset; this means that the subtype uses a different weak
reference list head than the base type. Since the list head is always found via
tp_weaklistoffset, this should not be a problem.
An optional pointer to a function that returns an iterator for the
object. Its presence normally signals that the instances of this type are
iterable (although sequences may be iterable without this function).
An optional pointer to a function that returns the next item in an
iterator. The signature is:
PyObject*tp_iternext(PyObject*self);
When the iterator is exhausted, the tp_iternext function must
return NULL; a StopIteration exception may or may not be set.
When another error occurs, it must return NULL too.
The presence of tp_iternext signals that the instances of this type are
iterators.
Iterator types should also define the tp_iter function, and that
function should return the iterator instance itself (not a new iterator
instance).
This function has the same signature as PyIter_Next().
An optional pointer to a static NULL-terminated array of PyMemberDef
structures, declaring regular data members (fields or slots) of instances of
this type.
For each entry in the array, an entry is added to the type’s dictionary (see
tp_dict below) containing a member descriptor.
Inheritance:
This field is not inherited by subtypes (members are inherited through a
different mechanism).
An optional pointer to a base type from which type properties are inherited. At
this level, only single inheritance is supported; multiple inheritance require
dynamically creating a type object by calling the metatype.
Slot initialization is subject to the rules of initializing globals.
C99 requires the initializers to be “address constants”. Function
designators like PyType_GenericNew(), with implicit conversion
to a pointer, are valid C99 address constants.
However, the unary ‘&’ operator applied to a non-static variable
like PyBaseObject_Type is not required to produce an address
constant. Compilers may support this (gcc does), MSVC does not.
Both compilers are strictly standard conforming in this particular
behavior.
Consequently, tp_base should be set in
the extension module’s init function.
Inheritance:
This field is not inherited by subtypes (obviously).
Default:
This field defaults to &PyBaseObject_Type (which to Python
programmers is known as the type object).
This field should normally be initialized to NULL before PyType_Ready is
called; it may also be initialized to a dictionary containing initial attributes
for the type. Once PyType_Ready() has initialized the type, extra
attributes for the type may be added to this dictionary only if they don’t
correspond to overloaded operations (like __add__()). Once
initialization for the type has finished, this field should be
treated as read-only.
Some types may not store their dictionary in this slot.
Use PyType_GetDict() to retrieve the dictionary for an arbitrary
type.
Changed in version 3.12: Internals detail: For static builtin types, this is always NULL.
Instead, the dict for such types is stored on PyInterpreterState.
Use PyType_GetDict() to get the dict for an arbitrary type.
Inheritance:
This field is not inherited by subtypes (though the attributes defined in here
are inherited through a different mechanism).
Default:
If this field is NULL, PyType_Ready() will assign a new
dictionary to it.
Warning
It is not safe to use PyDict_SetItem() on or otherwise modify
tp_dict with the dictionary C-API.
While this field is still supported, Py_TPFLAGS_MANAGED_DICT should be
used instead, if at all possible.
If the instances of this type have a dictionary containing instance variables,
this field is non-zero and contains the offset in the instances of the type of
the instance variable dictionary; this offset is used by
PyObject_GenericGetAttr().
Do not confuse this field with tp_dict; that is the dictionary for
attributes of the type object itself.
The value specifies the offset of the dictionary from the start of the instance structure.
The tp_dictoffset should be regarded as write-only.
To get the pointer to the dictionary call PyObject_GenericGetDict().
Calling PyObject_GenericGetDict() may need to allocate memory for the
dictionary, so it is may be more efficient to call PyObject_GetAttr()
when accessing an attribute on the object.
This field is inherited by subtypes. A subtype should not override this offset;
doing so could be unsafe, if C code tries to access the dictionary at the
previous offset.
To properly support inheritance, use Py_TPFLAGS_MANAGED_DICT.
Default:
This slot has no default. For static types, if the
field is NULL then no __dict__ gets created for instances.
An optional pointer to an instance initialization function.
