domain.py

"""Implements the `json` domain and `json:schema` directive."""

import dataclasses
import json
import os
import re
from typing import (
    List,
    Any,
    Optional,
    Dict,
    cast,
    NamedTuple,
    Tuple,
    Iterator,
    Set,
    Union,
    Iterable,
)

import docutils
import docutils.parsers.rst.directives

try:
    import jsonschema.validators

    _jsonschema_validation_supported = True
except ImportError:
    _jsonschema_validation_supported = False

import sphinx
import sphinx.addnodes
import sphinx.application
import sphinx.directives
import sphinx.domains
import sphinx.environment
import sphinx.project
import sphinx.roles
import sphinx.util
import sphinx.util.docutils
import sphinx.util.logging
import sphinx.util.matching
import yaml  # pylint: disable=import-error

from .. import object_description_options
from . import json_pprint
from ... import sphinx_utils
from .. import apigen_utils

logger = sphinx.util.logging.getLogger(__name__)


class JsonSchemaMapEntry(NamedTuple):
    path: str
    pointer: str
    schema: Any
    id: str
    top_level_schema: Any


JsonSchema = Any
"""JSON schema represented as a JSON object (i.e. dict)."""

YamlSourceInfoMap = Dict[int, Tuple[str, int]]
"""For each literal string parsed from a schema YAML file, maps `id(s)` to the
`(source_file, lineno)`.

The lineno is 1-based.

When embedding strings from the YAML files as reStructuredText content in the
documentation, this is used to include proper source information to Sphinx,
which in turn can include it in error messages.
"""


@dataclasses.dataclass
class LoadedSchemaData:
    id_map: Dict[str, JsonSchemaMapEntry] = dataclasses.field(default_factory=dict)
    """Maps JSON schema ids to the loaded schema entry.

    This contains an entry for both top-level and sub-schemas, including schemas
    for individual properties.

    The same schema entry may be contained under multiple keys.  In addition to
    the canonical id (defined by the $id member), if any, the schema is also
    present under a JSON pointer key `<parent>#/json/pointer/path`, if applicable.

    Additionally, as schemas are rendered to a document tree, additional aliases
    are added for member property ids `<parent>.<member>` and for string constant
    oneOfs `<parent>.<value>`.
    """

    source_info_map: YamlSourceInfoMap = dataclasses.field(default_factory=dict)
    """Combined source map for all schemas loaded in `id_map`."""

    identity_map: Dict[int, JsonSchemaMapEntry] = dataclasses.field(
        default_factory=dict
    )
    """Maps `id(entry.schema)` -> `entry` for every entry in `id_map`.

    This allows the entry to be determined from arbitrary nested schemas.
    """

    supertype_map: Dict[str, List[str]] = dataclasses.field(default_factory=dict)
    """Maps the canonical schema id to a list of types it "extends" (via allOf)."""

    subtype_map: Dict[str, List[str]] = dataclasses.field(default_factory=dict)
    """Maps the canonical schema id to a list of types that "extend" it).

    This is the inverse of `supertype_map`.
    """

    def register_subtype_relationships(self, schema: JsonSchema) -> None:
        """Registers subtype relationships involving `schema`.

        These are stored in the `subtype_map` and `schema_supertype_map` maps.

        This relationship information is used to render the "Extends" and "Subtypes"
        sections for a JSON schema description.

        :param schema: Schema to process.

        """
        schema_id = schema["$id"]
        # Don't show types with ids containing "#"
        if "#" in schema_id:
            return
        allof = schema.get("allOf")
        if not allof:
            return
        for supertype in allof:
            supertype_id = supertype.get("$ref")
            if supertype_id is None:
                continue
            supertype_entry = self.id_map.get(supertype_id)
            if supertype_entry is None:
                # Error was already previously emitted.
                continue
            if "$id" not in supertype_entry.schema:
                continue
            # Don't show types with ids containing "#"
            if "#" in supertype_entry.id:
                continue
            supertype_id = supertype_entry.id
            self.subtype_map.setdefault(supertype_id, []).append(schema_id)
            self.supertype_map.setdefault(schema_id, []).append(supertype_id)

    def load_json_schema(
        self, path: str, full_path: str, seen_ids: Dict[str, str], validate: bool
    ) -> None:
        """Loads a single JSON schema into the global registry.

        All relative schema ids are resolved to absolute ids.

        The top-level schema as well as any sub-schemas are included in the registry.

