structseq types should expose _fields

[abarnert]

BPO	20230
Nosy	@skrah, @meadori, @ethanfurman, @1st1
Superseder	bpo-1820: Enhance Object/structseq.c to match namedtuple and tuple api
Files	structseq.diff

^{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 2014-04-29.11:56:03.434>
created_at = <Date 2014-01-13.01:12:54.433>
labels = ['interpreter-core', 'type-feature', 'library']
title = 'structseq types should expose _fields'
updated_at = <Date 2014-04-29.11:58:18.088>
user = 'https://bugs.python.org/abarnert'

bugs.python.org fields:

activity = <Date 2014-04-29.11:58:18.088>
actor = 'skrah'
assignee = 'none'
closed = True
closed_date = <Date 2014-04-29.11:56:03.434>
closer = 'skrah'
components = ['Interpreter Core', 'Library (Lib)']
creation = <Date 2014-01-13.01:12:54.433>
creator = 'abarnert'
dependencies = []
files = ['33436']
hgrepos = []
issue_num = 20230
keywords = ['patch']
message_count = 2.0
messages = ['207991', '207993']
nosy_count = 5.0
nosy_names = ['skrah', 'meador.inge', 'ethan.furman', 'yselivanov', 'abarnert']
pr_nums = []
priority = 'normal'
resolution = 'duplicate'
stage = 'resolved'
status = 'closed'
superseder = '1820'
type = 'enhancement'
url = 'https://bugs.python.org/issue20230'
versions = ['Python 3.5']

[abarnert]

A PyStructSequence type doesn't expose its field names in any way to Python code (except indirectly, via the repr, which you could parse if you really wanted to…).

The docs claim that "Struct sequence objects are the C equivalent of namedtuple() objects", so it seems like the obvious way to expose them is the exact same way namedtuple types do—as a class attribute named "_fields".

Also, in bpo-7796, and in a recent thread on python-ideas, multiple people suggested that "named tuple" is a protocol, not a type, and duck typing on _fields is the obvious way to detect types matching that protocol.

Only the "sequence fields" need to be exposed—fields that don't match up to sequence indices aren't named tuple fields, they're just extra instance attributes (that don't even show up in the repr).

Patch included. For the tests, I tested the fields in struct_time and stat_result in effectively the same way test_repr tests their repr, so it should pass on all the same platforms/builds.

[abarnert]

After more searching (I should have searched on structseq instead of PyStructSequence…), this might be a dup of bpo-1820. If so, apologies.