feat: support JSON schema

README.md

@@ -30,3 +30,4 @@ Please refer to test files for detailed usage.
 - [PEP 484](https://peps.python.org/pep-0484/) type annotations
 - Native integration with [Sphinx](https://github.com/sphinx-doc/sphinx), [DP-GUI](https://github.com/deepmodeling/dpgui), and [Jupyter Notebook](https://jupyter.org/)
 - JSON encoder for `Argument` and `Variant` classes
+- Generate [JSON schema](https://json-schema.org/) from an `Argument`, which can be further integrated with JSON editors such as [Visual Studio Code](https://code.visualstudio.com/)

dargs/dargs.py

@@ -460,11 +460,15 @@ def _check_data(self, value: Any, path=None):
             )
 
     def _check_strict(self, value: dict, path=None):
-        allowed_keys = self.flatten_sub(value, path).keys()
+        allowed_keys = set(self.flatten_sub(value, path).keys())
         # curpath = [*path, self.name]
         if not len(allowed_keys):
             # no allowed keys defined, allow any keys
             return
+        # A special case to allow $schema in any dict to be compatible with vscode + json schema
+        # https://code.visualstudio.com/docs/languages/json#_mapping-in-the-json
+        # considering usually it's not a typo of users when they use $schema
+        allowed_keys.add("$schema")
         for name in value.keys():
             if name not in allowed_keys:
                 dym_message = did_you_mean(name, allowed_keys)

dargs/json_schema.py

@@ -0,0 +1,158 @@
+"""Generate JSON schema from a given dargs.Argument."""
+
+from __future__ import annotations
+
+from typing import Any
+
+from dargs.dargs import Argument, _Flags
+
+try:
+    from typing import get_origin
+except ImportError:
+    from typing_extensions import get_origin
+
+
+def generate_json_schema(argument: Argument, id: str = "") -> dict:
+    """Generate JSON schema from a given dargs.Argument.
+
+    Parameters
+    ----------
+    argument : Argument
+        The argument to generate JSON schema.
+    id : str, optional
+        The URL of the schema, by default "".
+
+    Returns
+    -------
+    dict
+        The JSON schema. Use :func:`json.dump` to save it to a file
+        or :func:`json.dumps` to get a string.
+
+    Examples
+    --------
+    Dump the JSON schema of DeePMD-kit to a file:
+
+    >>> from dargs.json_schema import generate_json_schema
+    >>> from deepmd.utils.argcheck import gen_args
+    >>> import json
+    >>> from dargs import Argument
+    >>> a = Argument("DeePMD-kit", dtype=dict, sub_fields=gen_args())
+    >>> schema = generate_json_schema(a)
+    >>> with open("deepmd.json", "w") as f:
+    ...     json.dump(schema, f, indent=2)
+    """
+    schema = {
+        "$schema": "https://json-schema.org/draft/2020-12/schema",
+        "$id": id,
+        "title": argument.name,
+        **_convert_single_argument(argument),
+    }
+    return schema
+
+
+def _convert_single_argument(argument: Argument) -> dict:
+    """Convert a single argument to JSON schema.
+
+    Parameters
+    ----------
+    argument : Argument
+        The argument to convert.
+
+    Returns
+    -------
+    dict
+        The JSON schema of the argument.
+    """
+    data = {
+        "description": argument.doc,
+        "type": list({_convert_types(tt) for tt in argument.dtype}),
+    }
+    if argument.default is not _Flags.NONE:
+        data["default"] = argument.default
+    properties = {
+        **{
+            nn: _convert_single_argument(aa)
+            for aa in argument.sub_fields.values()
+            for nn in (aa.name, *aa.alias)
+        },
+        **{
+            vv.flag_name: {
+                "type": "string",
+                "enum": list(vv.choice_dict.keys()) + list(vv.choice_alias.keys()),
+                "default": vv.default_tag,
+                "description": vv.doc,
+            }
+            for vv in argument.sub_variants.values()
+        },
+    }
+    required = [
+        aa.name
+        for aa in argument.sub_fields.values()
+        if not aa.optional and not aa.alias
+    ] + [vv.flag_name for vv in argument.sub_variants.values() if not vv.optional]
+    allof = [
+        {
+            "if": {
+                "oneOf": [
+                    {
+                        "properties": {vv.flag_name: {"const": kkaa}},
+                    }
+                    for kkaa in (kk, *aa.alias)
+                ],
+                "required": [vv.flag_name]
+                if not (vv.optional and vv.default_tag == kk)
+                else [],
+            },
+            "then": _convert_single_argument(aa),
+        }
+        for vv in argument.sub_variants.values()
+        for kk, aa in vv.choice_dict.items()
+    ]
+    allof += [
+        {"oneOf": [{"required": [nn]} for nn in (aa.name, *aa.alias)]}
+        for aa in argument.sub_fields.values()
+        if not aa.optional and aa.alias
+    ]
+    if not argument.repeat:
+        data["properties"] = properties
+        data["required"] = required
+        if allof:
+            data["allOf"] = allof
+    else:
+        data["items"] = {
+            "type": "object",
+            "properties": properties,
+            "required": required,
+        }
+        if allof:
+            data["items"]["allOf"] = allof
+    return data
+
+
+def _convert_types(T: type | Any | None) -> str:
+    """Convert a type to JSON schema type.
+
+    Parameters
+    ----------
+    T : type | Any | None
+        The type to convert.
+
+    Returns
+    -------
+    str
+        The JSON schema type.
+    """
+    # string, number, integer, object, array, boolean, null
+    if T is None or T is type(None):
+        return "null"
+    elif T is str:
+        return "string"
+    elif T in (int, float):
+        return "number"
+    elif T is bool:
+        return "boolean"
+    elif T is list or get_origin(T) is list:
+        return "array"
+    elif T is dict or get_origin(T) is dict:
+        return "object"
+    raise ValueError(f"Unknown type: {T}")

