hoosh 1.2.0

//! Configuration file loading for hoosh.
//!
//! Loads `hoosh.toml` from the current directory or a specified path.
//! Environment variables in API keys are resolved at load time.

use std::path::Path;

use serde::Deserialize;

use crate::budget::TokenPool;
use crate::cache::CacheConfig;
use crate::provider::ProviderType;
use crate::router::{ProviderRoute, RoutingStrategy};
use crate::server::ServerConfig;

/// Top-level configuration file structure.
#[derive(Debug, Deserialize)]
pub struct HooshConfig {
    #[serde(default)]
    pub server: ServerSection,
    #[serde(default)]
    pub cache: CacheSection,
    #[serde(default)]
    pub providers: Vec<ProviderSection>,
    #[serde(default)]
    pub budgets: Vec<BudgetPoolSection>,
    #[serde(default)]
    pub whisper: WhisperSection,
    #[serde(default)]
    pub tts: TtsSection,
    #[serde(default)]
    pub audit: AuditSection,
    #[serde(default)]
    pub auth: AuthConfig,
    #[serde(default)]
    pub telemetry: TelemetrySection,
    #[serde(default)]
    pub context: ContextSection,
    #[serde(default)]
    pub retry: crate::provider::retry::RetryConfig,
    #[serde(default)]
    pub hardware: HardwareSection,
}

#[derive(Debug, Default, Deserialize)]
pub struct WhisperSection {
    /// Path to whisper model file (e.g. "models/ggml-base.en.bin").
    pub model: Option<String>,
}

#[derive(Debug, Default, Deserialize)]
pub struct TtsSection {
    /// URL of the TTS backend (e.g. "http://localhost:5500" for openedai-speech).
    pub url: Option<String>,
}

#[derive(Default, Deserialize)]
pub struct AuditSection {
    /// Enable audit logging. Defaults to false.
    #[serde(default)]
    pub enabled: bool,
    /// HMAC signing key. Generated randomly if not set.
    pub signing_key: Option<String>,
    /// Max entries to keep in memory. Defaults to 10000.
    #[serde(default = "default_audit_max")]
    pub max_entries: usize,
}

fn default_audit_max() -> usize {
    10_000
}

impl std::fmt::Debug for AuditSection {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("AuditSection")
            .field("enabled", &self.enabled)
            .field(
                "signing_key",
                &self.signing_key.as_ref().map(|_| "[REDACTED]"),
            )
            .field("max_entries", &self.max_entries)
            .finish()
    }
}

/// Context management configuration.
#[derive(Debug, Deserialize)]
pub struct ContextSection {
    /// Token ratio threshold (0.0–1.0) at which context compaction triggers.
    /// Default 0.8 = compact when messages exceed 80% of the model's context window.
    #[serde(default = "default_compaction_threshold")]
    pub compaction_threshold: f64,
    /// Number of most-recent messages to keep when truncating.
    #[serde(default = "default_keep_last")]
    pub keep_last_messages: usize,
    /// Enable context compaction. Defaults to true.
    #[serde(default = "default_true")]
    pub enabled: bool,
}

impl Default for ContextSection {
    fn default() -> Self {
        Self {
            compaction_threshold: default_compaction_threshold(),
            keep_last_messages: default_keep_last(),
            enabled: true,
        }
    }
}

fn default_compaction_threshold() -> f64 {
    0.8
}
fn default_keep_last() -> usize {
    10
}

/// Hardware acceleration configuration.
#[derive(Debug, Deserialize)]
pub struct HardwareSection {
    /// Disk cache TTL in seconds for detection results. Default: 300 (5 min).
    #[serde(default = "default_hw_cache_ttl")]
    pub cache_ttl_secs: u64,
    /// Backend names to disable during detection (e.g. \["vulkan", "tpu"\]).
    #[serde(default)]
    pub disabled_backends: Vec<String>,
    /// VRAM bytes to reserve for non-inference GPU workloads (mabda compute, etc.).
    #[serde(default)]
    pub vram_reserve_bytes: u64,
    /// How often (seconds) to refresh GPU telemetry. Default: 30.
    #[serde(default = "default_hw_refresh")]
    pub refresh_interval_secs: u64,
}

impl Default for HardwareSection {
    fn default() -> Self {
        Self {
            cache_ttl_secs: default_hw_cache_ttl(),
            disabled_backends: Vec::new(),
            vram_reserve_bytes: 0,
            refresh_interval_secs: default_hw_refresh(),
        }
    }
}

fn default_hw_cache_ttl() -> u64 {
    300
}
fn default_hw_refresh() -> u64 {
    30
}

