`multiprocessing.Array` and `multiprocessing.managers.SyncManager.Array` APIs are similar but not the same

[jthistle]

BPO	41423
Nosy	@pitrou, @applio, @bensimner, @jthistle

^{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 = None
created_at = <Date 2020-07-28.18:34:04.301>
labels = ['3.8', 'type-bug', 'library']
title = '`multiprocessing.Array` and `multiprocessing.managers.SyncManager.Array` APIs are similar but not the same'
updated_at = <Date 2020-07-28.18:36:30.664>
user = 'https://github.com/jthistle'

bugs.python.org fields:

activity = <Date 2020-07-28.18:36:30.664>
actor = 'xtreak'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2020-07-28.18:34:04.301>
creator = 'jthistle'
dependencies = []
files = []
hgrepos = []
issue_num = 41423
keywords = []
message_count = 1.0
messages = ['374519']
nosy_count = 4.0
nosy_names = ['pitrou', 'davin', 'bjs', 'jthistle']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue41423'
versions = ['Python 3.8']

[jthistle]

I stumbled across this by trying to implement a call to the latter, while reading the docs for the former.

I think this is quite confusing and unnecessary that the APIs between these two definitions should differ. The same goes for multiprocessing.Value and multiprocessing.managers.SyncManager.Value.

This is especially exacerbated by the fact that the docs are incorrect. On this page [1], under 'Server process', a link to 'Array' is made. If it were correct, it would link to multiprocessing.managers.SyncManager.Array, since it's talking about a manager object - the kind returned by Manager(). But it links to multiprocessing.Array. Of course, the simple solution would be to change the link to link to the correct place, but I think this shows an unnecessary inconsistency in the API itself.

I don't have a PR for this yet, nor have I fully investigated, but should it be feasible I would like to implement it myself. I'm interested to hear what people think.

[1] https://docs.python.org/3/library/multiprocessing.html#sharing-state-between-processes

[picnixz]

I'm not sure we should change how they behave. For multiprocessing.Array, we really "[r]eturn a ctypes array allocated from shared memory" while for manager-based objects, the purpose is to share the objects around, so we need proxy objects. The purpose being different, the API may differ as well.

I would however agree that we should probably change the links in the docs.

cc @gpshead as mp expert