weakref proxy doesn't support the matrix multiplication operator

[mr-nfamous]

BPO	36669
Nosy	@freddrake, @mdickinson, @njsmith, @mr-nfamous
PRs	bpo-36669: Add matmul support for weakref.proxy #12932

^{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 = 'https://github.com/mdickinson'
closed_at = <Date 2019-04-26.08:46:22.259>
created_at = <Date 2019-04-19.14:08:57.116>
labels = ['3.8', 'type-feature', 'library']
title = "weakref proxy doesn't support the matrix multiplication operator"
updated_at = <Date 2019-04-26.08:46:22.258>
user = 'https://github.com/mr-nfamous'

bugs.python.org fields:

activity = <Date 2019-04-26.08:46:22.258>
actor = 'mark.dickinson'
assignee = 'mark.dickinson'
closed = True
closed_date = <Date 2019-04-26.08:46:22.259>
closer = 'mark.dickinson'
components = ['Library (Lib)']
creation = <Date 2019-04-19.14:08:57.116>
creator = 'bup'
dependencies = []
files = []
hgrepos = []
issue_num = 36669
keywords = ['patch']
message_count = 6.0
messages = ['340553', '340760', '340763', '340767', '340885', '340894']
nosy_count = 5.0
nosy_names = ['fdrake', 'mark.dickinson', 'njs', 'SilentGhost', 'bup']
pr_nums = ['12932']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue36669'
versions = ['Python 3.8']

[SilentGhost]

It's not obvious why it should. Do you care to show a use case you had in mind?

[mdickinson]

It's not obvious why it should.

I'd say that it's not obvious why it shouldn't. The "weakproxy" type currently supports every single "PyNumberMethods" slot *except* matrix multiplication: see

cpython/Objects/weakrefobject.c

Lines 610 to 645 in bf94cc7

	static PyNumberMethods proxy_as_number = {
	proxy_add, /*nb_add*/
	proxy_sub, /*nb_subtract*/
	proxy_mul, /*nb_multiply*/
	proxy_mod, /*nb_remainder*/
	proxy_divmod, /*nb_divmod*/
	proxy_pow, /*nb_power*/
	proxy_neg, /*nb_negative*/
	proxy_pos, /*nb_positive*/
	proxy_abs, /*nb_absolute*/
	(inquiry)proxy_bool, /*nb_bool*/
	proxy_invert, /*nb_invert*/
	proxy_lshift, /*nb_lshift*/
	proxy_rshift, /*nb_rshift*/
	proxy_and, /*nb_and*/
	proxy_xor, /*nb_xor*/
	proxy_or, /*nb_or*/
	proxy_int, /*nb_int*/
	0, /*nb_reserved*/
	proxy_float, /*nb_float*/
	proxy_iadd, /*nb_inplace_add*/
	proxy_isub, /*nb_inplace_subtract*/
	proxy_imul, /*nb_inplace_multiply*/
	proxy_imod, /*nb_inplace_remainder*/
	proxy_ipow, /*nb_inplace_power*/
	proxy_ilshift, /*nb_inplace_lshift*/
	proxy_irshift, /*nb_inplace_rshift*/
	proxy_iand, /*nb_inplace_and*/
	proxy_ixor, /*nb_inplace_xor*/
	proxy_ior, /*nb_inplace_or*/
	proxy_floor_div, /*nb_floor_divide*/
	proxy_true_div, /*nb_true_divide*/
	proxy_ifloor_div, /*nb_inplace_floor_divide*/
	proxy_itrue_div, /*nb_inplace_true_divide*/
	proxy_index, /*nb_index*/
	};

It seems likely that this was an unintentional omission when matrix multiplication was added. I can't think of any good reason to support all but one (well, two, including the in-place method) of those PyNumberMethods. Supporting all of them gives a (conceptually) smaller, simpler object.

I think this is a consistency bug that should be fixed.

[mdickinson]

Adding Nathaniel Smith (the matrix multiplication PEP author), just in case he knows some reason why weakref.proxy objects should _not_ support the matrix multiplication operator.

[njsmith]

Yeah, seems like a simple oversight to me. (Actually this is the first I've heard of weakref.proxy...)

[mdickinson]

New changeset 7abb6c0 by Mark Dickinson in branch 'master':
bpo-36669: add matmul support to weakref.proxy (GH-12932)
7abb6c0

[mdickinson]

Now fixed for 3.8. Although this was an oversight, I'm erring on the side of regarding this as a new feature for 3.8 (and I think the omission is unlikely to bite anyone for Python 3.7 or 3.6), so I wouldn't recommend backporting this.

Thanks for the report!