#[derive(Debug, Default, Deserialize)]
pub struct TelemetrySection {
    /// OTLP endpoint (e.g. "http://localhost:4317"). Enables OpenTelemetry when set.
    pub otlp_endpoint: Option<String>,
    /// Service name for traces. Defaults to "hoosh".
    #[serde(default = "default_service_name")]
    pub service_name: String,
}

fn default_service_name() -> String {
    "hoosh".into()
}

#[derive(Default, Deserialize)]
pub struct AuthConfig {
    /// Bearer tokens that are allowed to access the API.
    #[serde(default)]
    pub tokens: Vec<String>,
}

impl std::fmt::Debug for AuthConfig {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("AuthConfig")
            .field("tokens", &format!("[{} configured]", self.tokens.len()))
            .finish()
    }
}

impl std::fmt::Debug for ProviderSection {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("ProviderSection")
            .field("provider_type", &self.provider_type)
            .field("base_url", &self.base_url)
            .field("api_key", &self.api_key.as_ref().map(|_| "[REDACTED]"))
            .field("priority", &self.priority)
            .field("models", &self.models)
            .field("enabled", &self.enabled)
            .finish()
    }
}

#[derive(Debug, Deserialize)]
pub struct BudgetPoolSection {
    /// Pool name (e.g. "default", "agent-1").
    pub name: String,
    /// Maximum tokens allowed in this pool.
    pub capacity: u64,
}

#[derive(Debug, Deserialize)]
pub struct ServerSection {
    #[serde(default = "default_bind")]
    pub bind: String,
    #[serde(default = "default_port")]
    pub port: u16,
    #[serde(default)]
    pub strategy: StrategyValue,
    /// Health check interval in seconds. 0 = disabled. Defaults to 30.
    #[serde(default = "default_health_interval")]
    pub health_check_interval_secs: u64,
}

impl Default for ServerSection {
    fn default() -> Self {
        Self {
            bind: default_bind(),
            port: default_port(),
            strategy: StrategyValue::default(),
            health_check_interval_secs: default_health_interval(),
        }
    }
}

impl From<StrategyValue> for RoutingStrategy {
    fn from(v: StrategyValue) -> Self {
        match v {
            StrategyValue::Priority => RoutingStrategy::Priority,
            StrategyValue::RoundRobin => RoutingStrategy::RoundRobin,
            StrategyValue::LowestLatency => RoutingStrategy::LowestLatency,
            StrategyValue::Direct => RoutingStrategy::Direct,
        }
    }
}

#[derive(Debug, Default, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum StrategyValue {
    #[default]
    Priority,
    RoundRobin,
    LowestLatency,
    Direct,
}

#[derive(Debug, Deserialize)]
pub struct CacheSection {
    #[serde(default = "default_cache_max")]
    pub max_entries: usize,
    #[serde(default = "default_cache_ttl")]
    pub ttl_secs: u64,
    #[serde(default = "default_true")]
    pub enabled: bool,
}

#[derive(Deserialize)]
pub struct ProviderSection {
    /// Provider type name (e.g. "ollama", "openai", "anthropic").
    #[serde(rename = "type")]
    pub provider_type: ProviderType,
    /// Base URL. Uses provider-specific default if omitted.
    pub base_url: Option<String>,
    /// API key — literal string or `"$ENV_VAR"` to read from environment.
    pub api_key: Option<String>,
    /// Priority (lower = preferred). Defaults to 10.
    #[serde(default = "default_priority")]
    pub priority: u32,
    /// Model patterns this provider handles.
    #[serde(default)]
    pub models: Vec<String>,
    /// Whether this provider is enabled. Defaults to true.
    #[serde(default = "default_true")]
    pub enabled: bool,
    /// Maximum tokens per request for this provider.
    #[serde(default)]
    pub max_tokens_limit: Option<u32>,
    /// Maximum requests per minute for this provider.
    #[serde(default)]
    pub rate_limit_rpm: Option<u32>,
    /// Paths to PEM certificates to pin for this provider's TLS.
    #[serde(default)]
    pub tls_pinned_certs: Vec<String>,
    /// Path to client certificate for mTLS.
    pub client_cert: Option<String>,
    /// Path to client key for mTLS.
    pub client_key: Option<String>,
}