This function corresponds to the __init__() method of classes. Like
__init__(), it is possible to create an instance without calling
__init__(), and it is possible to reinitialize an instance by calling its
__init__() method again.
The self argument is the instance to be initialized; the args and kwds
arguments represent positional and keyword arguments of the call to
__init__().
The tp_init function, if not NULL, is called when an instance is
created normally by calling its type, after the type’s tp_new function
has returned an instance of the type. If the tp_new function returns an
instance of some other type that is not a subtype of the original type, no
tp_init function is called; if tp_new returns an instance of a
subtype of the original type, the subtype’s tp_init is called.
Returns 0 on success, -1 and sets an exception on error.
Inheritance:
This field is inherited by subtypes.
Default:
For static types this field does not have a default.
The subtype argument is the type of the object being created; the args and
kwds arguments represent positional and keyword arguments of the call to the
type. Note that subtype doesn’t have to equal the type whose tp_new
function is called; it may be a subtype of that type (but not an unrelated
type).
The tp_new function should call subtype->tp_alloc(subtype,nitems)
to allocate space for the object, and then do only as much further
initialization as is absolutely necessary. Initialization that can safely be
ignored or repeated should be placed in the tp_init handler. A good
rule of thumb is that for immutable types, all initialization should take place
in tp_new, while for mutable types, most initialization should be
deferred to tp_init.
This field is inherited by subtypes, except it is not inherited by
static types whose tp_base
is NULL or &PyBaseObject_Type.
Default:
For static types this field has no default.
This means if the slot is defined as NULL, the type cannot be called
to create new instances; presumably there is some other way to create
instances, like a factory function.
An optional pointer to an instance deallocation function. Its signature is:
voidtp_free(void*self);
This function must free the memory allocated by
tp_alloc.
Inheritance:
Static subtypes inherit this slot, which will be PyObject_Free() if
inherited from object. Exception: If the type supports garbage
collection (i.e., the Py_TPFLAGS_HAVE_GC flag is set in
tp_flags) and it would inherit
PyObject_Free(), then this slot is not inherited but instead defaults
to PyObject_GC_Del().
An optional pointer to a function called by the garbage collector.
The garbage collector needs to know whether a particular object is collectible
or not. Normally, it is sufficient to look at the object’s type’s
tp_flags field, and check the Py_TPFLAGS_HAVE_GC flag bit. But
some types have a mixture of statically and dynamically allocated instances, and
the statically allocated instances are not collectible. Such types should
define this function; it should return 1 for a collectible instance, and
0 for a non-collectible instance. The signature is:
inttp_is_gc(PyObject*self);
(The only example of this are types themselves. The metatype,
PyType_Type, defines this function to distinguish between statically
and dynamically allocated types.)
Inheritance:
This field is inherited by subtypes.
Default:
This slot has no default. If this field is NULL,
Py_TPFLAGS_HAVE_GC is used as the functional equivalent.
Multiple inheritance does not work well for statically defined types.
If you set tp_bases to a tuple, Python will not raise an error,
but some slots will only be inherited from the first base.
Weak reference list head, for weak references to this type object. Not
inherited. Internal use only.
Changed in version 3.12: Internals detail: For the static builtin types this is always NULL,
even if weakrefs are added. Instead, the weakrefs for each are stored
on PyInterpreterState. Use the public C-API or the internal
_PyObject_GET_WEAKREFS_LISTPTR() macro to avoid the distinction.
The corresponding slot IDPy_tp_finalize is part of the Stable ABI since version 3.5.
An optional pointer to an instance finalization function. This is the C
implementation of the __del__() special method. Its signature
is:
voidtp_finalize(PyObject*self);
The primary purpose of finalization is to perform any non-trivial cleanup
that must be performed before the object is destroyed, while the object and
any other objects it directly or indirectly references are still in a
consistent state. The finalizer is allowed to execute
arbitrary Python code.
Before Python automatically finalizes an object, some of the object’s direct
or indirect referents might have themselves been automatically finalized.
