Rewrite _reduce and _reconstructor in C

[gvanrossum]

BPO	614555
Nosy	@gvanrossum, @tim-one, @brettcannon, @avassalotti

^{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/gvanrossum'
closed_at = <Date 2008-01-04.01:13:17.995>
created_at = <Date 2002-09-25.18:15:01.000>
labels = ['interpreter-core', 'type-feature']
title = 'Rewrite _reduce and _reconstructor in C'
updated_at = <Date 2008-01-04.01:13:17.994>
user = 'https://github.com/gvanrossum'

bugs.python.org fields:

activity = <Date 2008-01-04.01:13:17.994>
actor = 'gvanrossum'
assignee = 'gvanrossum'
closed = True
closed_date = <Date 2008-01-04.01:13:17.995>
closer = 'gvanrossum'
components = ['Interpreter Core']
creation = <Date 2002-09-25.18:15:01.000>
creator = 'gvanrossum'
dependencies = []
files = []
hgrepos = []
issue_num = 614555
keywords = []
message_count = 7.0
messages = ['12478', '12479', '12480', '12481', '12482', '55903', '59200']
nosy_count = 4.0
nosy_names = ['gvanrossum', 'tim.peters', 'brett.cannon', 'alexandre.vassalotti']
pr_nums = []
priority = 'normal'
resolution = 'wont fix'
stage = None
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue614555'
versions = ['Python 2.6', 'Python 2.5']

[gvanrossum]

The copy_reg defines two functions, _reduce() and
_reconstructor(), that are used for the pickling and
unpickling of new-style classes. These were originally
written in Python because the right implementation
wasn't entirely clear; _reduce() is actually called
from a C-level __reduce__ method defined in 'object',
and it references _reconstructor() in its return tuple.
It is now time to move both back into C for efficiency.

[gvanrossum]

Logged In: YES
user_id=6380

I'd like to add some comments from Jim Fulton on this (or a
related) issue so they don't get lost.

"""
I still need to think of a good way to handle this.
Currently, the pattern is a reduce return value of the form:

reconstructor, (some_class, object, None), some_state

but there is a more general case where an uninitialized
object can be gotten by calling "type.__new__(some_class)",
or even: "some_class.__new__(some_class)".

There are really two issues:

• It's too hard to write __reduce__ functions for this case, and

• Too much data needs to be stored in pickles.

A new pickling code would handler the later, but I also want
to solve the former problem.

One thought is to return: (copy_reg.new, (some_class, ),
some_state) where copy_reg.new is:

   def new(class_): return class.__new__(class_)

The pickler could easily spot reduce returns with
copy_reg.new as the first value and generate a special
pickle code.
"""

[gvanrossum]

Logged In: YES
user_id=6380

Lowering priority; when you use pickle protocol 2, none of
this Python code is used any more.

Jim's comment has been addressed by __newobj__ and
__getnewargs__ (see PEP-307).

[tim-one]

Logged In: YES
user_id=31435

Except that the new Python code _better_reduce() is used
then. As discussed, I intend to rewrite _reduce and
_better_reduce in C, inside object's __reduce__
implementation, so I assigned this to me and boosted the
priority. I don't intend to recode _reconstructor in C,
though.

[gvanrossum]

Logged In: YES
user_id=6380

This isn't going to be done for 2.3a2; we'll revisit who
gets to do it closer to 2.3b1.

[brettcannon]

Classifying as an RFE since this is not critical (as shown by it not
happening since early 2003).

[gvanrossum]

I just don't see this happening. Possibly it's out of date. Lots of
things have changed in the pickle world since 2002.