        :param path: Path relative to documentation source dir.
        :param full_path: Full path to schema file.
        :param seen_ids: Dictionary of schema ids to populate.
        """
        with open(full_path, "r", encoding="utf-8") as f:
            top_level_schema, source_info_map = yaml_load(f, source_path=full_path)
        self.source_info_map.update(source_info_map)
        if validate:
            jsonschema.validators.validator_for(top_level_schema).check_schema(
                top_level_schema
            )
        top_level_id = top_level_schema.get("$id")
        _fix_jsonschema_ids(
            top_level_schema, top_level_id, path, seen_ids, self.source_info_map
        )

        # Register an individual schema/sub-schema.
        def _process_schema(schema: JsonSchema, pointer: str):
            pointer_based_id = top_level_id
            if pointer:
                pointer_based_id += "#" + pointer
            canonical_id = pointer_based_id
            if "$id" in schema:
                canonical_id = schema["$id"] = _normalize_jsonschema_id(
                    schema["$id"], top_level_id
                )
            schema_entry = JsonSchemaMapEntry(
                path=path,
                pointer=pointer,
                schema=schema,
                id=canonical_id,
                top_level_schema=top_level_schema,
            )
            self.identity_map[id(schema)] = schema_entry
            self.id_map[canonical_id] = schema_entry
            if canonical_id != pointer_based_id:
                self.id_map[pointer_based_id] = schema_entry

        for sub_schema, pointer in _traverse_sub_schemas(top_level_schema):
            _process_schema(sub_schema, pointer)

    def get_node_source_info_from_schema_string(
        self, source_string: str, default_filename: str
    ) -> Tuple[str, int]:
        filename, line = self.source_info_map.get(
            id(source_string), (default_filename, 0)
        )
        return (filename, line - 1)

    def get_location_from_schema_string(
        self, source_string: str, default_filename: str
    ) -> str:
        filename, line = self.source_info_map.get(
            id(source_string), (default_filename, 0)
        )
        if line > 0:
            return "%s:%d" % (filename, line)
        # Always include ":" in location even if line number is unknown, as
        # otherwise Sphinx interprets it as a docname.
        return "%s:" % (filename,)


def yaml_load(  # pylint: disable=invalid-name
    stream,
    source_path: str,
) -> Tuple[Any, YamlSourceInfoMap]:
    """Loads a yaml file, preserving object key order and source line information.

    :param stream: File-like stream to read from.
    :param source_path: Path to source file for inclusion in source info map.
    :param Loader: YAML loader class.

    :returns: Tuple of loaded YAML value and source info map.
    """
    source_info_map: YamlSourceInfoMap = {}

    class Loader(yaml.SafeLoader):
        def compose_scalar_node(self, anchor):
            line = self.line
            node = super().compose_scalar_node(anchor)
            if isinstance(node.value, str):
                source_info_map[id(node.value)] = (source_path, line)
            return node

    result = yaml.load(stream, Loader)
    return result, source_info_map


def _globs_to_re(globs: List[str]) -> re.Pattern:
    return re.compile(
        "|".join(
            "(?:" + sphinx.util.matching._translate_pattern(x) + ")" for x in globs
        )
    )


def _get_json_schema_files(app: sphinx.application.Sphinx):
    """Finds all schema files.

    :param app: Sphinx application object.

    :Yields: Each schema path relative to ``app.srcdir``.
    """
    include_globs = app.config.json_schemas
    if not include_globs:
        return

    exclude_globs = (
        app.config.exclude_patterns
        + app.config.templates_path
        + sphinx.project.EXCLUDE_PATHS
    )

    if sphinx.version_info >= (6, 0):
        yield from sphinx.util.matching.get_matching_files(
            app.srcdir, include_patterns=include_globs, exclude_patterns=exclude_globs
        )
        return

    # Use older predicate-based `get_matching_files`.
    include_re = _globs_to_re(include_globs)
    exclude_re = _globs_to_re(exclude_globs)
    matching_files = sphinx.util.get_matching_files(
        app.srcdir, (lambda s: exclude_re.fullmatch(s) is not None,)
    )
    for name in matching_files:
        if not include_re.fullmatch(name):
            continue
        yield name


def _populate_json_schema_id_map(app: sphinx.application.Sphinx):
    """Finds all schema files and loads them into `_json_schema_id_map`."""
    schema_data = LoadedSchemaData()
    setattr(app.env, "json_schema_data", schema_data)
    seen_ids: Dict[str, str] = {}
    all_paths = list(_get_json_schema_files(app))
    validate = app.config.json_schema_validate
    if all_paths and validate and not _jsonschema_validation_supported:
        logger.error("jsonschema package required for json_schema_validate option")
        validate = False
    for i, path in enumerate(all_paths):
        logger.info("[%d/%d] Loading JSON schema: %s", i + 1, len(all_paths), path)
        schema_data.load_json_schema(
            path=path,
            full_path=os.path.join(app.srcdir, path),
            seen_ids=seen_ids,
            validate=validate,
        )
    for key, path in seen_ids.items():
        if key in schema_data.id_map:
            continue
        logger.error(
            "Reference to undefined JSON schema: %r",
            key,
            location=schema_data.get_location_from_schema_string(key, path),
        )

    for schema_id, schema_entry in schema_data.id_map.items():
        if schema_entry.id != schema_id:
            # Don't process under an alias, in order to avoid processing the same type
            # more than once.
            continue
        schema = schema_entry.schema
        if "$id" not in schema:
            continue
        schema_data.register_subtype_relationships(schema)


def _is_object_with_properties(schema_node: JsonSchema) -> bool:
    """Returns `True` if `schema_node` is an object schema with properties.

    An `allOf` schema is also assumed to be an object schema.