However, none of the referents will have been automatically cleared
(tp_clear) yet.
Other non-finalized objects might still be using a finalized object, so the
finalizer must leave the object in a sane state (e.g., invariants are still
met).
Note
After Python automatically finalizes an object, Python might start
automatically clearing (tp_clear) the object
and its referents (direct and indirect). Cleared objects are not
guaranteed to be in a consistent state; a finalized object must be able
to tolerate cleared referents.
Note
An object is not guaranteed to be automatically finalized before its
destructor (tp_dealloc) is called. It is
recommended to call PyObject_CallFinalizerFromDealloc() at the
beginning of tp_dealloc to guarantee that the object is
always finalized before destruction.
Note
The tp_finalize function can be called from any
thread, although the GIL will be held.
Note
The tp_finalize function can be called during shutdown,
after some global variables have been deleted. See the documentation of
the __del__() method for details.
When Python finalizes an object, it behaves like the following algorithm:
Python might mark the object as finalized. Currently, Python always
marks objects whose type supports garbage collection (i.e., the
Py_TPFLAGS_HAVE_GC flag is set in
tp_flags) and never marks other types of
objects; this might change in a future version.
If the object is not marked as finalized and its
tp_finalize finalizer function is non-NULL, the
finalizer function is called.
If the finalizer function was called and the finalizer made the object
reachable (i.e., there is a reference to the object and it is not a
member of a cyclic isolate), then the finalizer is said to have
resurrected the object. It is unspecified whether the finalizer can
also resurrect the object by adding a new reference to the object that
does not make it reachable, i.e., the object is (still) a member of a
cyclic isolate.
If the finalizer resurrected the object, the object’s pending destruction
is canceled and the object’s finalized mark might be removed if
present. Currently, Python never removes the finalized mark; this
might change in a future version.
Automatic finalization refers to any finalization performed by Python
except via calls to PyObject_CallFinalizer() or
PyObject_CallFinalizerFromDealloc(). No guarantees are made about
when, if, or how often an object is automatically finalized, except:
Python will not automatically finalize an object if it is reachable, i.e.,
there is a reference to it and it is not a member of a cyclic
isolate.
Python will not automatically finalize an object if finalizing it would
not mark the object as finalized. Currently, this applies to objects
whose type does not support garbage collection, i.e., the
Py_TPFLAGS_HAVE_GC flag is not set. Such objects can still be
manually finalized by calling PyObject_CallFinalizer() or
PyObject_CallFinalizerFromDealloc().
Python will not automatically finalize any two members of a cyclic
isolate concurrently.
Python will not automatically finalize an object after it has
automatically cleared (tp_clear) the object.
If an object is a member of a cyclic isolate, Python will not
automatically finalize it after automatically clearing (see
tp_clear) any other member.
Python will automatically finalize every member of a cyclic
isolate before it automatically clears (see
tp_clear) any of them.
If Python is going to automatically clear an object
(tp_clear), it will automatically finalize the
object first.
Python currently only automatically finalizes objects that are members of a
cyclic isolate, but future versions might finalize objects regularly
before their destruction.
staticvoidfoo_finalize(PyObject*self){// Save the current exception, if any.PyObject*exc=PyErr_GetRaisedException();// ...if(do_something_that_might_raise()!=success_indicator){PyErr_WriteUnraisable(self);gotodone;}done:// Restore the saved exception. This silently discards any exception// raised above, so be sure to call PyErr_WriteUnraisable first if// necessary.PyErr_SetRaisedException(exc);}
Inheritance:
This field is inherited by subtypes.
Added in version 3.4.
Changed in version 3.8: Before version 3.8 it was necessary to set the
Py_TPFLAGS_HAVE_FINALIZE flags bit in order for this field to be
used. This is no longer required.
The corresponding slot IDPy_tp_vectorcall is part of the Stable ABI since version 3.14.
A vectorcall function to use for calls of this type
object (rather than instances).
In other words, tp_vectorcall can be used to optimize type.__call__,
which typically returns a new instance of type.
