`pydantic` aliases: `snake_case` or `kebab-case`?

[disinvite]

In our pydantic models for importing and exporting the reccmp-*.yml files, we use AliasChoice to allow multiple options for variable name. We mainly use this to allow both snake_case and kebab-case keys in the YML.

pydantic describes this feature as:

AliasChoices is used to specify a list of choices of aliases. Choices that appear first in the list will have higher priority during validation.

Observe that in this example, the key test-field is used regardless of order.

import ruamel.yaml
from pydantic import AliasChoices, BaseModel, Field

_yaml = ruamel.yaml.YAML()

text = """\
test-field: 123
test_field: 555
"""

reversed = """\
test_field: 555
test-field: 123
"""

class Sample(BaseModel):
    test_field: int = Field(
        validation_alias=AliasChoices("test-field", "test_field"),
    )

x = Sample.model_validate(_yaml.load(text))
y = Sample.model_validate(_yaml.load(reversed))

assert x.test_field == 123
assert y.test_field == 123

Do we want this behavior? It is at least consistent and predictable, but I would've expected a duplicate alias to raise an exception.

YAML forbids duplicate keys. JSON cautions against using them because the behavior is undefined and implementation-specific.

[madebr]

What is the reasoning behind allowing both snake and kebab casing?
When writing configuration files, I expect a single choice (e.g. GItHub workflows only have kebab casing).

[jonschz]

That is a valid question. I think the reasoning was that kebab case does not map well to Python, but is the more common format in yaml. If we want to reduce this down to one, I'd go for snake case.

[madebr]

Personally, I prefer the kebab style :)

It looks like projects use a combination of both: gta2_re, isle and dethrace.

reccmp-project creates projects using snake case.