    :param schema_node: JSON schema to check.
    """
    if not isinstance(schema_node, dict):
        return False
    return ("type" not in schema_node and "allOf" in schema_node) or (
        schema_node.get("type") == "object" and schema_node.get("properties", {})
    )


def _is_object_array_with_properties(schema_node: JsonSchema):
    return schema_node.get("type") == "array" and _is_object_with_properties(
        schema_node.get("items")
    )


def _is_object_or_object_array_with_properties(schema_node: JsonSchema):
    return _is_object_with_properties(schema_node) or _is_object_array_with_properties(
        schema_node
    )


def _normalize_jsonschema_id(id_str: str, top_level_id: Optional[str]) -> str:
    if id_str.startswith("#") and top_level_id is not None:
        return top_level_id + id_str
    return id_str


def _traverse_sub_schemas(
    schema: JsonSchema, pointer: str = ""
) -> Iterator[Tuple[JsonSchema, str]]:
    """Invokes `callback` on `schema` and any sub-schemas.

    :param schema: JSON schema.
    :param pointer: Initial JSON pointer.
    """
    yield (schema, pointer)

    def _handle_prop(prop):
        obj = schema.get(prop)
        if obj is None:
            return
        items: Iterable[Tuple[Union[int, str], JsonSchema]]
        if isinstance(obj, list):
            items = enumerate(obj)
        elif isinstance(obj, dict):
            items = obj.items()
        else:
            raise TypeError
        for key, sub_schema in items:
            yield from _traverse_sub_schemas(
                sub_schema, "%s/%s/%s" % (pointer, prop, key)
            )

    for prop in ("definitions", "properties", "allOf", "oneOf"):
        yield from _handle_prop(prop)
    items_data = schema.get("items")
    if items_data is not None:
        if isinstance(items_data, list):
            yield from _handle_prop("items")
        else:
            yield from _traverse_sub_schemas(items_data, "%s/items" % (pointer,))


def _fix_jsonschema_ids(
    schema: JsonSchema,
    top_level_id: Optional[str],
    path: str,
    seen_ids: Dict[str, str],
    source_info: YamlSourceInfoMap,
):
    """Makes relative `$id` and `$ref` members absolute.

    Also collects all ids that are seen in order to warn about unresolved
    references.

    :param schema: JSON schema, modified in place.
    :param top_level_id: Top-level id to use for resolving relative ids.
    :param path: Path to the file containing `schema`.
    :param seen_ids: Dictionary mapping seen ids to paths.
    :param source_info: Source info map to update.
    """

    def _fix_ids(sub_schema):
        for prop in ("$id", "$ref"):
            if prop in sub_schema:
                value = sub_schema[prop]
                normalized = _normalize_jsonschema_id(value, top_level_id)
                info = source_info.get(id(value))
                if info is not None:
                    source_info[id(normalized)] = info
                seen_ids.setdefault(normalized, path)
                sub_schema[prop] = normalized

    for sub_schema, pointer in _traverse_sub_schemas(schema):
        _fix_ids(sub_schema)


def _get_json_schema_node_id(fully_qualified_name: str) -> str:
    """Returns the reference id (i.e. HTML fragment id) for a schema."""
    return "json-%s" % (fully_qualified_name,)


def _schema_to_xref(schema_id: str) -> sphinx.addnodes.pending_xref:
    """Creates a pending_xref node that references a json schema.

    The result is similar to that of the json:schema role but just a bare
    pending_xref is constructed, without a code element.

    :param schema_id: Schema id.

    :returns: The pending_xref node.
    """
    return sphinx.addnodes.pending_xref(
        "",
        docutils.nodes.Text(schema_id),
        refdomain="json",
        reftype="schema",
        reftarget=schema_id,
    )


def _json_literal(self, j: Any) -> docutils.nodes.Node:
    return docutils.nodes.literal(
        text=json.dumps(j),
        classes=["code", "json"],
        language="json",
    )


class JsonSchemaDirective(sphinx.directives.ObjectDescription):
    """json:schema directive."""

    required_arguments = 1
    has_content = False

    objtype = "schema"

    option_spec = {
        "fully_qualified_name": str,
        "title": str,
        "toc_title": str,
        "exclude_from_toc": docutils.parsers.rst.directives.flag,
        "nested": docutils.parsers.rst.directives.flag,
        "noindex": docutils.parsers.rst.directives.flag,
    }

    _rendered_title: Optional[docutils.nodes.inline]

    def _inline_text(self, text: str) -> List[docutils.nodes.Node]:
        nodes, messages = self.state.inline_text(text, self.lineno)
        return nodes + cast(List[docutils.nodes.Node], messages)

    def _json_sig_type(self, name: str):
        return sphinx.addnodes.desc_type(name, name)

    def _get_type_description_line_ref(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        return [
            sphinx.addnodes.desc_type("", "", _schema_to_xref(schema_node.get("$ref")))
        ]

    def _get_type_description_line_oneof(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        result: List[docutils.nodes.Node] = []
        for x in schema_node.get("oneOf"):
            if result:
                result.append(sphinx.addnodes.desc_sig_punctuation("", " | "))
            part = self._get_type_description_line(x)
            if part is None:
                return None
            result += part
        return result

    def _get_type_description_line_const(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        return [_json_literal(self.state, schema_node["const"])]