fn default_bind() -> String {
    "127.0.0.1".into()
}
fn default_port() -> u16 {
    8088
}
fn default_cache_max() -> usize {
    1000
}
fn default_cache_ttl() -> u64 {
    300
}
fn default_true() -> bool {
    true
}
fn default_priority() -> u32 {
    10
}
fn default_health_interval() -> u64 {
    30
}

impl Default for CacheSection {
    fn default() -> Self {
        Self {
            max_entries: default_cache_max(),
            ttl_secs: default_cache_ttl(),
            enabled: true,
        }
    }
}

/// Resolve an API key value. If it starts with `$`, read from environment.
fn resolve_api_key(raw: &Option<String>) -> Option<String> {
    let raw = raw.as_ref()?;
    if let Some(var_name) = raw.strip_prefix('$') {
        match std::env::var(var_name) {
            Ok(val) => Some(val),
            Err(_) => {
                tracing::warn!(
                    "API key env var ${var_name} is not set — provider will have no API key"
                );
                None
            }
        }
    } else {
        Some(raw.clone())
    }
}

/// Default base URL for a provider type.
fn default_base_url(provider_type: ProviderType) -> &'static str {
    match provider_type {
        ProviderType::Ollama => "http://localhost:11434",
        ProviderType::LlamaCpp => "http://localhost:8080",
        ProviderType::Synapse => "http://localhost:5000",
        ProviderType::LmStudio => "http://localhost:1234",
        ProviderType::LocalAi => "http://localhost:8080",
        ProviderType::OpenAi => "https://api.openai.com",
        ProviderType::Anthropic => "https://api.anthropic.com",
        ProviderType::DeepSeek => "https://api.deepseek.com",
        ProviderType::Mistral => "https://api.mistral.ai",
        ProviderType::Groq => "https://api.groq.com/openai",
        ProviderType::OpenRouter => "https://openrouter.ai/api",
        ProviderType::Google => "https://generativelanguage.googleapis.com",
        ProviderType::Grok => "https://api.x.ai",
        ProviderType::Whisper => "http://localhost:8080",
    }
}

impl HooshConfig {
    /// Load configuration from a TOML file.
    pub fn load(path: impl AsRef<Path>) -> anyhow::Result<Self> {
        let contents = std::fs::read_to_string(path.as_ref())?;
        let config: HooshConfig = toml::from_str(&contents).map_err(|e| {
            // Redact error details that might contain API keys
            let msg = e.to_string();
            if msg.contains("api_key") {
                anyhow::anyhow!("failed to parse config: TOML syntax error near api_key field")
            } else {
                anyhow::anyhow!("failed to parse config: {e}")
            }
        })?;
        Ok(config)
    }

    /// Try loading from `hoosh.toml` in the current directory, or return defaults.
    pub fn load_or_default() -> Self {
        if Path::new("hoosh.toml").exists() {
            match Self::load("hoosh.toml") {
                Ok(config) => {
                    tracing::info!("loaded config from hoosh.toml");
                    config
                }
                Err(e) => {
                    tracing::error!("failed to load hoosh.toml: {e}");
                    std::process::exit(1);
                }
            }
        } else {
            Self {
                server: ServerSection::default(),
                cache: CacheSection::default(),
                providers: Vec::new(),
                budgets: Vec::new(),
                whisper: WhisperSection::default(),
                tts: TtsSection::default(),
                audit: AuditSection::default(),
                auth: AuthConfig::default(),
                telemetry: TelemetrySection::default(),
                context: ContextSection::default(),
                retry: crate::provider::retry::RetryConfig::default(),
                hardware: HardwareSection::default(),
            }
        }
    }

