Should shelve support dict union?

[Dominik1123]

BPO	43443
Nosy	@rhettinger, @ambv, @serhiy-storchaka, @miss-islington, @brandtbucher, @Dominik1123
PRs	bpo-43443: Clarify difference between shelve objects and dicts #27004 [3.10] bpo-43443: Clarify difference between shelve objects and dicts (GH-27004) #27369 [3.9] bpo-43443: Clarify difference between shelve objects and dicts (GH-27004) #27370

^{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 2021-07-26.19:33:20.289>
created_at = <Date 2021-03-09.08:46:35.421>
labels = ['type-bug', 'library', '3.9', '3.10', '3.11']
title = 'Should shelve support dict union?'
updated_at = <Date 2021-07-26.19:33:20.288>
user = 'https://github.com/Dominik1123'

bugs.python.org fields:

activity = <Date 2021-07-26.19:33:20.288>
actor = 'lukasz.langa'
assignee = 'none'
closed = True
closed_date = <Date 2021-07-26.19:33:20.289>
closer = 'lukasz.langa'
components = ['Library (Lib)']
creation = <Date 2021-03-09.08:46:35.421>
creator = 'Dominik V.'
dependencies = []
files = []
hgrepos = []
issue_num = 43443
keywords = ['patch']
message_count = 10.0
messages = ['388335', '388338', '388342', '388388', '388389', '388394', '388395', '398242', '398252', '398253']
nosy_count = 6.0
nosy_names = ['rhettinger', 'lukasz.langa', 'serhiy.storchaka', 'miss-islington', 'brandtbucher', 'Dominik V.']
pr_nums = ['27004', '27369', '27370']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue43443'
versions = ['Python 3.9', 'Python 3.10', 'Python 3.11']

[Dominik1123]

The docs of shelve mention that

Shelf objects support all methods supported by dictionaries. This eases the transition from dictionary based scripts to those requiring persistent storage.

However the |= operator is not implemented, preventing a seamless transition from dict to shelve. So should this be implemented for Shelf as well? | on the other hand doesn't make much sense.

Otherwise the docs could be updated.

[serhiy-storchaka]

The comment is outdated. Shelf objects also do not support methods copy and fromkey. Creating a new Shelve object without specifying a new underlying database object does not make much sense.

Maybe say that they implement the MutableMapping interface?

>>> sorted(set(dir(dict)) - set(dir(shelve.Shelf)))
['__ior__', '__or__', '__ror__', 'copy', 'fromkeys']
>>> sorted(set(dir(collections.abc.MutableMapping)) - set(dir(shelve.Shelf)))
[]

[Dominik1123]

Right, that seems like a good idea. But __ior__ could be implemented nevertheless?

[rhettinger]

Maybe say that they implement the MutableMapping interface?

Yes, that's a good idea (also add link to the MutableMapping docs).

Right, that seems like a good idea. But __ior__ could
be implemented nevertheless?

Off-hand, I don't see why not.

Adding Brandt to the nosy list to see what he thinks.

[rhettinger]

One other thought. While __ior__() could be used as a short-cut for update(), the related __or__() method is more problematic because it would need to create a new underlying DB. So maybe this is a can worms that we should not open. It would be weird to have __ior__() but not __or__().

[brandtbucher]

+1 for the MutableMapping comment.

We purposely omitted shelve when determining what classes should grow the new operators. Guido's thoughts:

I definitely think we should leave Shelf alone, it's a toy class from a different era.

(https://bugs.python.org/msg364196)

I personally have no experience with shelve, so I'd rather defer to the judgement of others here.

[Dominik1123]

It's true, having __ior__ but not __or__ would probably be weird. In the end it's just "nice to have", but I'm not even sure that this applies. Calling db.update(...) is still more explicit than db |= .... The docs mention that

This eases the transition from dictionary based scripts to those requiring persistent storage.

For my use cases, however, I always knew right from the beginning that I want object persistence between different runs of a script (e.g. for data analysis, caching the expensive results), so it was always clear that I'm working with a Shelf object and not a dict (i.e. no expectations on the availability of |=).

Primarily, this issue was meant to point out the mismatch of docs/implementation and not to get |= implemented for Shelf. In the end, I think updating the docs is all that is needed.