    def _get_type_description_line_enum(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        result: List[docutils.nodes.Node] = []
        for x in schema_node.get("enum"):
            if result:
                result.append(sphinx.addnodes.desc_sig_punctuation("", " | "))
            result.append(_json_literal(self.state, x))
        return result

    def _get_type_description_line_number(
        self, schema_node: JsonSchema, t: str
    ) -> Optional[List[docutils.nodes.Node]]:
        explicit_lower = True
        if schema_node.get("minimum") is not None:
            lower_punct = "["
            lower_node = _json_literal(self.state, schema_node.get("minimum"))
        elif schema_node.get("exclusiveMinimum") is not None:
            lower_punct = "("
            lower_node = _json_literal(self.state, schema_node.get("exclusiveMinimum"))
        else:
            lower_punct = "("
            lower_node = sphinx.addnodes.desc_sig_operator("", "-∞")
            explicit_lower = False

        explicit_upper = True
        if schema_node.get("maximum") is not None:
            upper_punct = "]"
            upper_node = _json_literal(self.state, schema_node.get("maximum"))
        elif schema_node.get("exclusiveMaximum") is not None:
            upper_punct = ")"
            upper_node = _json_literal(self.state, schema_node.get("exclusiveMaximum"))
        else:
            upper_punct = ")"
            upper_node = sphinx.addnodes.desc_sig_operator("", "+∞")
            explicit_upper = False

        result = [self._json_sig_type(t)]
        if explicit_lower or explicit_upper:
            result.append(
                docutils.nodes.subscript(
                    "",
                    "",
                    sphinx.addnodes.desc_sig_punctuation("", lower_punct),
                    lower_node,
                    sphinx.addnodes.desc_sig_punctuation("", ", "),
                    upper_node,
                    sphinx.addnodes.desc_sig_punctuation("", upper_punct),
                )
            )
        return result

    def _get_type_description_line_boolean(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        return [self._json_sig_type("boolean")]

    def _get_type_description_line_string(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        result = [self._json_sig_type("string")]
        subscript_parts: List[docutils.nodes.Node] = []
        if (
            schema_node.get("minLength") is not None
            or schema_node.get("maxLength") is not None
        ):
            subscript_parts.append(sphinx.addnodes.desc_sig_punctuation("", "["))
            min_length = schema_node.get("minLength")
            if min_length:
                subscript_parts.append(_json_literal(self.state, min_length))
            subscript_parts.append(sphinx.addnodes.desc_sig_punctuation("", ".."))
            max_length = schema_node.get("maxLength")
            if max_length:
                subscript_parts.append(_json_literal(self.state, max_length))
            subscript_parts.append(sphinx.addnodes.desc_sig_punctuation("", "]"))
            result.append(docutils.nodes.subscript("", "", *subscript_parts))
        return result

    def _get_type_description_line_array(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        items = schema_node.get("items")
        prefix = [self._json_sig_type("array")]
        if "minItems" in schema_node or "maxItems" in schema_node:
            prefix.append(sphinx.addnodes.desc_sig_punctuation("", "["))
            if schema_node.get("minItems") == schema_node.get("maxItems"):
                prefix.append(_json_literal(self.state, schema_node["minItems"]))
            else:
                if schema_node["minItems"]:
                    prefix.append(_json_literal(self.state, schema_node["minItems"]))
                prefix.append(sphinx.addnodes.desc_sig_punctuation("", ".."))
                if schema_node["maxItems"]:
                    prefix.append(_json_literal(self.state, schema_node["maxItems"]))
            prefix.append(sphinx.addnodes.desc_sig_punctuation("", "]"))

        if "items" in schema_node and isinstance(items, dict):
            items_desc = self._get_type_description_line(items)
            if items_desc is None:
                return None
            return prefix + [docutils.nodes.emphasis("", " of ")] + items_desc
        if "items" in schema_node and isinstance(items, list):
            result: List[docutils.nodes.Node] = []
            result.append(sphinx.addnodes.desc_sig_punctuation("", "["))
            for i, item in enumerate(items):
                if i != 0:
                    result.append(sphinx.addnodes.desc_sig_punctuation("", ","))
                    result.append(docutils.nodes.Text(" "))
                item_desc = self._get_type_description_line(item)
                if item_desc is None:
                    return prefix
                result.extend(item_desc)
            result.append(sphinx.addnodes.desc_sig_punctuation("", "]"))
            return result
        return prefix

    def _get_type_description_line_null(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        return [_json_literal(self.state, None)]

    def _get_type_description_line_object(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        return [self._json_sig_type("object")]

    def _get_type_description_line(
        self, schema_node: JsonSchema
    ) -> Optional[List[docutils.nodes.Node]]:
        """Renders a short type description for use in the signature."""
        if "$ref" in schema_node:
            return self._get_type_description_line_ref(schema_node)
        if "oneOf" in schema_node:
            return self._get_type_description_line_oneof(schema_node)
        if "const" in schema_node:
            return self._get_type_description_line_const(schema_node)
        if "enum" in schema_node:
            return self._get_type_description_line_enum(schema_node)
        t = schema_node.get("type")
        if "allOf" in schema_node and t is None:
            t = "object"
        if t in ("integer", "number"):
            return self._get_type_description_line_number(schema_node, t)
        if t == "boolean":
            return self._get_type_description_line_boolean(schema_node)
        if t == "string":
            return self._get_type_description_line_string(schema_node)
        if t == "array":
            return self._get_type_description_line_array(schema_node)
        if t == "null":
            return self._get_type_description_line_null(schema_node)
        if t == "object":
            return self._get_type_description_line_object(schema_node)
        return None

    def _collect_object_properties(
        self,
        schema_node: JsonSchema,
        properties: Dict[str, JsonSchema],
        required: Set[str],
    ) -> None:
        """Collects the object properties for a schema.

