Deprecate delegation of int to __trunc__

[mdickinson]

BPO	44977
Nosy	@rhettinger, @mdickinson, @serhiy-storchaka, @ZackerySpytz
PRs	bpo-44977: Deprecate delegation of int to __trunc__ #31031

^{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 = <Date 2022-02-03.09:44:44.771>
created_at = <Date 2021-08-22.09:17:57.051>
labels = ['interpreter-core', '3.11']
title = 'Deprecate delegation of int to __trunc__'
updated_at = <Date 2022-02-03.09:44:44.770>
user = 'https://github.com/mdickinson'

bugs.python.org fields:

activity = <Date 2022-02-03.09:44:44.770>
actor = 'serhiy.storchaka'
assignee = 'none'
closed = True
closed_date = <Date 2022-02-03.09:44:44.771>
closer = 'serhiy.storchaka'
components = ['Interpreter Core']
creation = <Date 2021-08-22.09:17:57.051>
creator = 'mark.dickinson'
dependencies = []
files = []
hgrepos = []
issue_num = 44977
keywords = ['patch']
message_count = 12.0
messages = ['400063', '400064', '400073', '400074', '400076', '400078', '400079', '400080', '400082', '412190', '412431', '412432']
nosy_count = 4.0
nosy_names = ['rhettinger', 'mark.dickinson', 'serhiy.storchaka', 'ZackerySpytz']
pr_nums = ['31031']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue44977'
versions = ['Python 3.11']

[mdickinson]

The int constructor, when applied to a general Python object obj, first looks for an int method, then for an index method, and then finally for a trunc method.

The delegation to __trunc__ used to be useful: it meant that users could write a custom class SomeNumber with the property that:

• SomeNumber instances supported 'int' calls, returning a truncated value, but
• SomeNumber instances weren't usable in indexing, chr() calls, and all the various other calls that implicitly invoked __int__.

class SomeNumber:
    def __trunc__(self):
       <return truncated value for self>

However, with Python >= 3.10, we no longer use __int__ implicitly for argument conversion in internal code. So the second point above is no longer a concern, and SomeNumber can now simply be written as

class SomeNumber:
    def __int__(self):
       <return truncated value for self>

This decouples int from __trunc__ and leaves __trunc__ as simply the support for the math.trunc function.

[mdickinson]

This decouples int from __trunc__ and leaves __trunc__ as simply the support for the math.trunc function.

Argh; copy and paste fail - I left out the crucial line.

I propose deprecating the delegation of int to __trunc__: calls to int(a) where type(a) implements __trunc__ but not __int__ or __index__ would raise a DeprecationWarning, and at some point in the future int(a) would become a TypeError.

[serhiy-storchaka]

Completely agree.

[rhettinger]

Afterwards, do __trunc__ and math.trunc() still need to exist?

They were mostly unused. Now they would also be unnecessary. Just having them around would be a point of confusion.

[mdickinson]

A GitHub code search shows a substantial number of uses of math.trunc out in the wild:

https://github.com/search?q=math.trunc+extension%3Apy&type=Code

So unfortunately, getting rid of math.trunc and __trunc__ looks a bit too much as though it would cause gratuitous breakage.

[rhettinger]

Possibly, we could have math.trunc() call __int__, letting us deprecate __trunc__.

That would let users avoid gratuitous aliasing such as that in _pydecimal.py:

   __trunc__ = __int__

[serhiy-storchaka]

I do not think that math.trunc() for UUID or IP4Address makes sense.

[mdickinson]

I think the needs are sufficiently different that __trunc__ still has value as its own thing, and it's a natural counterpart to __floor__ and __ceil__ (out of all the directed rounding modes, rounding towards zero probably turns up more often than rounding towards +inf or -inf).

__int__ has the disadvantage that it must return an int, so it's not useful for other Python-based numeric systems with their own rational number / integer pairing. For example, with gmpy2, truncating an mpq instance returns an mpz instance, which is what I'd expect and want - if I'm working with huge values then it would be annoying for math.trunc to do the inefficient thing and return a plain int.

>>> from gmpy2 import mpq
>>> math.trunc(mpq(22, 7))
mpz(3)

SymPy behaves similarly:

>>> from sympy import Rational
>>> type(math.trunc(Rational(22, 7)))
<class 'sympy.core.numbers.Integer'>

On the other side, there are conversions to integer that it doesn't make sense to think of as truncation. It would be odd for a UUID to be a valid input to math.trunc, for example:

>>> int(uuid.uuid4())
43341414945793370480422623795805641537

[rhettinger]

I retract the suggestion to deprecate __trunc__.

[ZackerySpytz]

I have created a patch for this issue. Please consider having a look.

[serhiy-storchaka]

New changeset b4bd1e1 by Zackery Spytz in branch 'main':
bpo-44977: Deprecate delegation of int to __trunc__ (GH-31031)
b4bd1e1

[serhiy-storchaka]

Thanks Zackery.