conda-forge.yml Pydantic Model: Single Values instead of Lists

[ytausch]

(Moved from #1865)

The conda-forge.yml schema contains fields like nooarch_platforms, remote_ci_setup that expect a list but some feedstocks just supply a single value, which is not supported by the model currently. Even our own documentation currently specifies a value that is not supported by the model (see below - Optional[List] does not support this).

conda-smithy/conda_smithy/schema.py

Lines 710 to 719 in 85ddb5d

	noarch_platforms: Optional[List[Platforms]] = Field(
	default_factory=lambda: ["linux_64"],
	description=cleandoc(
	"""
	Platforms on which to build noarch packages. The preferred default is a
	single build on `linux_64`.
	```yaml
	noarch_platforms: linux_64
	```

How to deal with this? One option would be to add a custom type that has a BeforeValidator converting single values to a list with only one item. I did a similar thing for the Pydantic models for cf-scripts:

https://github.com/ytausch/cf-scripts/blob/ea9cec3a988141e54d625169de60215bf3d8ff41/conda_forge_tick/models/common.py#L47-L58

The alternative would be to strictly support lists only and change the documentation and affected feedstocks.

@jaimergp

[beckermr]

What is in the model doesn't actually matter. You need to check the smithy source code to see if single values are allowed. If they are, then the model is wrong. The source code came first and is the source of truth on what is allowed for each field.

[ytausch]

conda-smithy supports single values for noarch_platforms. See the following snippet:

conda-smithy/conda_smithy/configure_feedstock.py

Lines 2107 to 2109 in 44acb93

	config["noarch_platforms"] = conda_build.utils.ensure_list(
	config["noarch_platforms"]
	)

It's also valid for remote_ci_setup:

conda-smithy/conda_smithy/configure_feedstock.py

Line 586 in 44acb93

remote_ci_setup_ = conda_build.utils.ensure_list(remote_ci_setup)

So this is actually an error in the model, which can be fixed like I proposed above.

[beckermr]

Great and thanks for tracking that down! Let's fix the model!

[jaimergp]

We can't use pydantic prevalidators because that's not used at runtime (we use jsonschema for the validation instead). What we can do is to turn the list[something] types into union[something, list[something]].