        :param schema_node: JSON schema to process.
        :param properties: Dict to be filled with members.
        :param required: Set to which required members are added.
        """
        schema_data: LoadedSchemaData = getattr(self.env, "json_schema_data")
        if "$ref" in schema_node:
            ref = schema_node["$ref"]
            referenced_entry = schema_data.id_map.get(ref)
            if referenced_entry is None:
                # Error was already emitted previously
                return
            schema_node = referenced_entry.schema
        if schema_node.get("type") == "object":
            properties.update(schema_node.get("properties", {}))
            required.update(schema_node.get("required", []))
        elif _is_object_array_with_properties(schema_node):
            self._collect_object_properties(schema_node["items"], properties, required)
        else:
            for sub_node in schema_node.get("allOf", []):
                self._collect_object_properties(sub_node, properties, required)

    def _add_sub_schema(
        self, entry: JsonSchemaMapEntry, **kwargs: Union[str, bool, None]
    ) -> List[docutils.nodes.Node]:
        """Renders a nested schema.

        :param entry: Entry for the nested schema.
        :param kwargs: Options for the json:schema directive.
        :returns: The rendered result as a list of docutils nodes.
        """
        return sphinx_utils.parse_rst(
            self.state,
            sphinx_utils.format_directive("json:schema", entry.id, options=kwargs),
            *self.get_source_info(),
        )

    def _make_field(
        self, label: str
    ) -> Tuple[docutils.nodes.field_list, docutils.nodes.field_body]:
        field_list = docutils.nodes.field_list()
        field = docutils.nodes.field()
        field_list += field
        field_name = docutils.nodes.field_name(label, label)
        field += field_name
        if not self._noindex and self._objdesc_options["include_fields_in_toc"]:
            field_name["ids"].append(
                docutils.nodes.make_id(label) + "-" + cast(str, self._node_id)
            )
            field_name["toc_title"] = label
        body = docutils.nodes.field_body(classes=["noindent"])
        field += body
        return field_list, body

    def _render_oneofs(self) -> List[docutils.nodes.Node]:
        """Renders a long-form list of oneof sub-schemas.

        :returns: The rendered result as a list of docutils nodes.
        """
        schema_data: LoadedSchemaData = getattr(self.env, "json_schema_data")
        field_list, body = self._make_field("One of")
        one_ofs = self._schema_entry.schema["oneOf"]
        # If all oneof options are constant strings, generate fully-qualified
        # ids.
        generate_oneof_ids = all(isinstance(x.get("const"), str) for x in one_ofs)
        for x in one_ofs:
            fully_qualified_name = None
            title = None
            if generate_oneof_ids:
                title = x["const"]
                fully_qualified_name = "%s.%s" % (self._fully_qualified_name, title)
            body.extend(
                self._add_sub_schema(
                    schema_data.identity_map[id(x)],
                    nested=True,
                    fully_qualified_name=fully_qualified_name,
                    exclude_from_toc=self._exclude_from_toc or not generate_oneof_ids,
                    noindex=self._noindex,
                    toc_title=title,
                )
            )
        return [field_list]

    def _get_node_source_info_from_schema_string(
        self, source_string: str
    ) -> Tuple[str, int]:
        """Obtains the source information for a schema string.

        If no source information is found, just returns the path of the current
        schema file without a valid line number.

        :param source_string: String literal obtained from a JSON schema.
        :returns: A tuple (path, line) specifying the source information.
        """

        schema_data: LoadedSchemaData = getattr(self.env, "json_schema_data")
        return schema_data.get_node_source_info_from_schema_string(
            source_string, self._schema_entry.path
        )

    def _parse_rst_from_schema_string(
        self, source_string: str
    ) -> List[docutils.nodes.Node]:
        """Renders RST markup obtained from a schema.

        :param source_string: String literal obtained from a JSON schema.