    /// Convert to provider routes.
    pub fn routes(&self) -> Vec<ProviderRoute> {
        self.providers
            .iter()
            .map(|p| {
                let base_url = p
                    .base_url
                    .clone()
                    .unwrap_or_else(|| default_base_url(p.provider_type).into());
                let tls_config = if !p.tls_pinned_certs.is_empty()
                    || p.client_cert.is_some()
                    || p.client_key.is_some()
                {
                    Some(crate::provider::TlsConfig {
                        pinned_certs: p.tls_pinned_certs.clone(),
                        client_cert: p.client_cert.clone(),
                        client_key: p.client_key.clone(),
                    })
                } else {
                    None
                };
                ProviderRoute {
                    provider: p.provider_type,
                    priority: p.priority,
                    model_patterns: p.models.clone(),
                    enabled: p.enabled,
                    base_url,
                    api_key: resolve_api_key(&p.api_key),
                    max_tokens_limit: p.max_tokens_limit,
                    rate_limit_rpm: p.rate_limit_rpm,
                    tls_config,
                }
            })
            .collect()
    }

    /// Convert to ServerConfig, merging CLI overrides.
    pub fn into_server_config(
        self,
        bind_override: Option<&str>,
        port_override: Option<u16>,
        config_path: Option<String>,
    ) -> ServerConfig {
        let routes = self.routes();
        let strategy: RoutingStrategy = self.server.strategy.into();
        let budget_pools = self
            .budgets
            .iter()
            .map(|b| TokenPool::new(&b.name, b.capacity))
            .collect();

        ServerConfig {
            bind: bind_override.map(String::from).unwrap_or(self.server.bind),
            port: port_override.unwrap_or(self.server.port),
            routes,
            strategy,
            cache_config: CacheConfig {
                max_entries: self.cache.max_entries,
                ttl_secs: self.cache.ttl_secs,
                enabled: self.cache.enabled,
            },
            budget_pools,
            whisper_model: self.whisper.model,
            tts_model: self.tts.url,
            audit_enabled: self.audit.enabled,
            audit_signing_key: resolve_api_key(&self.audit.signing_key),
            audit_max_entries: self.audit.max_entries,
            auth_tokens: self.auth.tokens,
            otlp_endpoint: self.telemetry.otlp_endpoint,
            telemetry_service_name: self.telemetry.service_name,
            health_check_interval_secs: self.server.health_check_interval_secs,
            config_path,
            context_config: self.context,
            retry_config: self.retry,
            #[cfg(feature = "hwaccel")]
            hardware_config: self.hardware,
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn parse_minimal_config() {
        let toml = "";
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.server.port, 8088);
        assert_eq!(config.server.bind, "127.0.0.1");
        assert!(config.providers.is_empty());
    }

    #[test]
    fn parse_full_config() {
        let toml = r#"
[server]
bind = "0.0.0.0"
port = 9000
strategy = "round_robin"

[cache]
max_entries = 500
ttl_secs = 600
enabled = false

[[providers]]
type = "Ollama"
base_url = "http://gpu-box:11434"
priority = 1
models = ["llama*", "mistral*"]

[[providers]]
type = "OpenAi"
api_key = "$OPENAI_API_KEY"
priority = 10
models = ["gpt-*"]
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.server.port, 9000);
        assert_eq!(config.server.bind, "0.0.0.0");
        assert_eq!(config.cache.max_entries, 500);
        assert!(!config.cache.enabled);
        assert_eq!(config.providers.len(), 2);
        assert_eq!(config.providers[0].provider_type, ProviderType::Ollama);
        assert_eq!(config.providers[1].provider_type, ProviderType::OpenAi);
        assert_eq!(
            config.providers[1].api_key.as_deref(),
            Some("$OPENAI_API_KEY")
        );
    }

    #[test]
    fn routes_from_config() {
        let toml = r#"
[[providers]]
type = "Ollama"
priority = 1
models = ["llama*"]