docs/json_schema.md

@@ -0,0 +1,42 @@
+## Generate JSON schema from an argument
+
+One can use {func}`dargs.json_schema_generate_json_schema` to generate [JSON schema](https://json-schema.org/).
+
+```py
+import json
+
+from dargs import Argument
+from dargs.json_schema import generate_json_schema
+from deepmd.utils.argcheck import gen_args
+
+
+a = Argument("DeePMD-kit", dtype=dict, sub_fields=gen_args())
+schema = generate_json_schema(a)
+with open("deepmd.json", "w") as f:
+    json.dump(schema, f, indent=2)
+```
+
+JSON schema can be used in several JSON editors. For example, in [Visual Studio Code](https://code.visualstudio.com/), you can [configure JSON schema](https://code.visualstudio.com/docs/languages/json#_json-schemas-and-settings) in the project `settings.json`:
+
+```json
+{
+    "json.schemas": [
+        {
+            "fileMatch": [
+                "/**/*.json"
+            ],
+            "url": "./deepmd.json"
+        }
+    ]
+}
+```
+
+VS Code also allows one to [specify the JSON schema in a JSON file](https://code.visualstudio.com/docs/languages/json#_mapping-in-the-json) with the `$schema` key.
+To be compatible, dargs will not throw an error for `$schema` in the strict mode even if `$schema` is not defined in the argument.
+
+```json
+{
+  "$schema": "./deepmd.json",
+  "model": {}
+}
+```

pyproject.toml

@@ -30,6 +30,7 @@ repository = "https://github.com/deepmodeling/dargs"
 [project.optional-dependencies]
 test = [
     "ipython",
+    "jsonschema",
 ]
 typecheck = [
     "basedpyright==1.12.2",

tests/dpmdargs.py

@@ -216,7 +216,7 @@ def descrpt_hybrid_args():
                     "type",
                     [
                         Argument("loc_frame", dict, descrpt_local_frame_args()),
-                        Argument("se_a", dict, descrpt_se_a_args()),
+                        Argument("se_e2_a", dict, descrpt_se_a_args(), alias=["se_a"]),
                         Argument("se_r", dict, descrpt_se_r_args()),
                         Argument(
                             "se_a_3be", dict, descrpt_se_a_3be_args(), alias=["se_at"]
@@ -764,8 +764,19 @@ def normalize(data):
     return data
 
 
+def gen_args() -> Argument:
+    ma = model_args()
+    lra = learning_rate_args()
+    la = loss_args()
+    ta = training_args()
+
+    base = Argument("base", dict, [ma, lra, la, ta])
+    return base
+
+
 example_json_str = """
 {
+    "$schema": "this should be ignored by dargs",
     "_comment": " model parameters",
     "model": {
         "type_map":	["O", "H"],

tests/test_json_schema.py

@@ -0,0 +1,30 @@
+from __future__ import annotations
+
+import json
+import unittest
+
+from jsonschema import validate
+
+from dargs.json_schema import _convert_types, generate_json_schema
+
+from .dpmdargs import example_json_str, gen_args
+
+
+class TestJsonSchema(unittest.TestCase):
+    def test_json_schema(self):
+        args = gen_args()
+        schema = generate_json_schema(args)
+        data = json.loads(example_json_str)
+        validate(data, schema)
+
+    def test_convert_types(self):
+        self.assertEqual(_convert_types(int), "number")
+        self.assertEqual(_convert_types(str), "string")
+        self.assertEqual(_convert_types(float), "number")
+        self.assertEqual(_convert_types(bool), "boolean")
+        self.assertEqual(_convert_types(None), "null")
+        self.assertEqual(_convert_types(type(None)), "null")
+        self.assertEqual(_convert_types(list), "array")
+        self.assertEqual(_convert_types(dict), "object")
+        with self.assertRaises(ValueError):
+            _convert_types(set)