As with any vectorcall function, if tp_vectorcall is NULL,
the tp_call protocol (Py_TYPE(type)->tp_call) is used instead.
Note
The vectorcall protocol requires that the vectorcall
function has the same behavior as the corresponding tp_call.
This means that type->tp_vectorcall must match the behavior of
Py_TYPE(type)->tp_call.
Specifically, if type uses the default metaclass,
type->tp_vectorcall must behave the same as
Py_TYPE(type)->tp_call, which:
calls type->tp_new,
if the result is a subclass of type, calls type->tp_init
on the result of tp_new, and
returns the result of tp_new.
Typically, tp_vectorcall is overridden to optimize this process
for specific tp_new and
tp_init.
When doing this for user-subclassable types, note that both can be
overridden (using __new__() and
__init__(), respectively).
Inheritance:
This field is never inherited.
Added in version 3.9: (the field exists since 3.8 but it’s only used since 3.9)
Traditionally, types defined in C code are static, that is,
a static PyTypeObject structure is defined directly in code
and initialized using PyType_Ready().
This results in types that are limited relative to types defined in Python:
Static types are limited to one base, i.e. they cannot use multiple
inheritance.
Static type objects (but not necessarily their instances) are immutable.
It is not possible to add or modify the type object’s attributes from Python.
Static type objects are shared across
sub-interpreters, so they should not
include any subinterpreter-specific state.
Also, since PyTypeObject is only part of the Limited API as an opaque struct, any extension modules using static types must be
compiled for a specific Python minor version.
An alternative to static types is heap-allocated types,
or heap types for short, which correspond closely to classes created by
Python’s class statement. Heap types have the Py_TPFLAGS_HEAPTYPE
flag set.
This structure holds pointers to the functions which an object uses to
implement the number protocol. Each function is used by the function of
similar name documented in the Number Protocol section.
Binary and ternary functions must check the type of all their operands,
and implement the necessary conversions (at least one of the operands is
an instance of the defined type). If the operation is not defined for the
given operands, binary and ternary functions must return
Py_NotImplemented, if another error occurred they must return NULL
and set an exception.
Note
The nb_reserved field should always be NULL. It
was previously called nb_long, and was renamed in
Python 3.0.1.
This function is used by PyMapping_Size() and
PyObject_Size(), and has the same signature. This slot may be set to
NULL if the object has no defined length.
This function is used by PySequence_Concat() and has the same
signature. It is also used by the + operator, after trying the numeric
addition via the nb_add slot.
This function is used by PySequence_Repeat() and has the same
signature. It is also used by the * operator, after trying numeric
multiplication via the nb_multiply slot.
Negative indexes are handled as follows: if the sq_length slot is
filled, it is called and the sequence length is used to compute a positive
index which is passed to sq_item. If sq_length is NULL,
the index is passed as is to the function.
This function is used by PySequence_SetItem() and has the same
signature. It is also used by PyObject_SetItem() and
PyObject_DelItem(), after trying the item assignment and deletion
via the mp_ass_subscript slot.
This slot may be left to NULL if the object does not support
item assignment and deletion.
This function may be used by PySequence_Contains() and has the same
signature. This slot may be left to NULL, in this case
PySequence_Contains() simply traverses the sequence until it
finds a match.
The corresponding slot IDPy_sq_inplace_concat is part of the Stable ABI.
This function is used by PySequence_InPlaceConcat() and has the same
signature. It should modify its first operand, and return it. This slot
may be left to NULL, in this case PySequence_InPlaceConcat()
will fall back to PySequence_Concat(). It is also used by the
augmented assignment +=, after trying numeric in-place addition
via the nb_inplace_add slot.
The corresponding slot IDPy_sq_inplace_repeat is part of the Stable ABI.
This function is used by PySequence_InPlaceRepeat() and has the same
signature. It should modify its first operand, and return it. This slot
may be left to NULL, in this case PySequence_InPlaceRepeat()
will fall back to PySequence_Repeat(). It is also used by the
augmented assignment *=, after trying numeric in-place multiplication
via the nb_inplace_multiply slot.