        :returns: The rendered result as a list of docutils nodes.
        """
        rst_input = docutils.statemachine.StringList()
        sphinx_utils.append_multiline_string_to_stringlist(
            rst_input,
            ".. highlight-push::\n\n" + self.env.config.json_schema_rst_prolog + "\n\n",
            "<json_schema_rst_prolog>",
            -2,
        )
        filename, offset = self._get_node_source_info_from_schema_string(source_string)
        sphinx_utils.append_multiline_string_to_stringlist(
            rst_input,
            source_string,
            filename,
            offset,
        )
        sphinx_utils.append_multiline_string_to_stringlist(
            rst_input,
            "\n\n"
            + self.env.config.json_schema_rst_epilog
            + "\n\n.. highlight-pop::\n\n",
            "<json_schema_rst_prolog>",
            -2,
        )
        with apigen_utils.save_rst_defaults(self.env):
            return sphinx_utils.parse_rst(
                self.state,
                rst_input,
                filename,
                offset,
            )

    def _render_body(self) -> List[docutils.nodes.Node]:
        """Renders the body of the schema."""
        schema_data: LoadedSchemaData = getattr(self.env, "json_schema_data")
        schema_node = self._schema_entry.schema
        result: List[docutils.nodes.Node] = []

        if self._rendered_title is not None:
            p = self._rendered_title
            p += docutils.nodes.Text("  ")
            result.append(p)
        if "description" in schema_node:
            result += self._parse_rst_from_schema_string(schema_node.get("description"))

        def _render_related_type_list(related: Optional[List[str]], label: str) -> None:
            if not related:
                return
            field_list, body = self._make_field(label)
            result.append(field_list)
            bullet_list = docutils.nodes.bullet_list()
            body += bullet_list
            for schema_id in related:
                related_schema = schema_data.id_map[schema_id].schema
                list_item = docutils.nodes.list_item()
                bullet_list += list_item
                p = docutils.nodes.paragraph()
                list_item += p
                p += self._inline_text(f":json:schema:`{schema_id}`")
                related_title = related_schema.get("title")
                if related_title:
                    p += self._inline_text(f" — {related_title}")

        _render_related_type_list(
            schema_data.supertype_map.get(self._schema_entry.id), "Extends"
        )
        subtypes = schema_data.subtype_map.get(self._schema_entry.id)
        if subtypes:
            subtypes = sorted(subtypes)
        _render_related_type_list(subtypes, "Subtypes")

        if "oneOf" in schema_node and any(
            ("title" in x or "description" in x) for x in schema_node["oneOf"]
        ):
            result.extend(self._render_oneofs())
        if _is_object_or_object_array_with_properties(schema_node):
            properties: Dict[str, JsonSchema] = {}
            required_properties: Set[str] = set()
            self._collect_object_properties(
                schema_node, properties, required_properties
            )
            for required in [True, False]:
                if not any(
                    (member_name in required_properties) == required
                    for member_name in properties
                ):
                    continue
                heading = "%s members" % ("Required" if required else "Optional")
                field_list, body = self._make_field(heading)
                result.append(field_list)
                for member_name, member_schema in properties.items():
                    if (member_name in required_properties) != required:
                        continue
                    body.extend(
                        self._add_sub_schema(
                            schema_data.identity_map[id(member_schema)],
                            title=member_name,
                            fully_qualified_name="%s.%s"
                            % (self._fully_qualified_name, member_name),
                            nested=True,
                            exclude_from_toc=self._exclude_from_toc,
                            noindex=self._noindex,
                        )
                    )

        def add_example(value, caption, admonition_class="example"):
            nonlocal result
            result += sphinx_utils.parse_rst(
                self.state,
                sphinx_utils.format_directive(
                    "admonition",
                    caption,
                    options={"class": admonition_class},
                    content=sphinx_utils.format_directive(
                        "code-block",
                        "json",
                        content=json_pprint.pformat(value, indent=2),
                    ),
                ),
                *self._get_node_source_info_from_schema_string(value),
            )

        if self._long_default_value:
            add_example(schema_node["default"], "Default", "note")

        if "examples" in schema_node:
            for example in schema_node["examples"]:
                add_example(example, "Example")
        return result

    def before_content(self):
        # Set ref_context information which will be used to resolve cross references
        # in the body.
        self.env.ref_context["json:schema"] = self._fully_qualified_name
        self.env.ref_context.setdefault("json:schemas", []).append(  # type: ignore
            self._fully_qualified_name
        )

    def after_content(self):
        # Clear the ref_context information set by `.before_context`.
        schemas = self.env.ref_context["json:schemas"]
        schemas.pop()
        self.env.ref_context["json:schema"] = schemas[-1] if schemas else None

    def transform_content(self, contentnode: sphinx.addnodes.desc_content) -> None:
        contentnode += self._render_body()

    def handle_signature(
        self, sig: str, signode: sphinx.addnodes.desc_signature
    ) -> Any:
        # Renders the header for the JSON schema.
        schema_entry = self._schema_entry
        if not self._nested:
            signode += sphinx.addnodes.desc_annotation("json ", "json ")
        type_line = self._type_line
        title = self._title
        if title is not None:
            signode += sphinx.addnodes.desc_name(title, title)
            if type_line:
                signode += sphinx.addnodes.desc_sig_punctuation("", " : ")
        signode += type_line
        if self._short_default_value:
            signode += sphinx.addnodes.desc_sig_punctuation("", " = ")
            signode += docutils.nodes.literal(
                text=self._short_default_value,
                classes=["code", "json"],
                language="json",
            )
        if not self._exclude_from_toc and self._toc_title:
            signode["toc_title"] = self._toc_title
        return schema_entry.id