[[providers]]
type = "OpenAi"
api_key = "sk-test-key"
priority = 5
models = ["gpt-*"]
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let routes = config.routes();
        assert_eq!(routes.len(), 2);
        assert_eq!(routes[0].base_url, "http://localhost:11434");
        assert!(routes[0].api_key.is_none());
        assert_eq!(routes[1].base_url, "https://api.openai.com");
        assert_eq!(routes[1].api_key.as_deref(), Some("sk-test-key"));
    }

    #[test]
    fn resolve_api_key_literal() {
        let key = Some("sk-literal".into());
        assert_eq!(resolve_api_key(&key).as_deref(), Some("sk-literal"));
    }

    #[test]
    fn resolve_api_key_env_var() {
        // SAFETY: test is single-threaded, no concurrent env access
        unsafe { std::env::set_var("HOOSH_TEST_KEY_1234", "sk-from-env") };
        let key = Some("$HOOSH_TEST_KEY_1234".into());
        assert_eq!(resolve_api_key(&key).as_deref(), Some("sk-from-env"));
        unsafe { std::env::remove_var("HOOSH_TEST_KEY_1234") };
    }

    #[test]
    fn resolve_api_key_missing_env() {
        let key = Some("$HOOSH_NONEXISTENT_KEY_999".into());
        assert!(resolve_api_key(&key).is_none());
    }

    #[test]
    fn resolve_api_key_none() {
        assert!(resolve_api_key(&None).is_none());
    }

    #[test]
    fn default_base_urls() {
        assert_eq!(
            default_base_url(ProviderType::Ollama),
            "http://localhost:11434"
        );
        assert_eq!(
            default_base_url(ProviderType::OpenAi),
            "https://api.openai.com"
        );
        assert_eq!(
            default_base_url(ProviderType::Anthropic),
            "https://api.anthropic.com"
        );
        assert_eq!(
            default_base_url(ProviderType::Groq),
            "https://api.groq.com/openai"
        );
    }

    #[test]
    fn into_server_config_with_overrides() {
        let toml = r#"
[server]
port = 9000
bind = "0.0.0.0"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let sc = config.into_server_config(Some("127.0.0.1"), Some(8080), None);
        assert_eq!(sc.bind, "127.0.0.1");
        assert_eq!(sc.port, 8080);
    }

    #[test]
    fn into_server_config_no_overrides() {
        let toml = r#"
[server]
port = 9000
bind = "0.0.0.0"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let sc = config.into_server_config(None, None, None);
        assert_eq!(sc.bind, "0.0.0.0");
        assert_eq!(sc.port, 9000);
    }

    #[test]
    fn parse_with_budgets() {
        let toml = r#"
[[budgets]]
name = "default"
capacity = 100000

[[budgets]]
name = "agents"
capacity = 50000
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.budgets.len(), 2);
        assert_eq!(config.budgets[0].name, "default");
        assert_eq!(config.budgets[0].capacity, 100000);
    }

    #[test]
    fn parse_with_whisper_and_tts() {
        let toml = r#"
[whisper]
model = "models/ggml-base.en.bin"

[tts]
url = "http://localhost:5500"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(
            config.whisper.model.as_deref(),
            Some("models/ggml-base.en.bin")
        );
        assert_eq!(config.tts.url.as_deref(), Some("http://localhost:5500"));
    }

    #[test]
    fn into_server_config_with_budgets() {
        let toml = r#"
[[budgets]]
name = "pool1"
capacity = 5000
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let sc = config.into_server_config(None, None, None);
        assert_eq!(sc.budget_pools.len(), 1);
        assert_eq!(sc.budget_pools[0].name, "pool1");
        assert_eq!(sc.budget_pools[0].capacity, 5000);
    }

    #[test]
    fn into_server_config_with_whisper_tts() {
        let toml = r#"
[whisper]
model = "model.bin"

[tts]
url = "http://tts:5500"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let sc = config.into_server_config(None, None, None);
        assert_eq!(sc.whisper_model.as_deref(), Some("model.bin"));
        assert_eq!(sc.tts_model.as_deref(), Some("http://tts:5500"));
    }

