Fix the wrong validator of ModulemdUnit.profiles [RHELDST-22116]

[lebouillon]

Currently, the wrong validator resulted in a TypeError, saying pubtools-pulplib is validating the after-conversion "profiles" field as a dict:

TypeError: "'profiles' must be <class 'dict'>
(got frozendict.frozendict({'test': ['secondRepoRPM']}) that is a <class 'frozendict.frozendict'>)."

The validator of ModulemdUnit.profiles should be a frozendict rather than a dict after the "converter=frozenlist_or_none_converter" conversion.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ddac319) 100.00% compared to head (4667b0c) 100.00%.

Additional details and impacted files

@@            Coverage Diff            @@
##            master      #217   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           47        47           
  Lines         3162      3162           
=========================================
  Hits          3162      3162           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rohanpm]

There seems to be a deeper story behind this issue. I find it hard to believe that the bug could be as described because it hasn't changed recently and how could it possibly not have broken anything until now?

Furthermore, these validators are documented as using "isinstance", so when we say optional_dict we're not saying the value has to be precisely a dict. It's allowed to be any subclass of dict, and I would expect frozendict to be a subclass of dict, so it should have been working fine.

Could you check the following in whatever environment you could reproduce this? For me, frozendict is an instance of dict so there is no problem with the validator.

>>> from frozendict.core import frozendict
>>> isinstance({}, dict)
True
>>> isinstance(frozendict({}), dict)
True

[rohanpm]

...and I do have a vague recollection that frozendict has both a C and a pure-python implementation, and if you end up using the C implementation it's not a subclass of dict, while if you use the pure-python implementation it is.

[lebouillon]

Could you check the following in whatever environment you could reproduce this? For me, frozendict is an instance of dict so there is no problem with the validator.

@rohanpm Sure. And it approved your assumption that environment is the cause.

I am running on Python 3.9.17 with frozendict 2.3.10
>>> from frozendict.core import frozendict
>>> isinstance({}, dict)
True
>>> isinstance(frozendict({}), dict)
False

But it seems both cannot allow the subclass to be True.

Here is a isinstance(frozendict({}), dict) check for the C Extension.

(python39) [dichen@arpeggio ~]$ CIBUILDWHEEL=1 pip install frozendict
Collecting frozendict
  Using cached frozendict-2.3.10-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (115 kB)
Installing collected packages: frozendict
Successfully installed frozendict-2.3.10

[notice] A new release of pip is available: 23.0.1 -> 23.3.1
[notice] To update, run: pip install --upgrade pip
(python39) [dichen@arpeggio ~]$ python
Python 3.9.17 (main, Jun  8 2023, 00:00:00) 
[GCC 13.1.1 20230511 (Red Hat 13.1.1-2)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from frozendict.core import frozendict
>>> isinstance(frozendict({}), dict)
False

Here is the isinstance(frozendict({}), dict) check for the pure py implementation.

(python39) [dichen@arpeggio ~]$ FROZENDICT_PURE_PY=1 pip install frozendict
Collecting frozendict
  Using cached frozendict-2.3.10-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (115 kB)
Installing collected packages: frozendict
Successfully installed frozendict-2.3.10

[notice] A new release of pip is available: 23.0.1 -> 23.3.1
[notice] To update, run: pip install --upgrade pip
(python39) [dichen@arpeggio ~]$ python
Python 3.9.17 (main, Jun  8 2023, 00:00:00) 
[GCC 13.1.1 20230511 (Red Hat 13.1.1-2)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from frozendict.core import frozendict
>>> isinstance(frozendict({}), dict)
False

I will keep doing some investigation for frozendict subclass issue.
PS: This same frozedict issue caused a hinderance for release-engineering/ubi-manifest#68 (comment)

[rohanpm]

I suppose we should still submit this despite that an issue outside of this library triggers the problem. It should work fine as long as we ensure that the converter and the validator are using the same type (i.e. either both using the pure-python frozendict or both using the C extension frozendict).