stygian-graph 0.9.2

//! Intelligent schema discovery service
//!
//! Automatically infers JSON Schema definitions from:
//! 1. **JSON/object examples** — type-inspects values to generate schema
//! 2. **HTML content** — uses an AI provider to suggest extraction schemas
//! 3. **User corrections** — applies overrides stored in a correction map
//! 4. **Evolution detection** — compares new inferences against cached schemas
//!
//! The service sits in the application layer and depends on `AIProvider` for
//! LLM-assisted HTML schema inference.
//!
//! # Example
//!
//! ```no_run
//! use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
//! use serde_json::json;
//!
//! # tokio::runtime::Runtime::new().unwrap().block_on(async {
//! let service = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None);
//!
//! // Infer from a JSON example
//! let example = json!({"name": "Alice", "age": 30, "active": true});
//! let schema = service.infer_from_example(&example);
//! assert_eq!(schema["type"].as_str().unwrap(), "object");
//! # });
//! ```

use std::collections::HashMap;
use std::sync::Arc;

use parking_lot::RwLock;
use serde_json::{Map, Value, json};
use tracing::debug;

/// Full standard GraphQL introspection query (6-level `ofType` depth for Relay
/// connection types such as `NON_NULL(LIST(NON_NULL(OBJECT)))`).
const INTROSPECTION_QUERY: &str = r"
query IntrospectionQuery {
  __schema {
    queryType { name }
    mutationType { name }
    subscriptionType { name }
    types { ...FullType }
    directives {
      name description locations
      args { ...InputValue }
    }
  }
}
fragment FullType on __Type {
  kind name description
  fields(includeDeprecated: true) {
    name description
    args { ...InputValue }
    type { ...TypeRef }
    isDeprecated deprecationReason
  }
  inputFields { ...InputValue }
  interfaces { ...TypeRef }
  enumValues(includeDeprecated: true) {
    name description isDeprecated deprecationReason
  }
  possibleTypes { ...TypeRef }
}
fragment InputValue on __InputValue {
  name description
  type { ...TypeRef }
  defaultValue
}
fragment TypeRef on __Type {
  kind name
  ofType {
    kind name
    ofType {
      kind name
      ofType {
        kind name
        ofType {
          kind name
          ofType {
            kind name
            ofType { kind name }
          }
        }
      }
    }
  }
}
";

use crate::domain::error::{ProviderError, Result, ServiceError, StygianError};
use crate::ports::{AIProvider, GraphQlAuth, ScrapingService, ServiceInput};

/// Discovered schema with metadata
#[derive(Debug, Clone)]
pub struct DiscoveredSchema {
    /// The JSON Schema object
    pub schema: Value,
    /// How the schema was derived
    pub source: SchemaSource,
    /// Confidence score [0.0, 1.0]
    pub confidence: f32,
    /// URL pattern or domain the schema is associated with
    pub url_pattern: Option<String>,
}

/// How a schema was derived
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum SchemaSource {
    /// Inferred by type-inspecting a JSON example
    ExampleInference,
    /// Generated by an AI provider analysing HTML structure
    AiInference,
    /// Loaded from cache
    Cached,
    /// Manually supplied / corrected by user
    UserCorrection,
}

/// Configuration for the schema discovery service
#[derive(Debug, Clone)]
pub struct SchemaDiscoveryConfig {
    /// Maximum HTML content length sent to AI provider
    pub max_content_chars: usize,
    /// Cache discovered schemas by URL pattern
    pub cache_schemas: bool,
    /// Minimum confidence to accept an AI-inferred schema
    pub min_ai_confidence: f32,
}

impl Default for SchemaDiscoveryConfig {
    fn default() -> Self {
        Self {
            max_content_chars: 16_000,
            cache_schemas: true,
            min_ai_confidence: 0.6,
        }
    }
}

/// Schema evolution diff — describes changes between two schema versions
#[derive(Debug, Clone)]
pub struct SchemaDiff {
    /// Fields present in new schema but not in old
    pub added_fields: Vec<String>,
    /// Fields present in old schema but not in new
    pub removed_fields: Vec<String>,
    /// Fields whose type changed
    pub type_changes: Vec<(String, String, String)>, // (field, old_type, new_type)
    /// Whether schemas are considered compatible
    pub is_compatible: bool,
}

/// Intelligent schema discovery service
///
/// Infers JSON Schemas from content using local heuristics and an optional
/// AI provider. Maintains a schema cache for URL patterns and supports
/// user corrections.
pub struct SchemaDiscoveryService {
    config: SchemaDiscoveryConfig,
    /// Optional AI provider for HTML-to-schema inference
    ai_provider: Option<Arc<dyn AIProvider>>,
    /// Schema cache: `url_pattern` → [`DiscoveredSchema`]
    cache: Arc<RwLock<HashMap<String, DiscoveredSchema>>>,
    /// User correction map: `field_name` → corrected schema fragment
    corrections: Arc<RwLock<HashMap<String, Value>>>,
    /// Optional GraphQL scraping service for introspection
    graphql_service: Option<Arc<dyn ScrapingService>>,
}

impl SchemaDiscoveryService {
    /// Create a new schema discovery service
    ///
    /// # Example
    ///
    /// ```
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
    ///
    /// let service = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None);
    /// ```
    pub fn new(config: SchemaDiscoveryConfig, ai_provider: Option<Arc<dyn AIProvider>>) -> Self {
        Self {
            config,
            ai_provider,
            cache: Arc::new(RwLock::new(HashMap::new())),
            corrections: Arc::new(RwLock::new(HashMap::new())),
            graphql_service: None,
        }
    }

    /// Attach a GraphQL scraping service used by [`infer_from_graphql`](Self::infer_from_graphql).
    ///
    /// # Example
    ///
    /// ```no_run
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
    /// use stygian_graph::adapters::graphql::GraphQlService;
    /// use std::sync::Arc;
    ///
    /// let gql = Arc::new(GraphQlService::new(Default::default(), None));
    /// let svc = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None)
    ///     .with_graphql_service(gql);
    /// ```
    #[must_use]
    pub fn with_graphql_service(mut self, service: Arc<dyn ScrapingService>) -> Self {
        self.graphql_service = Some(service);
        self
    }

    /// Infer a JSON Schema by type-inspecting a JSON example value.
    ///
    /// Supports objects, arrays, strings, numbers, booleans, and null.
    ///
    /// # Example
    ///
    /// ```
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
    /// use serde_json::json;
    ///
    /// let svc = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None);
    /// let example = json!({"name": "Alice", "score": 9.5, "active": true});
    /// let schema = svc.infer_from_example(&example);
    ///
    /// assert_eq!(schema["type"].as_str().unwrap(), "object");
    /// assert_eq!(schema["properties"]["name"]["type"].as_str().unwrap(), "string");
    /// assert_eq!(schema["properties"]["score"]["type"].as_str().unwrap(), "number");
    /// ```
    pub fn infer_from_example(&self, example: &Value) -> Value {
        Self::infer_schema_for_value(example)
    }

    /// Recursively infer a JSON Schema for a single JSON value
    fn infer_schema_for_value(value: &Value) -> Value {
        match value {
            Value::Object(map) => {
                let mut properties = Map::new();
                let mut required = Vec::new();
                for (key, val) in map {
                    properties.insert(key.clone(), Self::infer_schema_for_value(val));
                    required.push(json!(key));
                }
                json!({
                    "type": "object",
                    "properties": Value::Object(properties),
                    "required": required
                })
            }
            Value::Array(items) => {
                // Infer item schema from first element
                let item_schema = items
                    .first()
                    .map_or_else(|| json!({}), Self::infer_schema_for_value);
                json!({
                    "type": "array",
                    "items": item_schema
                })
            }
            Value::String(_) => json!({"type": "string"}),
            Value::Number(n) => {
                if n.is_i64() || n.is_u64() {
                    json!({"type": "integer"})
                } else {
                    json!({"type": "number"})
                }
            }
            Value::Bool(_) => json!({"type": "boolean"}),
            Value::Null => json!({"type": "null"}),
        }
    }

    /// Infer a schema from HTML content using an AI provider.
    ///
    /// Returns an error if no AI provider is configured.
    ///
    /// # Example
    ///
    /// ```no_run
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
    ///
    /// # tokio::runtime::Runtime::new().unwrap().block_on(async {
    /// let service = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None);
    /// // Without an AI provider, this returns Err
    /// let result = service.infer_from_html("<html>...</html>", None).await;
    /// assert!(result.is_err());
    /// # });
    /// ```
    pub async fn infer_from_html(
        &self,
        html: &str,
        url_pattern: Option<&str>,
    ) -> Result<DiscoveredSchema> {
        // Check cache first
        if self.config.cache_schemas
            && let Some(pattern) = url_pattern
        {
            let cache = self.cache.read();
            if let Some(cached) = cache.get(pattern) {
                debug!(pattern, "Schema cache hit");
                return Ok(DiscoveredSchema {
                    source: SchemaSource::Cached,
                    ..cached.clone()
                });
            }
        }

        let provider = self.ai_provider.as_ref().ok_or_else(|| {
            StygianError::Provider(ProviderError::ApiError(
                "No AI provider configured for HTML schema discovery".to_string(),
            ))
        })?;

        let truncated = if html.len() > self.config.max_content_chars {
            &html[..self.config.max_content_chars]
        } else {
            html
        };

        let discovery_schema = json!({
            "type": "object",
            "properties": {
                "schema": {
                    "type": "object",
                    "description": "A valid JSON Schema object describing the structured data in the HTML"
                },
                "confidence": {
                    "type": "number",
                    "minimum": 0.0,
                    "maximum": 1.0,
                    "description": "Confidence in the inferred schema [0.0, 1.0]"
                },
                "description": {
                    "type": "string",
                    "description": "Human-readable explanation of what data was detected"
                }
            },
            "required": ["schema", "confidence"]
        });

        let prompt = format!(
            "Analyze the following HTML page and infer a JSON Schema for the main structured \
             data it contains (e.g. product listings, articles, search results). \
             Return the schema definition.\n\nHTML:\n{truncated}"
        );

        let result = provider.extract(prompt, discovery_schema).await?;

        let schema = result
            .get("schema")
            .cloned()
            .unwrap_or_else(|| json!({"type": "object"}));

        #[allow(clippy::cast_possible_truncation)]
        let confidence = result
            .get("confidence")
            .and_then(Value::as_f64)
            .unwrap_or(0.7) as f32;

        // Apply any user corrections
        let schema = self.apply_corrections(schema);

        let discovered = DiscoveredSchema {
            schema,
            source: SchemaSource::AiInference,
            confidence,
            url_pattern: url_pattern.map(str::to_string),
        };

        // Cache the result
        if self.config.cache_schemas
            && let Some(pattern) = url_pattern
        {
            self.cache
                .write()
                .insert(pattern.to_string(), discovered.clone());
        }

        Ok(discovered)
    }

    /// Apply user corrections to an inferred schema
    fn apply_corrections(&self, mut schema: Value) -> Value {
        let corrections = self.corrections.read();
        if corrections.is_empty() {
            return schema;
        }

        if let Some(props) = schema.get_mut("properties").and_then(Value::as_object_mut) {
            for (field, correction) in corrections.iter() {
                if props.contains_key(field) {
                    props.insert(field.clone(), correction.clone());
                }
            }
        }
        schema
    }

    /// Register a user correction for a specific field type
    ///
    /// # Example
    ///
    /// ```
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
    /// use serde_json::json;
    ///
    /// let svc = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None);
    /// // Always use "integer" for "id" fields regardless of inferred type
    /// svc.add_correction("id".to_string(), json!({"type": "integer"}));
    /// ```
    pub fn add_correction(&self, field_name: String, schema_fragment: Value) {
        self.corrections.write().insert(field_name, schema_fragment);
    }

    /// Cache a schema for a URL pattern (useful for preloading known schemas)
    ///
    /// # Example
    ///
    /// ```
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig, SchemaSource};
    /// use serde_json::json;
    ///
    /// let svc = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None);
    /// svc.cache_schema(
    ///     "example.com/products".to_string(),
    ///     json!({"type": "object"}),
    ///     SchemaSource::UserCorrection,
    ///     1.0,
    /// );
    /// ```
    pub fn cache_schema(
        &self,
        url_pattern: String,
        schema: Value,
        source: SchemaSource,
        confidence: f32,
    ) {
        let discovered = DiscoveredSchema {
            schema,
            source,
            confidence,
            url_pattern: Some(url_pattern.clone()),
        };
        self.cache.write().insert(url_pattern, discovered);
    }

    /// Detect schema evolution between a known schema and a newly inferred one.
    ///
    /// Returns a diff describing added/removed/changed fields.
    ///
    /// # Example
    ///
    /// ```
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
    /// use serde_json::json;
    ///
    /// let svc = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None);
    ///
    /// let old = json!({
    ///     "type": "object",
    ///     "properties": {"name": {"type": "string"}, "price": {"type": "number"}}
    /// });
    /// let new = json!({
    ///     "type": "object",
    ///     "properties": {"name": {"type": "string"}, "discount": {"type": "number"}}
    /// });
    ///
    /// let diff = svc.detect_evolution(&old, &new);
    /// assert!(diff.added_fields.contains(&"discount".to_string()));
    /// assert!(diff.removed_fields.contains(&"price".to_string()));
    /// ```
    pub fn detect_evolution(&self, old_schema: &Value, new_schema: &Value) -> SchemaDiff {
        let old_props = old_schema
            .get("properties")
            .and_then(Value::as_object)
            .cloned()
            .unwrap_or_default();
        let new_props = new_schema
            .get("properties")
            .and_then(Value::as_object)
            .cloned()
            .unwrap_or_default();

        let added_fields: Vec<String> = new_props
            .keys()
            .filter(|k| !old_props.contains_key(*k))
            .cloned()
            .collect();

        let removed_fields: Vec<String> = old_props
            .keys()
            .filter(|k| !new_props.contains_key(*k))
            .cloned()
            .collect();

        let type_changes: Vec<(String, String, String)> = old_props
            .iter()
            .filter_map(|(k, old_v)| {
                let new_v = new_props.get(k)?;
                let old_t = old_v.get("type").and_then(Value::as_str).unwrap_or("any");
                let new_t = new_v.get("type").and_then(Value::as_str).unwrap_or("any");
                (old_t != new_t).then(|| (k.clone(), old_t.to_string(), new_t.to_string()))
            })
            .collect();

        let is_compatible = removed_fields.is_empty() && type_changes.is_empty();

        SchemaDiff {
            added_fields,
            removed_fields,
            type_changes,
            is_compatible,
        }
    }

    /// Discover a JSON Schema by introspecting a live GraphQL endpoint.
    ///
    /// Fires the standard `__schema` introspection query, converts the returned
    /// SDL types to a JSON Schema representation, and caches the result by
    /// endpoint URL.
    ///
    /// # Errors
    ///
    /// Returns `Err` if no GraphQL service is configured, if the endpoint is
    /// unreachable, if the response contains GraphQL errors, or if the
    /// introspection response cannot be parsed.
    ///
    /// # Example
    ///
    /// ```no_run
    /// use stygian_graph::application::schema_discovery::{SchemaDiscoveryService, SchemaDiscoveryConfig};
    /// use stygian_graph::adapters::graphql::GraphQlService;
    /// use std::sync::Arc;
    ///
    /// # tokio::runtime::Runtime::new().unwrap().block_on(async {
    /// let gql = Arc::new(GraphQlService::new(Default::default(), None));
    /// let svc = SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None)
    ///     .with_graphql_service(gql);
    /// let schema = svc.infer_from_graphql("https://api.example.com/graphql", None).await;
    /// # });
    /// ```
    #[allow(clippy::indexing_slicing)]
    pub async fn infer_from_graphql(
        &self,
        endpoint: &str,
        auth: Option<GraphQlAuth>,
    ) -> Result<DiscoveredSchema> {
        // ── 1. Cache check ────────────────────────────────────────────────
        if self.config.cache_schemas {
            let cache = self.cache.read();
            if let Some(cached) = cache.get(endpoint) {
                debug!(endpoint, "GraphQL introspection cache hit");
                return Ok(DiscoveredSchema {
                    source: SchemaSource::Cached,
                    ..cached.clone()
                });
            }
        }

        // ── 2. Require a GraphQL service ──────────────────────────────────
        let service = self.graphql_service.as_ref().ok_or_else(|| {
            StygianError::Service(ServiceError::Unavailable(
                "No GraphQL service configured for introspection".to_string(),
            ))
        })?;

        // ── 3. Build ServiceInput with introspection query ────────────────
        let mut params = json!({
            "query": INTROSPECTION_QUERY,
        });
        if let Some(a) = auth {
            params["auth"] = json!({
                "kind": match a.kind {
                    crate::ports::GraphQlAuthKind::Bearer => "bearer",
                    crate::ports::GraphQlAuthKind::ApiKey => "api_key",
                    crate::ports::GraphQlAuthKind::Header => "header",
                    crate::ports::GraphQlAuthKind::None => "none",
                },
                "token": a.token,
            });
            if let Some(header) = a.header_name {
                params["auth"]["header_name"] = json!(header);
            }
        }

        let input = ServiceInput {
            url: endpoint.to_string(),
            params,
        };

        // ── 4. Execute ────────────────────────────────────────────────────
        let output = service.execute(input).await?;

        // ── 5. Parse __schema ─────────────────────────────────────────────
        let response_json: Value = serde_json::from_str(&output.data)
            .map_err(|e| StygianError::Service(ServiceError::InvalidResponse(e.to_string())))?;
        let introspection = response_json.get("__schema").ok_or_else(|| {
            StygianError::Service(ServiceError::InvalidResponse(
                "introspection response missing __schema key".to_string(),
            ))
        })?;

        // ── 6. Build JSON Schema from types ───────────────────────────────
        let query_type_name = introspection
            .get("queryType")
            .and_then(|qt| qt.get("name"))
            .and_then(Value::as_str)
            .unwrap_or("Query")
            .to_string();

        let types = introspection
            .get("types")
            .and_then(Value::as_array)
            .cloned()
            .unwrap_or_default();

        let mut definitions = Map::new();
        for gql_type in &types {
            let name = gql_type.get("name").and_then(Value::as_str).unwrap_or("");
            // Skip built-in introspection types
            if name.is_empty() || name.starts_with('_') {
                continue;
            }
            let def = Self::convert_type_to_definition(gql_type);
            definitions.insert(name.to_string(), def);
        }

        let schema = json!({
            "type": "object",
            "description": format!("GraphQL schema for {endpoint}"),
            "definitions": Value::Object(definitions),
            "queryType": query_type_name,
        });

        let discovered = DiscoveredSchema {
            schema,
            source: SchemaSource::AiInference,
            confidence: 1.0,
            url_pattern: Some(endpoint.to_string()),
        };

        // ── 7. Cache ───────────────────────────────────────────────────────
        if self.config.cache_schemas {
            self.cache
                .write()
                .insert(endpoint.to_string(), discovered.clone());
        }

        Ok(discovered)
    }

    /// Convert a top-level GraphQL type definition to a JSON Schema object.
    #[allow(clippy::indexing_slicing)]
    fn convert_type_to_definition(gql_type: &Value) -> Value {
        let kind = gql_type.get("kind").and_then(Value::as_str).unwrap_or("");
        match kind {
            "OBJECT" | "INTERFACE" | "INPUT_OBJECT" => {
                let fields: Vec<Value> = gql_type
                    .get("fields")
                    .or_else(|| gql_type.get("inputFields"))
                    .and_then(Value::as_array)
                    .cloned()
                    .unwrap_or_default();

                let mut properties = Map::new();
                let mut required: Vec<Value> = Vec::new();

                for field in &fields {
                    let field_name = field.get("name").and_then(Value::as_str).unwrap_or("");
                    if field_name.is_empty() {
                        continue;
                    }
                    let type_ref = &field["type"];
                    // NON_NULL at the outermost level means the field is required
                    if type_ref.get("kind").and_then(Value::as_str) == Some("NON_NULL") {
                        required.push(json!(field_name));
                    }
                    properties.insert(field_name.to_string(), Self::convert_type_ref(type_ref));
                }

                let mut obj = json!({
                    "type": "object",
                    "properties": Value::Object(properties),
                });
                if !required.is_empty() {
                    obj["required"] = json!(required);
                }
                obj
            }
            "ENUM" => {
                let values: Vec<Value> = gql_type
                    .get("enumValues")
                    .and_then(Value::as_array)
                    .map(|vals| {
                        vals.iter()
                            .filter_map(|v| v.get("name").and_then(Value::as_str))
                            .map(|s| json!(s))
                            .collect()
                    })
                    .unwrap_or_default();
                json!({ "type": "string", "enum": values })
            }
            "SCALAR" => {
                let name = gql_type.get("name").and_then(Value::as_str).unwrap_or("");
                Self::scalar_to_json_schema(name)
            }
            _ => json!({}),
        }
    }

    /// Recursively convert a GraphQL type reference (field type) to JSON Schema.
    fn convert_type_ref(type_ref: &Value) -> Value {
        let kind = type_ref.get("kind").and_then(Value::as_str).unwrap_or("");
        match kind {
            "NON_NULL" => Self::convert_type_ref(&type_ref["ofType"]),
            "LIST" => json!({
                "type": "array",
                "items": Self::convert_type_ref(&type_ref["ofType"]),
            }),
            "OBJECT" | "INTERFACE" | "INPUT_OBJECT" | "ENUM" => {
                let name = type_ref.get("name").and_then(Value::as_str).unwrap_or("");
                json!({ "$ref": format!("#/definitions/{name}") })
            }
            "SCALAR" => {
                let name = type_ref.get("name").and_then(Value::as_str).unwrap_or("");
                Self::scalar_to_json_schema(name)
            }
            _ => json!({}),
        }
    }

    /// Map a GraphQL scalar name to its JSON Schema type.
    fn scalar_to_json_schema(name: &str) -> Value {
        match name {
            "String" | "ID" => json!({ "type": "string" }),
            "Int" => json!({ "type": "integer" }),
            "Float" => json!({ "type": "number" }),
            "Boolean" => json!({ "type": "boolean" }),
            s if s.to_ascii_lowercase().contains("datetime")
                || s.to_ascii_lowercase().contains("date") =>
            {
                json!({ "type": "string", "format": "date-time" })
            }
            _ => json!({}),
        }
    }

    /// Validate that a value is a structurally valid JSON Schema object.
    ///
    /// Checks for the presence of a `type` or `properties` field.
    ///
    /// # Example
    ///
    /// ```
    /// use stygian_graph::application::schema_discovery::SchemaDiscoveryService;
    /// use serde_json::json;
    ///
    /// assert!(SchemaDiscoveryService::is_valid_schema(&json!({"type": "object"})));
    /// assert!(!SchemaDiscoveryService::is_valid_schema(&json!("not a schema")));
    /// ```
    pub fn is_valid_schema(schema: &Value) -> bool {
        if !schema.is_object() {
            return false;
        }
        schema.get("type").is_some() || schema.get("properties").is_some()
    }
}

#[cfg(test)]
#[allow(
    clippy::unwrap_used,
    clippy::indexing_slicing,
    clippy::approx_constant,
    clippy::significant_drop_tightening,
    clippy::float_cmp,
    clippy::unnecessary_map_or
)]
mod tests {
    use super::*;
    use serde_json::json;

    fn svc() -> SchemaDiscoveryService {
        SchemaDiscoveryService::new(SchemaDiscoveryConfig::default(), None)
    }

    // --- infer_from_example ---

    #[test]
    fn test_infer_object_schema() {
        let s = svc();
        let schema = s.infer_from_example(&json!({"name": "Alice", "age": 30, "active": true}));
        assert_eq!(schema["type"].as_str().unwrap(), "object");
        assert_eq!(
            schema["properties"]["name"]["type"].as_str().unwrap(),
            "string"
        );
        assert_eq!(
            schema["properties"]["age"]["type"].as_str().unwrap(),
            "integer"
        );
        assert_eq!(
            schema["properties"]["active"]["type"].as_str().unwrap(),
            "boolean"
        );
    }

    #[test]
    fn test_infer_array_schema() {
        let s = svc();
        let schema = s.infer_from_example(&json!([{"id": 1}, {"id": 2}]));
        assert_eq!(schema["type"].as_str().unwrap(), "array");
        assert_eq!(schema["items"]["type"].as_str().unwrap(), "object");
    }

    #[test]
    fn test_infer_string_schema() {
        let s = svc();
        let schema = s.infer_from_example(&json!("hello"));
        assert_eq!(schema["type"].as_str().unwrap(), "string");
    }

    #[test]
    fn test_infer_number_float() {
        let s = svc();
        let schema = s.infer_from_example(&json!(3.14));
        assert_eq!(schema["type"].as_str().unwrap(), "number");
    }

    #[test]
    fn test_infer_null() {
        let s = svc();
        let schema = s.infer_from_example(&Value::Null);
        assert_eq!(schema["type"].as_str().unwrap(), "null");
    }

    // --- is_valid_schema ---

    #[test]
    fn test_valid_schema_with_type() {
        assert!(SchemaDiscoveryService::is_valid_schema(
            &json!({"type": "object"})
        ));
    }

    #[test]
    fn test_valid_schema_with_properties() {
        assert!(SchemaDiscoveryService::is_valid_schema(
            &json!({"properties": {"x": {"type": "string"}}})
        ));
    }

    #[test]
    fn test_invalid_schema_string() {
        assert!(!SchemaDiscoveryService::is_valid_schema(&json!(
            "not schema"
        )));
    }

    #[test]
    fn test_invalid_schema_empty_object() {
        // An empty object has neither "type" nor "properties"
        assert!(!SchemaDiscoveryService::is_valid_schema(&json!({})));
    }

    // --- detect_evolution ---

    #[test]
    fn test_evolution_added_field() {
        let s = svc();
        let old = json!({"type": "object", "properties": {"name": {"type": "string"}}});
        let new = json!({"type": "object", "properties": {"name": {"type": "string"}, "email": {"type": "string"}}});
        let diff = s.detect_evolution(&old, &new);
        assert!(diff.added_fields.contains(&"email".to_string()));
        assert!(diff.removed_fields.is_empty());
        assert!(diff.is_compatible);
    }

    #[test]
    fn test_evolution_removed_field() {
        let s = svc();
        let old = json!({"type": "object", "properties": {"name": {"type": "string"}, "age": {"type": "integer"}}});
        let new = json!({"type": "object", "properties": {"name": {"type": "string"}}});
        let diff = s.detect_evolution(&old, &new);
        assert!(diff.removed_fields.contains(&"age".to_string()));
        assert!(!diff.is_compatible);
    }

    #[test]
    fn test_evolution_type_change() {
        let s = svc();
        let old = json!({"type": "object", "properties": {"id": {"type": "string"}}});
        let new = json!({"type": "object", "properties": {"id": {"type": "integer"}}});
        let diff = s.detect_evolution(&old, &new);
        assert_eq!(diff.type_changes.len(), 1);
        assert_eq!(
            diff.type_changes[0],
            (
                "id".to_string(),
                "string".to_string(),
                "integer".to_string()
            )
        );
        assert!(!diff.is_compatible);
    }

    #[test]
    fn test_evolution_no_change() {
        let s = svc();
        let schema = json!({"type": "object", "properties": {"x": {"type": "string"}}});
        let diff = s.detect_evolution(&schema, &schema);
        assert!(diff.added_fields.is_empty());
        assert!(diff.removed_fields.is_empty());
        assert!(diff.is_compatible);
    }

    // --- corrections ---

    #[test]
    fn test_correction_overrides_inferred_type() {
        let s = svc();
        s.add_correction("id".to_string(), json!({"type": "integer"}));

        let schema = json!({
            "type": "object",
            "properties": {
                "id": {"type": "string"},
                "name": {"type": "string"}
            }
        });
        let corrected = s.apply_corrections(schema);
        assert_eq!(
            corrected["properties"]["id"]["type"].as_str().unwrap(),
            "integer"
        );
        assert_eq!(
            corrected["properties"]["name"]["type"].as_str().unwrap(),
            "string"
        );
    }

    // --- cache ---

    #[test]
    fn test_cache_and_retrieve() {
        let s = svc();
        let schema = json!({"type": "object", "properties": {"title": {"type": "string"}}});
        s.cache_schema(
            "example.com".to_string(),
            schema.clone(),
            SchemaSource::UserCorrection,
            1.0,
        );

        let cache = s.cache.read();
        let entry = cache.get("example.com").unwrap();
        assert_eq!(entry.schema, schema);
        assert_eq!(entry.source, SchemaSource::UserCorrection);
    }

    // --- HTML inference without provider ---

    #[tokio::test]
    async fn test_infer_from_html_no_provider() {
        let s = svc();
        let result = s
            .infer_from_html("<html><body>test</body></html>", None)
            .await;
        assert!(result.is_err());
        assert!(result.unwrap_err().to_string().contains("No AI provider"));
    }

    // --- HTML inference with cache hit ---

    #[tokio::test]
    async fn test_infer_from_html_cache_hit() {
        let s = svc();
        let schema = json!({"type": "object"});
        s.cache_schema(
            "example.com".to_string(),
            schema.clone(),
            SchemaSource::Cached,
            0.9,
        );

        let result = s
            .infer_from_html("<html/>", Some("example.com"))
            .await
            .unwrap();
        assert_eq!(result.source, SchemaSource::Cached);
        assert_eq!(result.schema, schema);
    }

    // ── GraphQL introspection tests ──────────────────────────────────────────

    /// Minimal mock [`ScrapingService`] that returns a pre-baked `Value` as `data`.
    struct MockGraphQlService(Value);

    #[async_trait::async_trait]
    impl ScrapingService for MockGraphQlService {
        fn name(&self) -> &'static str {
            "mock_graphql"
        }
        async fn execute(
            &self,
            _input: crate::ports::ServiceInput,
        ) -> Result<crate::ports::ServiceOutput> {
            Ok(crate::ports::ServiceOutput {
                data: serde_json::to_string(&self.0).unwrap_or_default(),
                metadata: Value::default(),
            })
        }
    }

    impl SchemaDiscoveryService {
        /// Test-only constructor: pre-bake a GraphQL introspection response.
        ///
        /// The `response` value should represent the `data` object that a real
        /// GraphQL service would return (i.e. `{ "__schema": { ... } }`).
        fn with_mock_graphql(config: SchemaDiscoveryConfig, response: Value) -> Self {
            let svc = Arc::new(MockGraphQlService(response));
            Self::new(config, None).with_graphql_service(svc)
        }
    }

    fn make_introspection_response_simple() -> Value {
        json!({
            "__schema": {
                "queryType": { "name": "Query" },
                "mutationType": null,
                "subscriptionType": null,
                "types": [
                    {
                        "kind": "OBJECT",
                        "name": "Query",
                        "fields": [
                            {
                                "name": "user",
                                "type": { "kind": "OBJECT", "name": "User", "ofType": null }
                            }
                        ],
                        "inputFields": null,
                        "interfaces": [],
                        "enumValues": null,
                        "possibleTypes": null
                    },
                    {
                        "kind": "OBJECT",
                        "name": "User",
                        "fields": [
                            {
                                "name": "id",
                                "type": { "kind": "NON_NULL", "name": null, "ofType": { "kind": "SCALAR", "name": "ID", "ofType": null } }
                            },
                            {
                                "name": "name",
                                "type": { "kind": "SCALAR", "name": "String", "ofType": null }
                            }
                        ],
                        "inputFields": null,
                        "interfaces": [],
                        "enumValues": null,
                        "possibleTypes": null
                    },
                    {
                        "kind": "SCALAR",
                        "name": "String",
                        "fields": null,
                        "inputFields": null,
                        "interfaces": null,
                        "enumValues": null,
                        "possibleTypes": null
                    },
                    {
                        "kind": "SCALAR",
                        "name": "ID",
                        "fields": null,
                        "inputFields": null,
                        "interfaces": null,
                        "enumValues": null,
                        "possibleTypes": null
                    }
                ],
                "directives": []
            }
        })
    }

    #[tokio::test]
    async fn introspection_no_graphql_service() {
        let s = svc(); // no graphql_service configured
        let result = s
            .infer_from_graphql("https://api.example.com/graphql", None)
            .await;
        assert!(result.is_err());
        assert!(
            result
                .unwrap_err()
                .to_string()
                .contains("No GraphQL service")
        );
    }

    #[test]
    fn graphql_type_to_json_schema_scalar() {
        assert_eq!(
            SchemaDiscoveryService::scalar_to_json_schema("String"),
            json!({"type": "string"})
        );
        assert_eq!(
            SchemaDiscoveryService::scalar_to_json_schema("Int"),
            json!({"type": "integer"})
        );
        assert_eq!(
            SchemaDiscoveryService::scalar_to_json_schema("Float"),
            json!({"type": "number"})
        );
        assert_eq!(
            SchemaDiscoveryService::scalar_to_json_schema("Boolean"),
            json!({"type": "boolean"})
        );
        assert_eq!(
            SchemaDiscoveryService::scalar_to_json_schema("ID"),
            json!({"type": "string"})
        );
        assert_eq!(
            SchemaDiscoveryService::scalar_to_json_schema("DateTime"),
            json!({"type": "string", "format": "date-time"})
        );
        // Unknown scalar → permissive empty schema
        assert_eq!(
            SchemaDiscoveryService::scalar_to_json_schema("JSON"),
            json!({})
        );
    }

    #[test]
    fn graphql_type_to_json_schema_object() {
        let gql_type = json!({
            "kind": "OBJECT",
            "name": "User",
            "fields": [
                {
                    "name": "id",
                    "type": { "kind": "SCALAR", "name": "ID", "ofType": null }
                },
                {
                    "name": "email",
                    "type": { "kind": "SCALAR", "name": "String", "ofType": null }
                }
            ]
        });
        let schema = SchemaDiscoveryService::convert_type_to_definition(&gql_type);
        assert_eq!(schema["type"].as_str().unwrap(), "object");
        assert_eq!(
            schema["properties"]["id"]["type"].as_str().unwrap(),
            "string"
        );
        assert_eq!(
            schema["properties"]["email"]["type"].as_str().unwrap(),
            "string"
        );
        // Neither field is NON_NULL, so no required array
        assert!(
            schema.get("required").is_none()
                || schema["required"].as_array().is_none_or(Vec::is_empty)
        );
    }

    #[test]
    fn graphql_non_null_adds_to_required() {
        let gql_type = json!({
            "kind": "OBJECT",
            "name": "Product",
            "fields": [
                {
                    "name": "sku",
                    "type": {
                        "kind": "NON_NULL",
                        "name": null,
                        "ofType": { "kind": "SCALAR", "name": "String", "ofType": null }
                    }
                },
                {
                    "name": "description",
                    "type": { "kind": "SCALAR", "name": "String", "ofType": null }
                }
            ]
        });
        let schema = SchemaDiscoveryService::convert_type_to_definition(&gql_type);
        let required = schema["required"].as_array().unwrap();
        assert!(required.contains(&json!("sku")));
        assert!(!required.contains(&json!("description")));
    }

    #[tokio::test]
    async fn introspection_result_cached() {
        let response = make_introspection_response_simple();
        let s =
            SchemaDiscoveryService::with_mock_graphql(SchemaDiscoveryConfig::default(), response);

        // First call — fetches from mock service
        let first = s
            .infer_from_graphql("https://api.example.com/graphql", None)
            .await
            .unwrap();
        assert_eq!(first.source, SchemaSource::AiInference);
        assert_eq!(first.confidence, 1.0);

        // Second call — served from cache (source changes to Cached)
        let second = s
            .infer_from_graphql("https://api.example.com/graphql", None)
            .await
            .unwrap();
        assert_eq!(second.source, SchemaSource::Cached);
        assert_eq!(second.schema["queryType"].as_str().unwrap(), "Query");
    }
}