SubConfig now catches failures and warnings

[smarie]

Fixes #2709

[oprypin]

Sorry for the delay

[smarie]

Great, no worries @oprypin !

[oprypin]

Oops, unfortunately this makes a test fail.

https://github.com/mkdocs/mkdocs/runs/5702600210?check_suite_focus=true#step:5:19

FAIL: test_extra_context (mkdocs.tests.build_tests.BuildTests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/runner/work/mkdocs/mkdocs/mkdocs/tests/build_tests.py", line 181, in test_extra_context
    cfg = load_config(extra={'a': 1})
  File "/home/runner/work/mkdocs/mkdocs/mkdocs/tests/base.py", line 40, in load_config
    assert(errors_warnings == ([], [])), errors_warnings
AssertionError: ([], [('extra', ('a', 'Unrecognised configuration name: a'))])

That makes me think that there are probably unforeseen negative implications from this change, so I'll have to revert it until further investigation.

[smarie]

Indeed, sorry. Sad that the CI was not executed on PR :( . Maybe something to consider in the future ?

[smarie]

Update : I was able to test outside of tox, directly in PyCharm.
This is what happens :

• the extra option in the default config schema is defined as an empty SubConfig() (without any options defined inside it)

  mkdocs/mkdocs/config/defaults.py

  Lines 111 to 116 in 17ff566

  	# extra is a mapping/dictionary of data that is passed to the template.
  	# This allows template authors to require extra configuration that not
  	# relevant to all themes and doesn't need to be explicitly supported by
  	# MkDocs itself. A good example here would be including the current
  	# project version.
  	('extra', config_options.SubConfig()),

• in the failing test , the configuration is created with extra={'a': 1},

  mkdocs/mkdocs/tests/build_tests.py

  Lines 180 to 183 in 17ff566

  	def test_extra_context(self):
  	cfg = load_config(extra={'a': 1})
  	context = build.get_context(mock.Mock(), mock.Mock(), cfg)
  	self.assertEqual(context['config']['extra']['a'], 1)

Therefore there was an error in the subconfig Unrecognised configuration name: a.
This error was until now, hidden : it was already there but was not caught nor shown to the user.

The fix is now doing its job correctly, showing the error to the user.

Now the decision to take with this test is "should the 'extra' config option in the default schema accept any kind of options inside it ?". If the answer is yes, we should probably define a constructor flag in SubConfig to either allow for unvalidated options or not. If the answer is no, in that case the test should be modified so that extra is passed an empty dict.

Let me know !

[oprypin]

@smarie If I understand correctly, the extra option is now widely used as a freeform container of undefined options that themes can access.
Example: https://squidfunk.github.io/mkdocs-material/setup/setting-up-versioning/#versioning
Whatever change you develop, make sure this still works without warnings.

[smarie]

@oprypin I opened #2807 that fixes the failing test.