3.15 regression: Unset default namespace version suddenly raises 404

[peterthomassen]

• The intention of Fix NamespaceVersioning ignoring DEFAULT_VERSION on non-None namespaces #7278 seems to be to deal with situations where DEFAULT_VERSION is ignored. The description of the fix relates to non-None values of DEFAULT_VERSION.
• When DEFAULT_VERSION is None and ALLOWED_VERSIONS is non-empty, the new code now raises 404.
• In 3.14, None would be returned in this case. This suddenly raises 404 where previously the view was properly run.

This is a regression that significantly changes the API. The change should be reverted.

[auvipy]

#9335

[Kostia-K]

It's possible I interpreted the documentation wrong, but here it says

ALLOWED_VERSIONS. If set, this value will restrict the set of versions that may be returned by the versioning scheme, and will raise an error if the provided version is not in this set. Note that the value used for the DEFAULT_VERSION setting is always considered to be part of the ALLOWED_VERSIONS set (unless it is None). Defaults to None.

The "unless it is None" part refers to DEFAULT_VERSION. Doesn't make sense for it to refer to ALLOWED_VERSIONS because if it's None, anything is allowed.

This is also consistent with how other versioning schemes behave, see TestAcceptHeaderAllowedAndDefaultVersion.test_missing_without_default and TestAcceptHeaderAllowedAndDefaultVersion.test_missing_without_default_but_none_allowed. I added equivalent tests in TestNamespaceAllowedAndDefaultVersion.

To fix the problem, you have to add None to ALLOWED_VERSIONS as shown in the unit tests.

I could be wrong but please review this again. You might be relying on incorrect, undocumented behavior. The API changes only make the behavior consistent with the documentation and with other versioning schemes.

[peterthomassen]

Fair enough, it's actually not a big deal to make that change. The issue is more about the unexpected breakage, instead of going through the usual deprecation stages.