Update documentation for tp_basicsize & tp_itemsize

- Add alignment requirement
- Mention that ob_size is unreliable if you don't control it
- Add some links for context
- basicsize should include the base type in generaly not just PyObject

This adds a “by-the-way” link to `PyObject_New`, which shouldn't be
used for GC types. In order to be comfortable linking to it, I also
add a link to `PyObject_GC_New` from its docs. And the same for
`*Var` variants, while I'm here.

Author: encukou

Doc/c-api/allocation.rst

@@ -35,6 +35,10 @@ Allocating Objects on the Heap
    The size of the memory allocation is determined from the
    :c:member:`~PyTypeObject.tp_basicsize` field of the type object.
 
+   Note that this function is unsuitable if *typeobj* has
+   :c:macro:`Py_TPFLAGS_HAVE_GC` set. For such objects,
+   use :c:func:`PyObject_GC_New` instead.
+
 
 .. c:macro:: PyObject_NewVar(TYPE, typeobj, size)
 
@@ -49,6 +53,10 @@ Allocating Objects on the Heap
    fields into the same allocation decreases the number of allocations,
    improving the memory management efficiency.
 
+   Note that this function is unsuitable if *typeobj* has
+   :c:macro:`Py_TPFLAGS_HAVE_GC` set. For such objects,
+   use :c:func:`PyObject_GC_NewVar` instead.
+
 
 .. c:function:: void PyObject_Del(void *op)
 

Doc/c-api/type.rst

@@ -456,6 +456,9 @@ The following functions and structs are used to create
       class need *in addition* to the superclass.
       Use :c:func:`PyObject_GetTypeData` to get a pointer to subclass-specific
       memory reserved this way.
+      For negative :c:member:`!basicsize`, Python will insert padding when
+      needed to meet :c:member:`~PyTypeObject.tp_basicsize`'s alignment
+      requirements.
 
       .. versionchanged:: 3.12
 

Doc/c-api/typeobj.rst

@@ -587,47 +587,81 @@ and :c:data:`PyType_Type` effectively act as defaults.)
 
 
 .. c:member:: Py_ssize_t PyTypeObject.tp_basicsize
-             Py_ssize_t PyTypeObject.tp_itemsize
+              Py_ssize_t PyTypeObject.tp_itemsize
 
    These fields allow calculating the size in bytes of instances of the type.
 
    There are two kinds of types: types with fixed-length instances have a zero
-   :c:member:`~PyTypeObject.tp_itemsize` field, types with variable-length instances have a non-zero
-   :c:member:`~PyTypeObject.tp_itemsize` field.  For a type with fixed-length instances, all
-   instances have the same size, given in :c:member:`~PyTypeObject.tp_basicsize`.
+   :c:member:`!tp_itemsize` field, types with variable-length instances have a non-zero
+   :c:member:`!tp_itemsize` field.  For a type with fixed-length instances, all
+   instances have the same size, given in :c:member:`!tp_basicsize`.
+   (Exceptions to this rule can be made using
+   :c:func:`PyUnstable_Object_GC_NewWithExtraData`.)
 