    #[test]
    fn into_server_config_all_strategies() {
        for (strategy_str, _) in [
            ("priority", "Priority"),
            ("round_robin", "RoundRobin"),
            ("lowest_latency", "LowestLatency"),
            ("direct", "Direct"),
        ] {
            let toml = format!("[server]\nstrategy = \"{strategy_str}\"");
            let config: HooshConfig = toml::from_str(&toml).unwrap();
            let sc = config.into_server_config(None, None, None);
            // Just verify it doesn't panic
            let _ = sc.strategy;
        }
    }

    #[test]
    fn all_default_base_urls_covered() {
        let types = [
            (ProviderType::Ollama, "http://localhost:11434"),
            (ProviderType::LlamaCpp, "http://localhost:8080"),
            (ProviderType::Synapse, "http://localhost:5000"),
            (ProviderType::LmStudio, "http://localhost:1234"),
            (ProviderType::LocalAi, "http://localhost:8080"),
            (ProviderType::OpenAi, "https://api.openai.com"),
            (ProviderType::Anthropic, "https://api.anthropic.com"),
            (ProviderType::DeepSeek, "https://api.deepseek.com"),
            (ProviderType::Mistral, "https://api.mistral.ai"),
            (ProviderType::Groq, "https://api.groq.com/openai"),
            (ProviderType::OpenRouter, "https://openrouter.ai/api"),
            (
                ProviderType::Google,
                "https://generativelanguage.googleapis.com",
            ),
            (ProviderType::Grok, "https://api.x.ai"),
            (ProviderType::Whisper, "http://localhost:8080"),
        ];
        for (pt, expected) in types {
            assert_eq!(default_base_url(pt), expected, "mismatch for {pt}");
        }
    }

    #[test]
    fn provider_with_max_tokens_limit() {
        let toml = r#"
[[providers]]
type = "OpenAi"
api_key = "sk-test"
max_tokens_limit = 4096
models = ["gpt-*"]
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.providers[0].max_tokens_limit, Some(4096));
        let routes = config.routes();
        assert_eq!(routes[0].max_tokens_limit, Some(4096));
    }

    #[test]
    fn provider_defaults() {
        let toml = r#"
[[providers]]
type = "Ollama"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let p = &config.providers[0];
        assert_eq!(p.priority, 10);
        assert!(p.enabled);
        assert!(p.models.is_empty());
        assert!(p.base_url.is_none());
    }

    #[test]
    fn routes_with_tls_config() {
        let toml = r#"
[[providers]]
type = "OpenAi"
api_key = "sk-test"
models = ["gpt-*"]
tls_pinned_certs = ["/path/to/cert.pem"]
client_cert = "/path/to/client.pem"
client_key = "/path/to/client-key.pem"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let routes = config.routes();
        assert_eq!(routes.len(), 1);
        let tls = routes[0].tls_config.as_ref().unwrap();
        assert_eq!(tls.pinned_certs.len(), 1);
        assert_eq!(tls.client_cert.as_deref(), Some("/path/to/client.pem"));
        assert_eq!(tls.client_key.as_deref(), Some("/path/to/client-key.pem"));
    }

    #[test]
    fn routes_without_tls_config() {
        let toml = r#"
[[providers]]
type = "Ollama"
models = ["llama*"]
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let routes = config.routes();
        assert!(routes[0].tls_config.is_none());
    }

    #[test]
    fn routes_with_rate_limit() {
        let toml = r#"
[[providers]]
type = "OpenAi"
api_key = "sk-test"
models = ["gpt-*"]
rate_limit_rpm = 60
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let routes = config.routes();
        assert_eq!(routes[0].rate_limit_rpm, Some(60));
    }

    #[test]
    fn routes_disabled_provider() {
        let toml = r#"
[[providers]]
type = "Ollama"
enabled = false
models = ["llama*"]
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let routes = config.routes();
        assert!(!routes[0].enabled);
    }

    #[test]
    fn parse_audit_section() {
        let toml = r#"
[audit]
enabled = true
signing_key = "my-secret-key"
max_entries = 5000
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert!(config.audit.enabled);
        assert_eq!(config.audit.signing_key.as_deref(), Some("my-secret-key"));
        assert_eq!(config.audit.max_entries, 5000);
    }

