[C API] How many slots of static types should be exposed in PyType_GetSlot()

[shihai1991]

BPO	41618
Nosy	@vstinner, @encukou, @shihai1991

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2020-08-23.08:48:18.095>
labels = ['expert-C-API', 'type-feature', '3.10']
title = '[C API] How many slots of static types should be exposed in PyType_GetSlot()'
updated_at = <Date 2021-01-16.05:36:16.127>
user = 'https://github.com/shihai1991'

bugs.python.org fields:

activity = <Date 2021-01-16.05:36:16.127>
actor = 'shihai1991'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['C API']
creation = <Date 2020-08-23.08:48:18.095>
creator = 'shihai1991'
dependencies = []
files = []
hgrepos = []
issue_num = 41618
keywords = []
message_count = 10.0
messages = ['375808', '375809', '376568', '379110', '379948', '380303', '380414', '380510', '383973', '385135']
nosy_count = 3.0
nosy_names = ['vstinner', 'petr.viktorin', 'shihai1991']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue41618'
versions = ['Python 3.10']

[shihai1991]

In order to resolve bpo-41073, we want to extend PyType_GetSlot() to accept static types. But there have another question we should be considered: How many slots should be exposed in PyType_GetSlot().

petr's opinion from PR-21931:I would not want to expose the Py_tp_as_* slots at all; their contents are covered in other slots that should be used instead. And slots like Py_tp_version_tag are specific to CPython; I don't think they should be exposed in the general C API.

[shihai1991]

related bpo: bpo-41073

[encukou]

IMO, a slot should be exposed if there is some reason for exposing it.

bpo-41073 talks about tp_dealloc, which is already exposed as Py_tp_dealloc.
Recently, Py_bf_* slots were exposed (but not as part of the limited API), in bpo-40724 which links to the reasons for exposing them.

Which other slots are we missing?

[encukou]

Also: PyType_GetSlot should return a function pointer, not data.

For some slots (Py_tp_doc, Py_tp_methods, Py_tp_members, Py_tp_getset) that ship has sailed, but for new ones, we should honor the distinction between data and function pointers. The C standard says they're distinct, so it might bite us in the future (e.g. compiling to WebAssembly).

[shihai1991]

IMO, a slot should be exposed if there is some reason for exposing it.

Can we write those info to docs? In case some developer want to exposing slot in future.

Which other slots are we missing?

nb_inserved in PyNumberMethods should be removed?

[encukou]

Can we write those info to docs? In case some developer want to exposing slot in future.

Do you mean something like "only expose a slot if you have a reason for exposing it"? That sounds like a tautology.
Where in the docs would this go?

nb_inserved in PyNumberMethods should be removed?

You mean nb_reserved, right?
It can't be removed; old-style initialization is positional.
But it could be replaced according to PEP-387:

• in 3.x, raise a warning if it is not set to NULL (in static types)
• in 3.x+2, throw an error if it is not set to NULL (in static types)
• in 3.x+3 or later, it can be replaced by another slot, if there's something we need to add to PyNumberMethods

[encukou]

My views on individual slots:

Not a function, shouldn't be used with PyType_Slot:

• tp_dict - I'd add a PyType_GetDict if there is a need to get this
• tp_mro - I'd add a PyType_GetMRO if there is a need to get this
  (There are existing slots that aren't functions; I'd like to deprecate them in the long term.)

internal use only, do not expose:

• tp_cache
• tp_subclasses
• tp_weaklist
• tp_as_*

Not part of limited API, do not expose:

• tp_vectorcall

Can be exposed if we deem all of the related API stable:

• bf_getbuffer
• bf_releasebuffer

[shihai1991]

Do you mean something like "only expose a slot if you have a reason for > exposing it"? That sounds like a tautology.
Where in the docs would this go?

Add a friend Note in docs of PyType_GetSlot MAYBE?

You mean nb_reserved, right?
Yes, thanks for your description.

[encukou]

Documentation for developers *of* CPython should generally go in the dev guide, a comment in the code, or in issues like this.