    def add_target_and_index(
        self, name: Any, sig: str, signode: sphinx.addnodes.desc_signature
    ) -> None:
        del name
        if self._noindex:
            return
        signode["ids"].append(self._node_id)

    def run(self) -> List[docutils.nodes.Node]:
        schema_data: LoadedSchemaData = self.env.json_schema_data  # type: ignore
        schema_id = self.arguments[0]
        schema_entry = schema_data.id_map.get(schema_id)
        if schema_entry is None:
            logger.error(
                "Undefined JSON schema: %r", schema_id, location=self.get_source_info()
            )
            return []
        self._schema_entry = schema_entry

        self._fully_qualified_name = self.options.get("fully_qualified_name")
        self._noindex = "noindex" in self.options
        self._exclude_from_toc = "exclude_from_toc" in self.options
        self._nested = "nested" in self.options
        self._objtype = "schema" if not self._nested else "subschema"
        self._objdesc_options = (
            object_description_options.get_object_description_options(
                self.env, "json", self._objtype
            )
        )
        if self._fully_qualified_name:
            schema_data.id_map.setdefault(
                self._fully_qualified_name, self._schema_entry
            )
        else:
            self._fully_qualified_name = self._schema_entry.id

        if "#/" in self._fully_qualified_name:
            self._noindex = True

        if self._noindex:
            self._exclude_from_toc = True

        self._node_id = None
        if not self._noindex:
            self._node_id = _get_json_schema_node_id(self._fully_qualified_name)

        schema_node = self._schema_entry.schema
        type_line = self._get_type_description_line(schema_node)

        # Determine how to display the default value.  If it is short enough, it is
        # displayed inline after the type.  Otherwise, it is displayed in a separate
        # block after the description.
        self._long_default_value = None
        self._short_default_value = None
        has_default_value = "default" in schema_node
        if has_default_value:
            formatted_default = json_pprint.pformat(
                schema_node["default"], indent=2
            ).strip()
            if "\n" not in formatted_default and len(formatted_default) < 40:
                self._short_default_value = formatted_default
                has_default_value = False
            else:
                self._long_default_value = formatted_default
        self._type_line = type_line

        if "title" in self.options:
            self._title = self.options["title"]
        elif not self._nested:
            self._title = self._fully_qualified_name
        else:
            self._title = None

        self._toc_title = self.options.get("toc_title") or self._title

        schema = self._schema_entry.schema
        if "title" in schema:
            self._rendered_title = docutils.nodes.inline()
            self._rendered_title += self._parse_rst_from_schema_string(schema["title"])
        else:
            self._rendered_title = None

        if self._node_id:
            domain = cast(JsonSchemaDomain, self.env.get_domain("json"))
            generate_synopses = self._objdesc_options["generate_synopses"]
            synopsis: Optional[str]
            if generate_synopses is not None and self._rendered_title:
                synopsis = sphinx_utils.summarize_element_text(
                    self._rendered_title, generate_synopses
                )
            else:
                synopsis = None
            domain.schemas[self._fully_qualified_name] = DomainSchemaEntry(
                docname=self.env.docname,
                node_id=self._node_id,
                schema_id=self._schema_entry.id,
                synopsis=synopsis,
                objtype=self._objtype,
            )
        return super().run()


class JsonSchemaRole(sphinx.roles.XRefRole):
    """json:schema role."""

    def process_link(
        self,
        env: sphinx.environment.BuildEnvironment,
        refnode: docutils.nodes.Element,
        has_explicit_title: bool,
        title: str,
        target: str,
    ) -> Tuple[str, str]:
        # Based on PyXRefRole in sphinx/domains/python.py
        refnode["json:schema"] = env.ref_context.get("json:schema")
        if not has_explicit_title:
            title = title.lstrip(".")  # Only has a meaning for the target
            target = target.lstrip("~")  # Only has a meaning for the title
            # If the first character is a tilde, don't display the initial parts
            if title[0:1] == "~":
                title = title[1:]
                dot = title.rfind(".")
                if dot != -1:
                    title = title[dot + 1 :]
        # if the first character is a dot, search more specific namespaces first
        # else search builtins first
        if target[0:1] == ".":
            target = target[1:]
            refnode["refspecific"] = True
        return title, target


class DomainSchemaEntry(NamedTuple):
    """Entry representing a schema in the JSON domain."""

    docname: str
    """Document id where schema is defined."""

    node_id: str
    """Sphinx reference id"""

    schema_id: str
    """Canonical schema id"""

    synopsis: Optional[str]
    """Rendered text content of synopsis (title property)"""

    objtype: str
    """Object type, either 'schema' or 'subschema'"""


# Sphinx object priority values

OBJECT_PRIORITY_DEFAULT = 1
OBJECT_PRIORITY_IMPORTANT = 0
OBJECT_PRIORITY_UNIMPORTANT = 2
OBJECT_PRIORITY_EXCLUDE_FROM_SEARCH = -1


class JsonSchemaDomain(sphinx.domains.Domain):
    """Domain for JSON schemas."""

