Choice prefers Enum.name even for StrEnum where it should use the value
#2911
Comments
|
Could this be the culprit: https://github.com/pallets/click/pull/2796/files#diff-dcb534e6a7591b92836537d4655ddbd2f18e3b293c3420144c30a9ca08f65c4eR298 ? And if so, is this behaviour intentional? If not, I'd be happy to open a PR with an additional test for |
Choice in Click 8.2.0 stringifies enums by name, even when they're str subclassesChoice stringifies enums by name, even when they're str subclasses
|
I've replicated the example as a test and I guess the follow-up question is whether this default is desirable. I could see quite a few class SomeStrEnum(StrEnum)
FOO = "foo"
BAR = "bar"
class SomeOtherStrEnum(str, Enum)
BLA = "bla"
ETC = "etc"that are currently being used in We experienced a similar, but not exactly the same, issue for @dataclass
class CLIOptions
foo_preference: Annotated[
SomeStrEnum,
typer.Option( "-f", "--foo-preference"),
] = SomeStrEnum.FOObroke as EDIT: Looping in @tiangolo in case reports of scenarios above come in for Typer as well and the click connection isn't immediately clear |
|
|
Just guessing based on the default value and the way the function is written. Apologies, I should not have stated it so definitively. I am not a developer on |
|
@qdegraaf #2796 added support for A good solution would perhaps be to get the starting value in def normalize_choice(self, choice: ParamTypeValue, ctx: Context | None) -> str:
"""
Normalize a choice value, used to map a passed string to a choice.
Each choice must have a unique normalized value.
By default uses :meth:`Context.token_normalize_func` and if not case
sensitive, convert it to a casefolded value.
.. versionadded:: 8.2.0
"""
- normed_value = choice.name if isinstance(choice, enum.Enum) else str(choice)
+ match choice:
+ case str(choice):
+ normed_value = choice
+ case Enum(choice):
+ normed_value = choice.name
+ case _:
+ normed_value = str(choice)
if ctx is not None and ctx.token_normalize_func is not None:
normed_value = ctx.token_normalize_func(normed_value)
if not self.case_sensitive:
normed_value = normed_value.casefold()
return normed_value |
Choice stringifies enums by name, even when they're str subclassesChoice prefers Enum.name even for StrEnum where it should use the value
|
Thanks for clarifying! Because the tests used a In the case of a class that is both assert "Usage: cli [OPTIONS] {foo-value|bar-value|baz-value}\nTry 'cli --help' for help.\n\nError: Invalid value for '{foo-value|bar-value|baz-value}': 'foo' is not one of 'foo-value', 'bar-value', 'baz-value'.\n" == 'foo-value\n'Which makes sense of course, but could undermine the original intention. |
|
Working on this (PyCon 2025) |
|
I don't like the idea of using |
|
@davidism What about just using |
|
That will still result in unexpected inconsistencies in enums. I'm marking this as a documentation fix rather than something we'll change. If custom behavior is needed, then a custom type or callback can be written to do exactly what that application needs. |
Consider the following code:
With Click 8.1.8, running this script with
-C redworks as expected, but with Click 8.2.0, it instead fails with:This was observed with Python 3.13.3 on macOS 14.7.5.
The text was updated successfully, but these errors were encountered: