from __future__ import annotations as _annotations

import os
import warnings
from dataclasses import dataclass
from typing import TypeAlias, overload

from pydantic_ai import ModelProfile
from pydantic_ai._http import AsyncHTTPClient, create_async_httpx2_client
from pydantic_ai.profiles import merge_profile
from pydantic_ai.profiles.anthropic import AnthropicModelProfile, anthropic_model_profile
from pydantic_ai.providers import Provider, missing_api_key_error
from pydantic_ai.providers._bedrock_model_names import split_bedrock_model_id

from .._json_schema import JsonSchema, JsonSchemaTransformer

try:
    from anthropic import (
        AsyncAnthropic,
        AsyncAnthropicBedrock,  # pyright: ignore[reportPrivateImportUsage]
        AsyncAnthropicBedrockMantle,  # pyright: ignore[reportPrivateImportUsage]
        AsyncAnthropicFoundry,
        AsyncAnthropicVertex,  # pyright: ignore[reportPrivateImportUsage]
    )
except ImportError as _import_error:
    raise ImportError(
        'Please install the `anthropic` package to use the Anthropic provider, '
        'you can use the `anthropic` optional group — `pip install "pydantic-ai-slim[anthropic]"`'
    ) from _import_error

# Below the guard on purpose: `anthropic` requires `httpx2`, so without the extra the error above
# is what users should see, not `ModuleNotFoundError: httpx2`.
import httpx2

AsyncAnthropicClient: TypeAlias = (
    AsyncAnthropic | AsyncAnthropicBedrock | AsyncAnthropicBedrockMantle | AsyncAnthropicFoundry | AsyncAnthropicVertex
)

_INLINE_SYSTEM_PROMPT_MODEL_PREFIXES = (
    'claude-fable-5',
    'claude-mythos-5',
    'claude-opus-4-8',
    'claude-opus-5',
)
"""Models that honor a `{'role': 'system'}` entry inside the Messages API's `messages` array.

This is the list Anthropic publishes for the feature, and accepting the entry is not the same as
honoring it. Older models reject it outright (`role 'system' is not supported on this model`), but
`claude-sonnet-5` returns 200 and then ignores it — the docs say the feature "is not available on
Claude Sonnet 5; use the top-level `system` field instead", and measuring agrees: asked to lift a
restriction the top-level prompt set, it refuses every time where Opus 5 complies every time, and on
a plain formatting instruction the `<system>`-tagged fallback actually lands more often than the
entry does. So Sonnet 5 is deliberately absent, and a 200 is not evidence for adding a model here.

`claude-mythos-5` is published as supported but isn't reachable with our credentials.
"""

_TOOL_AVAILABILITY_DELTA_MODEL_PREFIXES = (
    'claude-fable-5',
    'claude-mythos-5',
    'claude-opus-4-8',
    'claude-opus-5',
)
"""Models that accept `tool_addition` / `tool_removal` blocks on a `{'role': 'system'}` entry.

The list Anthropic publishes for the `mid-conversation-tool-changes-2026-07-01` beta, and it happens
to match `_INLINE_SYSTEM_PROMPT_MODEL_PREFIXES` — the two remain separate settings because they're
separate features, one GA and one beta, that could diverge again. Models predating the beta reject
the blocks with `requires a model that supports ...`, and `claude-sonnet-5` rejects them outright
(`tool_addition/tool_removal is not supported on this model`) rather than accepting and ignoring them
the way it does a plain system entry. Verified live per model except `claude-mythos-5`, which isn't
reachable with our credentials and is included on the strength of the published list.
"""


class AnthropicProvider(Provider[AsyncAnthropicClient]):
    """Provider for Anthropic API."""

    @property
    def name(self) -> str:
        return 'anthropic'

    @property
    def base_url(self) -> str:
        return str(self._client.base_url)

    @property
    def client(self) -> AsyncAnthropicClient:
        return self._client

    @staticmethod
    def model_profile(model_name: str) -> ModelProfile | None:
        # When the underlying client is `AsyncAnthropicBedrock`/`AsyncAnthropicBedrockMantle`, the model
        # name carries a Bedrock `anthropic.` provider segment (and, on the legacy InvokeModel API, a
        # `-v<n>(:<m>)?` version suffix and optional cross-region geo prefix), e.g.
        # `us.anthropic.claude-haiku-4-5-20251001-v1:0`. Strip it so `anthropic_model_profile`'s
        # `claude-...` prefix checks match; the full model name still goes on the wire via `AnthropicModel._model_name`.
        bedrock_provider, base_model_name = split_bedrock_model_id(model_name)
        if bedrock_provider == 'anthropic':
            model_name = base_model_name
        profile = anthropic_model_profile(model_name)
        return merge_profile(
            AnthropicModelProfile(json_schema_transformer=AnthropicJsonSchemaTransformer),
            profile,
            # Accepting a `{'role': 'system'}` entry is a fact about the Messages API, not about the
            # model family, so it's set here rather than in `anthropic_model_profile()`, which is
            # shared with the Bedrock Converse API and the OpenAI-compatible gateways that route the
            # same models. Only the transport verified to serve the role opts in; everywhere else the
            # flag stays `False` and `Model.prepare_messages` keeps applying the `<system>`-tagged
            # rendering, as it did before this was supported anywhere.
            AnthropicModelProfile(
                supports_inline_system_prompts=model_name.startswith(_INLINE_SYSTEM_PROMPT_MODEL_PREFIXES),
            ),
            AnthropicModelProfile(tool_addition_mode='by_reference')
            if model_name.startswith(_TOOL_AVAILABILITY_DELTA_MODEL_PREFIXES)
            else AnthropicModelProfile(),
        )

    @overload
    def __init__(self, *, anthropic_client: AsyncAnthropicClient | None = None) -> None: ...

    @overload
    def __init__(
        self,
        *,
        api_key: str | None = None,
        base_url: str | None = None,
        http_client: httpx2.AsyncClient | None = None,
    ) -> None: ...

    def __init__(
        self,
        *,
        api_key: str | None = None,
        base_url: str | None = None,
        anthropic_client: AsyncAnthropicClient | None = None,
        http_client: httpx2.AsyncClient | None = None,
    ) -> None:
        """Create a new Anthropic provider.

        Args:
            api_key: The API key to use for authentication, if not provided, the `ANTHROPIC_API_KEY` environment variable
                will be used if available.
            base_url: The base URL to use for the Anthropic API.
            anthropic_client: An existing Anthropic client to use. Accepts
                [`AsyncAnthropic`](https://github.com/anthropics/anthropic-sdk-python),
                [`AsyncAnthropicBedrock`](https://platform.claude.com/docs/en/build-with-claude/claude-on-amazon-bedrock-legacy),
                [`AsyncAnthropicBedrockMantle`](https://platform.claude.com/docs/en/build-with-claude/claude-in-amazon-bedrock),
                [`AsyncAnthropicFoundry`](https://platform.claude.com/docs/en/build-with-claude/claude-in-microsoft-foundry), or
                [`AsyncAnthropicVertex`](https://docs.anthropic.com/en/api/claude-on-vertex-ai).
                If provided, the `api_key` and `http_client` arguments will be ignored.
            http_client: An existing `httpx2.AsyncClient` to use for making HTTP requests.
        """
        if anthropic_client is not None:
            assert http_client is None, 'Cannot provide both `anthropic_client` and `http_client`'
            assert api_key is None, 'Cannot provide both `anthropic_client` and `api_key`'
            self._client = anthropic_client
        else:
            api_key = api_key or os.getenv('ANTHROPIC_API_KEY')
            if not api_key:
                raise missing_api_key_error(
                    'Set the `ANTHROPIC_API_KEY` environment variable or pass it via `AnthropicProvider(api_key=...)`'
                    ' to use the Anthropic provider.'
                )
            if http_client is not None:
                self._client = AsyncAnthropic(api_key=api_key, base_url=base_url, http_client=http_client)
            else:
                http_client = create_async_httpx2_client()
                self._own_http_client = http_client
                self._http_client_factory = create_async_httpx2_client
                self._client = AsyncAnthropic(api_key=api_key, base_url=base_url, http_client=http_client)

    def _set_http_client(self, http_client: AsyncHTTPClient) -> None:
        assert isinstance(http_client, httpx2.AsyncClient)
        self._client._client = http_client  # pyright: ignore[reportPrivateUsage]


@dataclass(init=False)
class AnthropicJsonSchemaTransformer(JsonSchemaTransformer):
    """Transforms schemas to the subset supported by Anthropic structured outputs.

    Transformation is applied when:
    - `NativeOutput` is used as the `output_type` of the Agent
    - `strict=True` is set on the `Tool`

    The behavior of this transformer differs from the OpenAI one in that it sets `Tool.strict=False` by default when not explicitly set to True.

    Example:
        ```python
        from pydantic_ai import Agent

        agent = Agent('anthropic:claude-sonnet-4-6')

        @agent.tool_plain  # -> defaults to strict=False
        def my_tool(x: str) -> dict[str, int]:
            ...
        ```

    Anthropic's SDK `transform_schema()` automatically:
    - Adds `additionalProperties: false` to all objects (required by API)
    - Removes unsupported constraints (minLength, pattern, etc.)
    - Moves removed constraints to description field
    - Removes title and $schema fields
    """

    def walk(self) -> JsonSchema:
        schema = super().walk()

        # The caller (pydantic_ai.models._customize_tool_def or _customize_output_object) coalesces
        # - output_object.strict = self.is_strict_compatible
        # - tool_def.strict = self.is_strict_compatible
        # the reason we don't default to `strict=True` is that the transformation could be lossy
        # so in order to change the behavior (default to True), we need to come up with logic that will check for lossiness
        # https://github.com/pydantic/pydantic-ai/issues/3541
        self.is_strict_compatible = self.strict is True  # not compatible when strict is False/None

        if self.strict is True:
            from anthropic import transform_schema

            return transform_schema(schema)
        else:
            return schema

    def _handle_object(self, schema: JsonSchema) -> JsonSchema:
        schema = super()._handle_object(schema)
        if self.strict is True:
            additional_properties = schema.get('additionalProperties')
            if isinstance(additional_properties, dict) or additional_properties is True:
                warnings.warn(
                    '`dict` fields are not supported by Anthropic in strict mode '
                    '(including `NativeOutput`, which is automatically selected when thinking is enabled). '
                    "Anthropic's schema transformation sets `additionalProperties` to `false`, "
                    'which forces the model to return `{}`. '
                    'Use a `list` of `tuple[str, str]`, or a `TypedDict` or `dataclass` '
                    'with explicit `key` and `value` fields, '
                    'or set `output_type=PromptedOutput(...)`.',
                    UserWarning,
                )
        return schema

    def transform(self, schema: JsonSchema) -> JsonSchema:
        schema.pop('title', None)
        schema.pop('$schema', None)
        return schema