Generated facade code types are too wide

[dimaqq]

Description

According to
https://github.com/dimaqq/juju-schema-analysis/blob/0acd07573bb2c192785796381f0594470ef296d8/schemas-juju-3.6.0.model-user.txt#L741C1-L753C12

Application.Get().charm is an str

However, the generated facade code accepts None|str|bytes:

python-libjuju/juju/client/_definitions.py

Lines 1967 to 1968 in a58645e

	if charm_ is not None and not isinstance(charm_, (bytes, str)):
	raise Exception(f"Expected charm_ to be a str, received: {type(charm_)}")

The type checker rightfully infers None|str|bytes for the .charm attribute.

Urgency

Casually reporting

Python-libjuju version

3.6.1.0

Juju version

any

Reproduce / Test

charm_name: str = (await Application.Get()).charm

[dimaqq]

P.S. I imagine that the codegen path is generic and may be used for other things that are:

• nullable (because, like, everything is in Juju or something?)
• not representable as a string (?)

Thus it may not be a trivial change.

[dimaqq]

Likewise, things are expected to be arrays of objects are also allowed to be:

• None (because everything is nullable?)
• str (eh? why?)
• bytes (because any str may also be bytes or what?)

For example:

python-libjuju/juju/client/_definitions.py

Lines 2132 to 2144 in a58645e

	class ApplicationMetricCredentials(Type):
	_toSchema = {"creds": "creds"}
	_toPy = {"creds": "creds"}
	def __init__(self, creds=None, **unknown_fields):
	"""Creds : typing.Sequence[~ApplicationMetricCredential]"""
	creds_ = [ApplicationMetricCredential.from_json(o) for o in creds or []]
	# Validate arguments against known Juju API types.
	if creds_ is not None and not isinstance(creds_, (bytes, str, list)):
	raise Exception(
	f"Expected creds_ to be a Sequence, received: {type(creds_)}"
	)

Schema:

https://github.com/dimaqq/juju-schema-analysis/blob/0acd07573bb2c192785796381f0594470ef296d8/schemas-juju-3.6.0.model-user.txt#L968-L973

P.S. would this actually be valid for arguments (and not results), is it OK to pass formatted JSON as str or bytes?

The same code appears to generate this for results as well:

Schema: https://github.com/dimaqq/juju-schema-analysis/blob/0acd07573bb2c192785796381f0594470ef296d8/schemas-juju-3.6.0.model-user.txt#L1094-L1098

Code:

python-libjuju/juju/client/_definitions.py

Lines 3989 to 3998 in a58645e

	class BlockResults(Type):
	_toSchema = {"results": "results"}
	_toPy = {"results": "results"}
	def __init__(self, results=None, **unknown_fields):
	"""Results : typing.Sequence[~BlockResult]"""
	results_ = [BlockResult.from_json(o) for o in results or []]
	# Validate arguments against known Juju API types.
	if results_ is not None and not isinstance(results_, (bytes, str, list)):

Here, the extra types don't make sense.
It could be the artefact of same codegen being used for arguments and results, maybe?