Supporting Cyclic Garbage Collection

Python's support for detecting and collecting garbage which involves circular references requires support from object types which are "containers" for other objects which may also be containers. Types which do not store references to other objects, or which only store references to atomic types (such as numbers or strings), do not need to provide any explicit support for garbage collection.

To create a container type, the ~PyTypeObject.tp_flags field of the type object must include the Py_TPFLAGS_HAVE_GC and provide an implementation of the ~PyTypeObject.tp_traverse handler. If instances of the type are mutable, a ~PyTypeObject.tp_clear implementation must also be provided.

Py_TPFLAGS_HAVE_GC
Objects with a type with this flag set must conform with the rules documented here. For convenience these objects will be referred to as container objects.

Constructors for container types must conform to two rules:

1. The memory for the object must be allocated using PyObject_GC_New or PyObject_GC_NewVar.
2. Once all the fields which may contain references to other containers are initialized, it must call PyObject_GC_Track.

Similarly, the deallocator for the object must conform to a similar pair of rules:

1. Before fields which refer to other containers are invalidated, PyObject_GC_UnTrack must be called.

2. The object's memory must be deallocated using PyObject_GC_Del.

   [!WARNING] If a type adds the Py_TPFLAGS_HAVE_GC, then it must implement at least a ~PyTypeObject.tp_traverse handler or explicitly use one from its subclass or subclasses.

   When calling PyType_Ready or some of the APIs that indirectly call it like PyType_FromSpecWithBases or PyType_FromSpec the interpreter will automatically populate the ~PyTypeObject.tp_flags, ~PyTypeObject.tp_traverse and ~PyTypeObject.tp_clear fields if the type inherits from a class that implements the garbage collector protocol and the child class does not include the Py_TPFLAGS_HAVE_GC flag.

Analogous to PyObject_New but for container objects with the Py_TPFLAGS_HAVE_GC flag set.

Do not call this directly to allocate memory for an object; call the type's ~PyTypeObject.tp_alloc slot instead.

When populating a type's ~PyTypeObject.tp_alloc slot, PyType_GenericAlloc is preferred over a custom function that simply calls this macro.

Memory allocated by this macro must be freed with PyObject_GC_Del (usually called via the object's ~PyTypeObject.tp_free slot).

• PyObject_GC_Del
• PyObject_New
• PyType_GenericAlloc
• ~PyTypeObject.tp_alloc

Analogous to PyObject_NewVar but for container objects with the Py_TPFLAGS_HAVE_GC flag set.

Do not call this directly to allocate memory for an object; call the type's ~PyTypeObject.tp_alloc slot instead.

When populating a type's ~PyTypeObject.tp_alloc slot, PyType_GenericAlloc is preferred over a custom function that simply calls this macro.

Memory allocated by this macro must be freed with PyObject_GC_Del (usually called via the object's ~PyTypeObject.tp_free slot).

• PyObject_GC_Del
• PyObject_NewVar
• PyType_GenericAlloc
• ~PyTypeObject.tp_alloc

Analogous to PyObject_GC_New but allocates extra_size bytes at the end of the object (at offset ~PyTypeObject.tp_basicsize). The allocated memory is initialized to zeros, except for the Python object header <PyObject>.

The extra data will be deallocated with the object, but otherwise it is not managed by Python.

Memory allocated by this function must be freed with PyObject_GC_Del (usually called via the object's ~PyTypeObject.tp_free slot).

[!WARNING] The function is marked as unstable because the final mechanism for reserving extra data after an instance is not yet decided. For allocating a variable number of fields, prefer using PyVarObject and ~PyTypeObject.tp_itemsize instead.

3.12

Resize an object allocated by PyObject_GC_NewVar. Returns the resized object of type TYPE* (refers to any C type) or NULL on failure.

op must be of type PyVarObject * and must not be tracked by the collector yet. newsize must be of type Py_ssize_t.

Adds the object op to the set of container objects tracked by the collector. The collector can run at unexpected times so objects must be valid while being tracked. This should be called once all the fields followed by the ~PyTypeObject.tp_traverse handler become valid, usually near the end of the constructor.

Returns non-zero if the object implements the garbage collector protocol, otherwise returns 0.

The object cannot be tracked by the garbage collector if this function returns 0.