    #[test]
    fn parse_audit_defaults() {
        let config: HooshConfig = toml::from_str("").unwrap();
        assert!(!config.audit.enabled);
        assert!(config.audit.signing_key.is_none());
        // AuditSection default() uses #[serde(default = "default_audit_max")] = 10_000
        // but when deserialized from empty TOML, [audit] section is absent, so Default is used
    }

    #[test]
    fn parse_context_section() {
        let toml = r#"
[context]
compaction_threshold = 0.6
keep_last_messages = 5
enabled = false
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert!((config.context.compaction_threshold - 0.6).abs() < f64::EPSILON);
        assert_eq!(config.context.keep_last_messages, 5);
        assert!(!config.context.enabled);
    }

    #[test]
    fn parse_context_defaults() {
        let config: HooshConfig = toml::from_str("").unwrap();
        assert!((config.context.compaction_threshold - 0.8).abs() < f64::EPSILON);
        assert_eq!(config.context.keep_last_messages, 10);
        assert!(config.context.enabled);
    }

    #[test]
    fn parse_telemetry_section() {
        let toml = r#"
[telemetry]
otlp_endpoint = "http://localhost:4317"
service_name = "my-hoosh"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(
            config.telemetry.otlp_endpoint.as_deref(),
            Some("http://localhost:4317")
        );
        assert_eq!(config.telemetry.service_name, "my-hoosh");
    }

    #[test]
    fn parse_telemetry_defaults() {
        let config: HooshConfig = toml::from_str("").unwrap();
        assert!(config.telemetry.otlp_endpoint.is_none());
    }

    #[test]
    fn parse_telemetry_with_service_name() {
        let toml = r#"
[telemetry]
service_name = "hoosh"
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.telemetry.service_name, "hoosh");
    }

    #[test]
    fn parse_auth_section() {
        let toml = r#"
[auth]
tokens = ["token1", "token2"]
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.auth.tokens.len(), 2);
    }

    #[test]
    fn parse_retry_section() {
        let toml = r#"
[retry]
max_retries = 5
base_delay_ms = 1000
max_delay_ms = 60000
jitter_factor = 0.3
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.retry.max_retries, 5);
        assert_eq!(config.retry.base_delay_ms, 1000);
        assert_eq!(config.retry.max_delay_ms, 60_000);
    }

    #[test]
    fn into_server_config_full() {
        let toml = r#"
[server]
port = 9000
bind = "0.0.0.0"
strategy = "lowest_latency"
health_check_interval_secs = 60

[cache]
max_entries = 500
ttl_secs = 600
enabled = false

[audit]
enabled = true
max_entries = 5000

[telemetry]
otlp_endpoint = "http://otel:4317"
service_name = "test-hoosh"

[auth]
tokens = ["tok1"]

[context]
compaction_threshold = 0.9
keep_last_messages = 20

[retry]
max_retries = 5
base_delay_ms = 500
max_delay_ms = 30000
jitter_factor = 0.5

[[budgets]]
name = "default"
capacity = 100000

[[providers]]
type = "Ollama"
priority = 1
models = ["llama*"]
rate_limit_rpm = 120
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        let sc = config.into_server_config(None, None, Some("/path/to/config.toml".into()));
        assert_eq!(sc.port, 9000);
        assert_eq!(sc.bind, "0.0.0.0");
        assert_eq!(sc.cache_config.max_entries, 500);
        assert!(!sc.cache_config.enabled);
        assert!(sc.audit_enabled);
        assert_eq!(sc.audit_max_entries, 5000);
        assert_eq!(sc.otlp_endpoint.as_deref(), Some("http://otel:4317"));
        assert_eq!(sc.telemetry_service_name, "test-hoosh");
        assert_eq!(sc.auth_tokens.len(), 1);
        assert_eq!(sc.health_check_interval_secs, 60);
        assert_eq!(sc.config_path.as_deref(), Some("/path/to/config.toml"));
        assert!((sc.context_config.compaction_threshold - 0.9).abs() < f64::EPSILON);
        assert_eq!(sc.context_config.keep_last_messages, 20);
        assert_eq!(sc.retry_config.max_retries, 5);
        assert_eq!(sc.budget_pools.len(), 1);
        assert_eq!(sc.routes.len(), 1);
    }