This structure holds pointers to the functions required by the
Buffer protocol. The protocol defines how
an exporter object can expose its internal data to consumer objects.
The export counter increment in step (3) is atomic.
The underlying buffer data remains valid and at a stable memory
location for the lifetime of all exports.
For objects that support resizing or reallocation (such as
bytearray), the export counter is checked atomically before
such operations, and BufferError is raised if exports exist.
The function is safe to call concurrently from multiple threads.
If exporter is part of a chain or tree of buffer providers, two main
schemes can be used:
Re-export: Each member of the tree acts as the exporting object and
sets view->obj to a new reference to itself.
Redirect: The buffer request is redirected to the root object of the
tree. Here, view->obj will be a new reference to the root
object.
The individual fields of view are described in section
Buffer structure, the rules how an exporter
must react to specific requests are in section
Buffer request types.
All memory pointed to in the Py_buffer structure belongs to
the exporter and must remain valid until there are no consumers left.
format, shape,
strides, suboffsets
and internal
are read-only for the consumer.
PyBuffer_FillInfo() provides an easy way of exposing a simple
bytes buffer while dealing correctly with all request types.
PyObject_GetBuffer() is the interface for the consumer that
wraps this function.
The corresponding slot IDPy_bf_releasebuffer is part of the Stable ABI since version 3.11.
The signature of this function is:
void(PyObject*exporter,Py_buffer*view);
Handle a request to release the resources of the buffer. If no resources
need to be released, PyBufferProcs.bf_releasebuffer may be
NULL. Otherwise, a standard implementation of this function will take
these optional steps:
Decrement an internal counter for the number of exports.
If the counter is 0, free all memory associated with view.
The export counter decrement in step (1) must be atomic.
Resource cleanup when the counter reaches zero must be done atomically,
as the final release may race with concurrent releases from other
threads and deallocation must only happen once.
The exporter MUST use the internal field to keep
track of buffer-specific resources. This field is guaranteed to remain
constant, while a consumer MAY pass a copy of the original buffer as the
view argument.
This function MUST NOT decrement view->obj, since that is
done automatically in PyBuffer_Release() (this scheme is
useful for breaking reference cycles).
PyBuffer_Release() is the interface for the consumer that
wraps this function.
The purpose of this function is to separate memory allocation from memory
initialization. It should return a pointer to a block of memory of adequate
length for the instance, suitably aligned, and initialized to zeros, but with
ob_refcnt set to 1 and ob_type set to the type argument. If
the type’s tp_itemsize is non-zero, the object’s ob_size field
should be initialized to nitems and the length of the allocated memory block
should be tp_basicsize+nitems*tp_itemsize, rounded up to a multiple of
sizeof(void*); otherwise, nitems is not used and the length of the block
should be tp_basicsize.
This function should not do any other instance initialization, not even to
allocate additional memory; that should be done by tp_new.
A str subclass that cannot be subclassed and cannot be called
to create instances (e.g. uses a separate factory func) using
Py_TPFLAGS_DISALLOW_INSTANTIATION flag:
typedefstruct{PyUnicodeObjectraw;char*extra;}MyStr;staticPyTypeObjectMyStr_Type={PyVarObject_HEAD_INIT(NULL,0).tp_name="mymod.MyStr",.tp_basicsize=sizeof(MyStr),.tp_base=NULL,// set to &PyUnicode_Type in module init.tp_doc=PyDoc_STR("my custom str"),.tp_flags=Py_TPFLAGS_DEFAULT|Py_TPFLAGS_DISALLOW_INSTANTIATION,.tp_repr=(reprfunc)myobj_repr,};
The simplest static type with fixed-length instances:
The Python Software Foundation is a non-profit corporation.
Please donate.
Last updated on Nov 12, 2026 (18:27 UTC).
Found a bug?
Created using Sphinx 9.1.1.
You are visitor 01402503 — GeezerVison does not store the pages you look up. Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.