What happened?
The Python models that crossplane dependency update-cache (and crossplane project build) generate have many optional fields that type checkers report as required.
datamodel-code-generator writes the default of a field as a positional argument of Field when the field also has other arguments:
continue_: str | None = Field(None, alias='continue')
metadata: v1.ObjectMeta | None = Field({}, validate_default=True)
Pyright (and so Pylance, the default Python language server of VS Code) reads a default only from the default= keyword. It reports each of these fields as a required argument:
from models.io.k8s.api.apps.v1 import Deployment, DeploymentSpec
from models.io.k8s.api.core.v1 import PodTemplateSpec
from models.io.k8s.api.networking.v1 import IPBlock, NetworkPolicyIngressRule
from models.io.k8s.apimachinery.pkg.apis.core.meta.v1 import LabelSelector, ListMeta
Deployment()
DeploymentSpec(selector=LabelSelector(), template=PodTemplateSpec())
IPBlock(cidr="10.0.0.0/8")
NetworkPolicyIngressRule()
ListMeta()
t.py:6:1 - error: Arguments missing for parameters "metadata", "spec", "status" (reportCallIssue)
t.py:7:1 - error: Argument missing for parameter "strategy" (reportCallIssue)
t.py:7:51 - error: Arguments missing for parameters "metadata", "spec" (reportCallIssue)
t.py:8:1 - error: Argument missing for parameter "except" (reportCallIssue)
t.py:9:1 - error: Argument missing for parameter "from" (reportCallIssue)
t.py:10:1 - error: Argument missing for parameter "continue" (reportCallIssue)
6 errors, 0 warnings, 0 informations
Pydantic itself accepts these calls, because the defaults are correct at runtime. Only static type checks fail. In the Kubernetes v1.36.4 core API models, 146 fields use Field({}, ...), plus the fields renamed from Python keywords (continue_, except_, from_, and int_/bool_ since #63).
Expected: no errors. The fields have a default, so they are optional.
How can we reproduce it?
cat > crossplane-project.yaml <<'EOF'
apiVersion: dev.crossplane.io/v1alpha1
kind: Project
metadata:
name: repro
spec:
repository: example.com/repro
schemas:
languages: [python]
dependencies:
- type: k8s
k8s:
version: v1.36.4
EOF
crossplane dependency update-cache
grep -rn "Field(None, alias=\|Field({}" schemas/python/models | head
Then run pyright on the snippet above with schemas/python on the Python path (for example pyrightconfig.json with "extraPaths": ["schemas/python"]).
What environment did it happen in?
- Crossplane CLI version: v2.5.0
- Platform (e.g., linux/amd64): linux/amd64
- Crossplane version (if applicable): n/a
- datamodel-code-generator: 0.59.0 (pinned by the CLI), pyright 1.1.406, pydantic 2
What happened?
The Python models that
crossplane dependency update-cache(andcrossplane project build) generate have many optional fields that type checkers report as required.datamodel-code-generatorwrites the default of a field as a positional argument ofFieldwhen the field also has other arguments:Pyright (and so Pylance, the default Python language server of VS Code) reads a default only from the
default=keyword. It reports each of these fields as a required argument:Pydantic itself accepts these calls, because the defaults are correct at runtime. Only static type checks fail. In the Kubernetes v1.36.4 core API models, 146 fields use
Field({}, ...), plus the fields renamed from Python keywords (continue_,except_,from_, andint_/bool_since #63).Expected: no errors. The fields have a default, so they are optional.
How can we reproduce it?
Then run pyright on the snippet above with
schemas/pythonon the Python path (for examplepyrightconfig.jsonwith"extraPaths": ["schemas/python"]).What environment did it happen in?