    For a type with variable-length instances, the instances must have an
-   :c:member:`~PyVarObject.ob_size` field, and the instance size is :c:member:`~PyTypeObject.tp_basicsize` plus N
-   times :c:member:`~PyTypeObject.tp_itemsize`, where N is the "length" of the object.  The value of
-   N is typically stored in the instance's :c:member:`~PyVarObject.ob_size` field.  There are
-   exceptions:  for example, ints use a negative :c:member:`~PyVarObject.ob_size` to indicate a
-   negative number, and N is ``abs(ob_size)`` there.  Also, the presence of an
-   :c:member:`~PyVarObject.ob_size` field in the instance layout doesn't mean that the instance
-   structure is variable-length (for example, the structure for the list type has
-   fixed-length instances, yet those instances have a meaningful :c:member:`~PyVarObject.ob_size`
-   field).
-
-   The basic size includes the fields in the instance declared by the macro
-   :c:macro:`PyObject_HEAD` or :c:macro:`PyObject_VAR_HEAD` (whichever is used to
-   declare the instance struct) and this in turn includes the  :c:member:`~PyObject._ob_prev` and
-   :c:member:`~PyObject._ob_next` fields if they are present.  This means that the only correct
-   way to get an initializer for the :c:member:`~PyTypeObject.tp_basicsize` is to use the
+   :c:member:`~PyVarObject.ob_size` field, and the instance size is
+   :c:member:`!tp_basicsize` plus N times :c:member:`!tp_itemsize`,
+   where N is the "length" of the object.
+
+   Functions like :c:func:`PyObject_NewVar` will take the value of N as an
+   argument, and store in the instance's :c:member:`~PyVarObject.ob_size` field.
+   Note that the :c:member:`~PyVarObject.ob_size` field may later be used for
+   other purposes. For example, :py:type:`int` instances use the bits of
+   :c:member:`~PyVarObject.ob_size` in an implementation-defined
+   way; the underlying storage and its size should be acessed using
+   :c:func:`PyLong_Export`.
+
+   Also, the presence of an :c:member:`~PyVarObject.ob_size` field in the
+   instance layout doesn't mean that the instance structure is variable-length.
+   For example, the :py:type:`list` type has fixed-length instances, yet those
+   instances have a :c:member:`~PyVarObject.ob_size` field.
+   (As with :py:type:`int`, avoid reading lists' :c:member:`!ob_size` directly.
+   Call :c:func:`PyList_Size` instead.)
+
+   The :c:member:`!tp_basicsize` includes size needed for data of the type's
+   :c:member:`~PyTypeObject.tp_base`, plus any extra data needed
+   by each instance.
+
+   The  correct way to set :c:member:`!tp_basicsize` is to use the
    ``sizeof`` operator on the struct used to declare the instance layout.
-   The basic size does not include the GC header size.
-
-   A note about alignment: if the variable items require a particular alignment,
-   this should be taken care of by the value of :c:member:`~PyTypeObject.tp_basicsize`.  Example:
-   suppose a type implements an array of ``double``. :c:member:`~PyTypeObject.tp_itemsize` is
-   ``sizeof(double)``. It is the programmer's responsibility that
-   :c:member:`~PyTypeObject.tp_basicsize` is a multiple of ``sizeof(double)`` (assuming this is the
-   alignment requirement for ``double``).
-
-   For any type with variable-length instances, this field must not be ``NULL``.
+   This struct must include the struct used to declare the base type.
+   In other words, :c:member:`!tp_basicsize` must be greater than or equal
+   to the base's :c:member:`!tp_basicsize`.
+
+   Since every type is a subtype of :py:type:`object`, this struct must
+   include :c:type:`PyObject` or :c:type:`PyVarObject` (depending on
+   whether :c:member:`~PyVarObject.ob_size` should be included). These are
+   usually defined by the macro :c:macro:`PyObject_HEAD` or
+   :c:macro:`PyObject_VAR_HEAD`, respectively.
+
+   The basic size does not include the GC header size, as that header is not
+   part of :c:macro:`PyObject_HEAD`.
+
+   For cases where struct used to declare the base type is unknown,
+   see :c:member:`PyType_Spec.basicsize` and :c:func:`PyType_FromMetaclass`.
+
+   Notes about alignment:
+
+   - :c:member:`!tp_basicsize` must be a multiple of ``_Alignof(PyObject)``.
+     When using ``sizeof`` on a ``struct`` that includes
+     :c:macro:`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,
+     :c:member:`!tp_basicsize` and :c:member:`!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.  If the base type has a
-   non-zero :c:member:`~PyTypeObject.tp_itemsize`, it is generally not safe to set
+   These fields are inherited separately by subtypes.
+   (That is, if the field is set to zero, :c:func:`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 :c:member:`~PyTypeObject.tp_itemsize`, it is generally not safe to set
    :c:member:`~PyTypeObject.tp_itemsize` to a different non-zero value in a subtype (though this
    depends on the implementation of the base type).