Returns 1 if the object type of op implements the GC protocol and op is being currently tracked by the garbage collector and 0 otherwise.

This is analogous to the Python function gc.is_tracked.

3.9

Returns 1 if the object type of op implements the GC protocol and op has been already finalized by the garbage collector and 0 otherwise.

This is analogous to the Python function gc.is_finalized.

3.9

Releases memory allocated to an object using PyObject_GC_New or PyObject_GC_NewVar.

Do not call this directly to free an object's memory; call the type's ~PyTypeObject.tp_free slot instead.

Do not use this for memory allocated by PyObject_New, PyObject_NewVar, or related allocation functions; use PyObject_Free instead.

• PyObject_Free is the non-GC equivalent of this function.
• PyObject_GC_New
• PyObject_GC_NewVar
• PyType_GenericAlloc
• ~PyTypeObject.tp_free

Remove the object op from the set of container objects tracked by the collector. Note that PyObject_GC_Track can be called again on this object to add it back to the set of tracked objects. The deallocator (~PyTypeObject.tp_dealloc handler) should call this for the object before any of the fields used by the ~PyTypeObject.tp_traverse handler become invalid.

The ~PyTypeObject.tp_traverse handler accepts a function parameter of this type:

Type of the visitor function passed to the ~PyTypeObject.tp_traverse handler. The function should be called with an object to traverse as object and the third parameter to the ~PyTypeObject.tp_traverse handler as arg. The Python core uses several visitor functions to implement cyclic garbage detection; it's not expected that users will need to write their own visitor functions.

The ~PyTypeObject.tp_clear handler must be of the inquiry type, or NULL if the object is immutable.

Drop references that may have created reference cycles. Immutable objects do not have to define this method since they can never directly create reference cycles. Note that the object must still be valid after calling this method (don't just call Py_DECREF on a reference). The collector will call this method if it detects that this object is involved in a reference cycle.

Traversal

The ~PyTypeObject.tp_traverse handler must have the following type:

Traversal function for a garbage-collected object, used by the garbage collector to detect reference cycles. Implementations must call the visit function for each object directly contained by self, with the parameters to visit being the contained object and the arg value passed to the handler. The visit function must not be called with a NULL object argument. If visit returns a non-zero value, that value should be returned immediately. Otherwise, return zero.

A typical !tp_traverse function calls the Py_VISIT convenience macro on each of the instance's members that are Python objects that the instance owns. For example, this is a (slightly outdated) traversal function for the threading.local class:

static int
local_traverse(PyObject *op, visitproc visit, void *arg)
{
    localobject *self = (localobject *) op;
    Py_VISIT(Py_TYPE(self));
    Py_VISIT(self->args);
    Py_VISIT(self->kw);
    Py_VISIT(self->dict);
    return 0;
}

[!NOTE] Py_VISIT requires the visit and arg parameters to !local_traverse to have these specific names; don't name them just anything.

Instances of heap-allocated types <heap-types> hold a reference to their type. Their traversal function must therefore visit the type:

Py_VISIT(Py_TYPE(self));

Alternately, the type may delegate this responsibility by calling tp_traverse of a heap-allocated superclass (or another heap-allocated type, if applicable). If they do not, the type object may not be garbage-collected.

If the Py_TPFLAGS_MANAGED_DICT bit is set in the ~PyTypeObject.tp_flags field, the traverse function must call PyObject_VisitManagedDict like this:

int err = PyObject_VisitManagedDict((PyObject*)self, visit, arg);
if (err) {
    return err;
}

Only the members that the instance owns (by having strong references <strong reference> to them) must be visited. For instance, if an object supports weak references via the ~PyTypeObject.tp_weaklist slot, the pointer supporting the linked list (what tp_weaklist points to) must not be visited as the instance does not directly own the weak references to itself.

The traversal function has a limitation:

[!WARNING] The traversal function must not have any side effects. Implementations may not modify the reference counts of any Python objects nor create or destroy any Python objects, directly or indirectly.

This means that most Python C API functions may not be used, since they can raise a new exception, return a new reference to a result object, have internal logic that uses side effects. Also, unless documented otherwise, functions that happen to not have side effects may start having them in future versions, without warning.

For a list of safe functions, see a separate section <duringgc-functions> below.