    #[test]
    fn audit_section_debug_redacts_key() {
        let section = AuditSection {
            enabled: true,
            signing_key: Some("super-secret".into()),
            max_entries: 1000,
        };
        let debug = format!("{section:?}");
        assert!(!debug.contains("super-secret"));
        assert!(debug.contains("[REDACTED]"));
    }

    #[test]
    fn auth_config_debug_shows_count() {
        let auth = AuthConfig {
            tokens: vec!["tok1".into(), "tok2".into()],
        };
        let debug = format!("{auth:?}");
        assert!(debug.contains("2 configured"));
        assert!(!debug.contains("tok1"));
    }

    #[test]
    fn provider_section_debug_redacts_key() {
        let section: ProviderSection = toml::from_str(
            r#"
type = "OpenAi"
api_key = "sk-secret-key"
models = ["gpt-*"]
"#,
        )
        .unwrap();
        let debug = format!("{section:?}");
        assert!(!debug.contains("sk-secret-key"));
        assert!(debug.contains("[REDACTED]"));
    }

    #[test]
    fn load_nonexistent_config_file() {
        let result = HooshConfig::load("/nonexistent/path/hoosh.toml");
        assert!(result.is_err());
    }

    #[test]
    fn load_invalid_toml() {
        let dir = std::env::temp_dir();
        let path = dir.join("hoosh_test_invalid.toml");
        std::fs::write(&path, "invalid {{{{ toml content").unwrap();
        let result = HooshConfig::load(&path);
        assert!(result.is_err());
        let _ = std::fs::remove_file(&path);
    }

    #[test]
    fn context_section_default() {
        let ctx = ContextSection::default();
        assert!((ctx.compaction_threshold - 0.8).abs() < f64::EPSILON);
        assert_eq!(ctx.keep_last_messages, 10);
        assert!(ctx.enabled);
    }

    #[test]
    fn server_section_default() {
        let s = ServerSection::default();
        assert_eq!(s.bind, "127.0.0.1");
        assert_eq!(s.port, 8088);
        assert_eq!(s.health_check_interval_secs, 30);
    }

    #[test]
    fn cache_section_default() {
        let c = CacheSection::default();
        assert_eq!(c.max_entries, 1000);
        assert_eq!(c.ttl_secs, 300);
        assert!(c.enabled);
    }

    #[test]
    fn hardware_section_default() {
        let h = HardwareSection::default();
        assert_eq!(h.cache_ttl_secs, 300);
        assert!(h.disabled_backends.is_empty());
        assert_eq!(h.vram_reserve_bytes, 0);
        assert_eq!(h.refresh_interval_secs, 30);
    }

    #[test]
    fn parse_hardware_section() {
        let toml = r#"
[hardware]
cache_ttl_secs = 600
disabled_backends = ["vulkan", "tpu"]
vram_reserve_bytes = 2147483648
refresh_interval_secs = 60
"#;
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.hardware.cache_ttl_secs, 600);
        assert_eq!(config.hardware.disabled_backends, vec!["vulkan", "tpu"]);
        assert_eq!(config.hardware.vram_reserve_bytes, 2_147_483_648);
        assert_eq!(config.hardware.refresh_interval_secs, 60);
    }

    #[test]
    fn parse_hardware_section_empty_defaults() {
        let toml = "[hardware]\n";
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.hardware.cache_ttl_secs, 300);
        assert!(config.hardware.disabled_backends.is_empty());
        assert_eq!(config.hardware.vram_reserve_bytes, 0);
        assert_eq!(config.hardware.refresh_interval_secs, 30);
    }

    #[test]
    fn parse_no_hardware_section() {
        let toml = "[server]\nport = 9000\n";
        let config: HooshConfig = toml::from_str(toml).unwrap();
        assert_eq!(config.hardware.cache_ttl_secs, 300); // default
    }
}