Backport next()

[birkenfeld]

BPO	2719
Nosy	@gvanrossum, @birkenfeld, @rhettinger, @abalkin
Files	nextbackport.diff: Support "bpo-" in Misc/NEWS #1

^{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 2008-04-30.19:47:35.211>
created_at = <Date 2008-04-29.21:02:25.876>
labels = ['interpreter-core', 'type-bug']
title = 'Backport next()'
updated_at = <Date 2008-04-30.19:47:35.013>
user = 'https://github.com/birkenfeld'

bugs.python.org fields:

activity = <Date 2008-04-30.19:47:35.013>
actor = 'georg.brandl'
assignee = 'gvanrossum'
closed = True
closed_date = <Date 2008-04-30.19:47:35.211>
closer = 'georg.brandl'
components = ['Interpreter Core']
creation = <Date 2008-04-29.21:02:25.876>
creator = 'georg.brandl'
dependencies = []
files = ['10141']
hgrepos = []
issue_num = 2719
keywords = ['patch', '26backport']
message_count = 13.0
messages = ['65980', '65992', '65993', '65995', '66000', '66004', '66005', '66006', '66007', '66008', '66009', '66010', '66017']
nosy_count = 4.0
nosy_names = ['gvanrossum', 'georg.brandl', 'rhettinger', 'belopolsky']
pr_nums = []
priority = 'critical'
resolution = 'accepted'
stage = None
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue2719'
versions = ['Python 2.6']

[birkenfeld]

Backporting 3.0's next() builtin.

There's no change w.r.t. __next__/next, that is tracked in bpo-2336.

[rhettinger]

ISTM, the only value added by next(g) is that it replaces g.next() with
a more conventional spelling, g.__next__(). Since 2.6 still has g.next
(),I don't see how this backport adds value. It does however create a
second way to do it that will be confusing to some remaining in the 2.x
world. I think we should avoid double spellings in 2.6 except in cases
where the 2-to-3 converter would need help.

[birkenfeld]

IMO having next() in 2.6 helps since if you use it consistently you
don't have to care about calling .next() or .__next__().

Also, I don't see how this is different from having e.g. reduce() and
functools.reduce() in 2.6.

[rhettinger]

The problem is with the "if you use it consistently" premise. That
will not hold in an environment with legacy code, multiple programmers,
lots of code in ASPN recipes and published materials, and third-party
modules. A patch like this dooms Py2.6 programmers to seeing both of
these forms intermixed throughout the code base. This is *not* a win.

[gvanrossum]

I think it's important to make this available in 2.6; it will let people
writing 3.0-oriented code in 2.6 (as several developers are planning to
do) do the right thing for this syntax.

[abalkin]

I thought new code is supposed to use Py_TYPE macro instead of ->ob_type:

+ "%.200s object is not an iterator", it->ob_type-

tp_name);
..
+ res = (*it->ob_type->tp_iternext)(it);

Py3k branch has the same issue.

[abalkin]

One more question: What is the rationale for

+ res = (*it->ob_type->tp_iternext)(it);
+ if (res == NULL) {
..
+ PyErr_SetNone(PyExc_StopIteration);
+ return NULL;

?

I would think tp_iternext failing to set an exception should not be
translated into stop iteration. Instead, builtin_next() should return
NULL without an exception set and thus trigger a SystemError.

[gvanrossum]

I would think tp_iternext failing to set an exception should not be
translated into stop iteration. Instead, builtin_next() should return
NULL without an exception set and thus trigger a SystemError.

Wrong; the iternext slot is designed to return NULL without setting an
exception. See e.g. listiter_next().

[gvanrossum]

+1 on this. I have a few nits about the code:

Line 1083: "%.200s object is not an iterator", it->ob_type->tp_name);
Line is too long.

Line 1088: if (res == NULL) {
How about

if (res != NULL)
return res;

?

Line 1089: if (def) {
if (def != NULL) {

Line 1093: PyErr_Clear();
I would only call this if PyErr_Occurred() returns true.

[abalkin]

On Wed, Apr 30, 2008 at 12:41 PM, Guido van Rossum
<report@bugs.python.org> wrote:

Guido van Rossum <guido@python.org> added the comment:

> .. builtin_next() should return
> NULL without an exception set and thus trigger a SystemError.

Wrong; the iternext slot is designed to return NULL without setting an
exception. See e.g. listiter_next().

I did not know that. Thanks for the explanation. In this case,
wouldn't it be cleaner to call PyIter_Next which is documented to
return NULL with no exception?

[abalkin]

On Wed, Apr 30, 2008 at 12:41 PM, Guido van Rossum
<report@bugs.python.org> wrote:

the iternext slot is designed to return NULL without setting an
exception.

This is not what the documentation says:

"""
iternextfunc PyTypeObject.tp_iternext
An optional pointer to a function that returns the next item in an
iterator, or raises StopIteration when the iterator is exhausted.
""" <http://docs.python.org/dev/c-api/typeobj.html#tp_iternext\>

It looks like documentation needs to be updated, but wouldn't it be
odd to specify that setting StopIteration exception is optional? It's
probably more logical to intercept StopIteration in slot_tp_iternext
rather than at every place where tp_iternext is called.

[gvanrossum]

Feel free to submit a patch to fix the docs.

Changing the API is not an option -- it's been like this since the
tp_iternext slot was added, and it's been designed like this for a
reason: so that in the common case of iterating over a built-in sequence
type no exception objects have to be created. In particular the
for-loop code would just discard the StopIteration instance again.

The requirement that the exception is *optional* is so that if you're
calling a Python iterator that *does* create the exception, the
exception object (with whatever data the creator might have attached to
it) doesn't get lost (or worse, have to be recreated).

Calling PyIter_Next() here instead of inlining it would not be
advantageous; it would just slow things down since we'd have to make a
redundant call to PyErr_Occurred() to distinguish the StopIteration case
from other errors.

[birkenfeld]

Updated and committed as r62599. Also fixed your nits in the original 3k
version in r62598.