If an error occurs during traversal, ignore it and return 0. If tp_traverse fails to visit some strong references, garbage reference cycles involving the skipped references will not be reliably broken, but correctness won't be affected. For example, if traversal requires a lock that cannot be acquired promptly, skip the traversal. The garbage collector will visit your object again in a future pass.

[!NOTE] The Py_VISIT call may be skipped for those members that provably cannot participate in reference cycles. In the local_traverse example above, there is also a self->key member, but it can only be NULL or a Python string and therefore cannot be part of a reference cycle.

On the other hand, even if you know a member can never be part of a cycle, as a debugging aid you may want to visit it anyway just so the gc module's ~gc.get_referents function will include it.

[!NOTE] The ~PyTypeObject.tp_traverse function can be called from any thread.

Garbage collection is a "stop-the-world" operation: even in free threading builds, only one thread state is attached <attached thread state> when !tp_traverse handlers run.

3.9

Heap-allocated types are expected to visit Py_TYPE(self) in tp_traverse. In earlier versions of Python, due to bug 40217, doing this may lead to crashes in subclasses.

To simplify writing ~PyTypeObject.tp_traverse handlers, a Py_VISIT macro is provided. In order to use this macro, the ~PyTypeObject.tp_traverse implementation must name its arguments exactly visit and arg:

If the PyObject * o is not NULL, call the visit callback, with arguments o and arg. If visit returns a non-zero value, then return it.

This corresponds roughly to:

#define Py_VISIT(o)                             \
   if (o) {                                    \
      int visit_result = visit(o, arg);         \
      if (visit_result != 0) {                  \
         return visit_result;                   \
      }                                         \
   }

Traversal-safe functions

The following functions and macros are safe to use in a ~PyTypeObject.tp_traverse handler:

• the visit function passed to tp_traverse
• Py_VISIT
• Py_SIZE
• Py_TYPE: if called from a !tp_traverse handler, !Py_TYPE's result will be valid for the duration of the handler call
• PyObject_VisitManagedDict
• PyObject_TypeCheck, PyType_IsSubtype, PyType_HasFeature
• Py{<type>}_Check and Py{<type>}_CheckExact -- for example, PyTuple_Check
• duringgc-functions

"DuringGC" functions

The following functions should only be used in a ~PyTypeObject.tp_traverse handler; calling them in other contexts may have unintended consequences.

These functions act like their counterparts without the _DuringGC suffix, but they are guaranteed to not have side effects, they do not set an exception on failure, and they return/set borrowed references <borrowed reference> as detailed in the individual documentation.

Note that these functions may fail (return NULL or -1), but as they do not set an exception, no error information is available. In some cases, failure is not distinguishable from a successful NULL result.

See duringgc-functions for common information.

Sets *result to a borrowed reference rather than a strong one. The reference is valid for the duration of the !tp_traverse handler call.

3.15

PyType_GetBaseByToken

Controlling the Garbage Collector State

The C-API provides the following functions for controlling garbage collection runs.

Perform a full garbage collection, if the garbage collector is enabled. (Note that gc.collect runs it unconditionally.)

Returns the number of collected + unreachable objects which cannot be collected. If the garbage collector is disabled or already collecting, returns 0 immediately. Errors during garbage collection are passed to sys.unraisablehook. This function does not raise exceptions.

Enable the garbage collector: similar to gc.enable. Returns the previous state, 0 for disabled and 1 for enabled.

3.10

Disable the garbage collector: similar to gc.disable. Returns the previous state, 0 for disabled and 1 for enabled.

3.10

Query the state of the garbage collector: similar to gc.isenabled. Returns the current state, 0 for disabled and 1 for enabled.

3.10

Querying Garbage Collector State

The C-API provides the following interface for querying information about the garbage collector.

Run supplied callback on all live GC-capable objects. arg is passed through to all invocations of callback.

[!WARNING] If new objects are (de)allocated by the callback it is undefined if they will be visited.

Garbage collection is disabled during operation. Explicitly running a collection in the callback may lead to undefined behaviour e.g. visiting the same objects multiple times or not at all.

3.12

Type of the visitor function to be passed to PyUnstable_GC_VisitObjects. arg is the same as the arg passed to PyUnstable_GC_VisitObjects. Return 1 to continue iteration, return 0 to stop iteration. Other return values are reserved for now so behavior on returning anything else is undefined.

3.12