    name = "json"
    label = "JSON"

    roles = {
        "schema": JsonSchemaRole(warn_dangling=True),
    }

    object_types: Dict[str, sphinx.domains.ObjType] = {
        "schema": sphinx.domains.ObjType("type", "schema"),
        "subschema": sphinx.domains.ObjType("member", "schema"),
    }

    directives = {
        "schema": JsonSchemaDirective,
    }

    initial_data: Dict[str, DomainSchemaEntry] = {
        "schemas": cast(DomainSchemaEntry, {}),
    }

    @property
    def schemas(self) -> Dict[str, DomainSchemaEntry]:
        return self.data["schemas"]

    def get_objects(self) -> Iterator[Tuple[str, str, str, str, str, int]]:
        for refname, obj in self.schemas.items():
            yield (
                refname,
                refname,
                obj.objtype,
                obj.docname,
                obj.node_id,
                OBJECT_PRIORITY_DEFAULT
                if obj.objtype == "schema"
                else OBJECT_PRIORITY_UNIMPORTANT,
            )

    def merge_domaindata(
        self, docnames: List[str], otherdata: Dict
    ) -> None:  # pylint: disable=g-bare-generic
        self.schemas.update(otherdata["schemas"])

    def _find_schema(
        self, target: str, parent_schema: Optional[str], refspecific: bool
    ) -> Optional[Tuple[str, DomainSchemaEntry]]:
        """Looks up a cross-referenced schema.

        :param target: Target schema identifier.
        :param parent_schema: Fully-qualified parent id.
        :param refspecific: Whether to search for relative targets first.

        :returns:
          If not found, returns `None`.  Otherwise, returns the
          ``(fully_qualified_id, entry)``
        """
        search_prefixes = [""]
        if parent_schema:
            parent_parts = parent_schema.split(".")
            for i in range(1, len(parent_parts) + 1):
                search_prefixes.append(".".join(parent_parts[:i]) + ".")
        if refspecific:
            search_prefixes = search_prefixes[::-1]
        for search_prefix in search_prefixes:
            full_name = search_prefix + target
            domain_entry = self.schemas.get(full_name)
            if domain_entry is not None:
                return (full_name, domain_entry)
        # Match not found
        return None

    def resolve_xref(
        self,
        env: sphinx.environment.BuildEnvironment,
        fromdocname: str,
        builder: sphinx.builders.Builder,
        typ: str,
        target: str,
        node: sphinx.addnodes.pending_xref,
        contnode: docutils.nodes.Element,
    ) -> Optional[docutils.nodes.Element]:
        del typ
        match = self._find_schema(
            target=target,
            parent_schema=node.get("json:schema"),
            refspecific=node.hasattr("refspecific"),
        )
        if match is None:
            return None
        return self._make_refnode(
            builder=builder, fromdocname=fromdocname, contnode=contnode, match=match
        )

    def _make_refnode(
        self,
        builder: sphinx.builders.Builder,
        fromdocname: str,
        contnode: docutils.nodes.Element,
        match: Tuple[str, DomainSchemaEntry],
    ) -> docutils.nodes.Element:
        full_name, domain_entry = match
        options = object_description_options.get_object_description_options(
            self.env, "json", domain_entry.objtype
        )
        tooltip = object_description_options.format_object_description_tooltip(
            self.env, options, full_name, domain_entry.synopsis
        )
        return sphinx.util.nodes.make_refnode(
            builder=builder,
            fromdocname=fromdocname,
            todocname=domain_entry.docname,
            targetid=domain_entry.node_id,
            child=contnode,
            title=tooltip,
        )

    def resolve_any_xref(
        self,
        env: sphinx.environment.BuildEnvironment,
        fromdocname: str,
        builder: sphinx.builders.Builder,
        target: str,
        node: sphinx.addnodes.pending_xref,
        contnode: docutils.nodes.Element,
    ) -> List[Tuple[str, docutils.nodes.Element]]:
        match = self._find_schema(
            target=target, parent_schema=node.get("json:schema"), refspecific=True
        )
        results: List[Tuple[str, docutils.nodes.Element]] = []
        if match is not None:
            results.append(
                (
                    "json:schema",
                    self._make_refnode(
                        builder=builder,
                        fromdocname=fromdocname,
                        contnode=contnode,
                        match=match,
                    ),
                )
            )
        return results

    def get_object_synopses(self) -> Iterator[Tuple[Tuple[str, str], str]]:
        for obj in self.schemas.values():
            synopsis = obj.synopsis
            if synopsis:
                yield ((obj.docname, obj.node_id), synopsis)


def _builder_inited(app: sphinx.application.Sphinx):
    _populate_json_schema_id_map(app)


def setup(app: sphinx.application.Sphinx):
    app.add_domain(JsonSchemaDomain)
    app.add_config_value("json_schemas", types=(List[str],), default=[], rebuild="env")
    app.add_config_value(
        "json_schema_validate", types=(bool,), default=False, rebuild="env"
    )
    app.add_config_value(
        "json_schema_rst_prolog", types=(str,), default="", rebuild="env"
    )
    app.add_config_value(
        "json_schema_rst_epilog", types=(str,), default="", rebuild="env"
    )
    app.connect("builder-inited", _builder_inited)
    return {"parallel_read_safe": True, "parallel_write_safe": True}