Changes in the inspect module for PEP 570

[pablogsal]

BPO	36751
Nosy	@brettcannon, @ncoghlan, @vstinner, @ambv, @serhiy-storchaka, @zooba, @hroncok, @pablogsal
PRs	bpo-36751: Deprecate getfullargspec and report positional-only args as regular args #13016 bpo-36751: Undeprecate getfullargspec #13245

^{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 2019-05-17.11:23:42.943>
created_at = <Date 2019-04-29.12:13:17.185>
labels = ['interpreter-core', '3.8']
title = 'Changes in the inspect module for PEP 570'
updated_at = <Date 2019-05-22.12:21:34.359>
user = 'https://github.com/pablogsal'

bugs.python.org fields:

activity = <Date 2019-05-22.12:21:34.359>
actor = 'ncoghlan'
assignee = 'none'
closed = True
closed_date = <Date 2019-05-17.11:23:42.943>
closer = 'pablogsal'
components = ['Interpreter Core']
creation = <Date 2019-04-29.12:13:17.185>
creator = 'pablogsal'
dependencies = []
files = []
hgrepos = []
issue_num = 36751
keywords = ['patch', '3.7regression']
message_count = 30.0
messages = ['341070', '341075', '341081', '341082', '341083', '341098', '341101', '341102', '341104', '341115', '341124', '341128', '341130', '342156', '342157', '342158', '342197', '342295', '342296', '342362', '342426', '342666', '342667', '342669', '342670', '342679', '342683', '342685', '342698', '343187']
nosy_count = 9.0
nosy_names = ['brett.cannon', 'ncoghlan', 'vstinner', 'zzzeek', 'lukasz.langa', 'serhiy.storchaka', 'steve.dower', 'hroncok', 'pablogsal']
pr_nums = ['13016', '13245']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue36751'
versions = ['Python 3.8']

[pablogsal]

This issue is to discuss how to handle the changes in the inspect module for PEP-570. In particular, how to handle:

• getfullargspec
• formatargspec

for positional-only parameters.

[vstinner]

Context: see #12701 discussion.

[zooba]

Marking as a release blocker and adding RM. The 3.8 branch now contains breaking API changes to these functions that we have to resolve before the next release (or revert while deciding how to handle the change) - PR 12701 has been merged.

[zooba]

My bad, there were PR changes that GitHub didn't show me. As you were

[zooba]

Nope, I was right the first time. The FullArgSpec tulle has changed indexes, and formatargspec has additional (undocumented) arguments.

Since formatargspec is deprecated, it should probably just not change.

The FullArgSpec tuple might have to become a concrete class implementation so that it can have positional-only args as a named attribute but not part of unpacking.

As for where the new information *can* go, I suspect the Signature object is the only place. Though that has limitations, and for C APIs in particular much of my code only uses the more reliable getfullargspec.

[brettcannon]

Steve's right that we can't change the indexes on those functions for pre-existing values, so appending to the end is the only real option in that case. I still think they should both be deprecated due to how limiting their APIs are.

Obviously signature() can be updated cleanly.

[vstinner]

IMHO getargs(), getargspec(), getfullargspec() should be deprecated in favor of signature() which is future-proof. But Nick Coghlan undeprecated these functions in 2017 https://bugs.python.org/issue20438 whereas these were deprecated since 2015. I didn't understand the subtle issues. That's why I asked Pablo to open a separated issue, to get Nick and others involved to get the context.

My proposal was to raise an exception if the input function has positonal arguments. There is good way to fix these legacy functions, their design cannot be extended, whereas signature() is fine.

[zooba]

My proposal was to raise an exception if the input function has positonal arguments

Counter-proposal - just report them as regular position-or-name arguments.

Users of inspect are most likely to be IDEs or editors getting information to present to users. And in my experience, most just ignore all exceptions because bugs cause some function objects to raise errors when trying to inspect them. Which means a new exception will just result in a worse experience for users and no indication of what's wrong.

On the other hand, if there is no change made, they'll just report a normal looking signature until they migrate to using updated Signature objects. Anyone using current Signature objects should similarly get the argument count right, even if they totally ignore the positional-only count. Since one of the arguments for supporting positional-only parameters is that they're for parameters that you'd never consider passing by name, but these tools still need to show a name, very few people should ever be tripped up by them being misreported. And when they do they'll be tripped up very quickly.

Basically, this is not a scenario that *requires* existing users to be broken. So we should avoid breaking them at all costs.

[brettcannon]

If I remember correctly the un-deprecation was because 2/3 code could only use getfullargspec() to get all relevant details. Since we're so close to being done w/ Python 2 I think this might be the last change to make to them, document their deprecation, tack this info on to the end of the tuple, and then start raising DeprecationWarning in 3.9.

[ambv]

I'm with Steve on this one. Report positional-only args as regular arguments in the old functions.