apr-cli 0.60.0

//! HTTP route definitions and middleware for APR serve command

#![allow(unused_imports)]
#![allow(unused_variables)]

use super::health_check;
use super::types::{
    ErrorResponse, GenerateRequest, GenerateResponse, HealthResponse, HealthStatus, ServerInfo,
    ServerState, StreamEvent, TranscribeResponse, MAX_REQUEST_SIZE,
};
use std::sync::Arc;
use std::time::Instant;

#[cfg(feature = "inference")]
use axum::{
    body::Body,
    extract::{Request, State},
    http::{header, Method, StatusCode},
    middleware::{self, Next},
    response::{IntoResponse, Response, Sse},
    routing::{get, post},
    Json, Router,
};

/// Middleware: Request size limit (SE02, EH05)
#[cfg(feature = "inference")]
async fn size_limit_middleware(request: Request, next: Next) -> Response {
    let content_length = request
        .headers()
        .get(header::CONTENT_LENGTH)
        .and_then(|v| v.to_str().ok())
        .and_then(|v| v.parse::<usize>().ok())
        .unwrap_or(0);

    if content_length > MAX_REQUEST_SIZE {
        return (
            StatusCode::PAYLOAD_TOO_LARGE,
            Json(ErrorResponse::new(
                "payload_too_large",
                format!("Request body exceeds {} bytes limit", MAX_REQUEST_SIZE),
            )),
        )
            .into_response();
    }

    next.run(request).await
}

/// Handler: GET / (SL09: Root endpoint returns semver)
#[cfg(feature = "inference")]
async fn root_handler(State(state): State<Arc<ServerState>>) -> Json<ServerInfo> {
    Json(ServerInfo {
        name: "apr-serve".to_string(),
        version: env!("CARGO_PKG_VERSION").to_string(),
        model_id: state.model_id.clone(),
    })
}

/// Handler: GET /health (HR01-HR10)
#[cfg(feature = "inference")]
async fn health_handler(
    State(state): State<Arc<ServerState>>,
) -> (StatusCode, Json<HealthResponse>) {
    contract_pre_timeout_honoring!();
    let health = health_check(&state);

    if state.config.verbose {
        eprintln!(
            "[VERBOSE] GET /health: status={:?}, uptime={}s",
            health.status, health.uptime_seconds
        );
    }

    let status_code = match health.status {
        HealthStatus::Healthy | HealthStatus::Degraded => StatusCode::OK,
        HealthStatus::Unhealthy => StatusCode::SERVICE_UNAVAILABLE,
    };

    contract_post_timeout_honoring!(&health);
    (status_code, Json(health))
}

/// Handler: GET /metrics (MA01-MA10)
#[cfg(feature = "inference")]
async fn metrics_handler(State(state): State<Arc<ServerState>>) -> impl IntoResponse {
    (
        StatusCode::OK,
        [("content-type", "text/plain; charset=utf-8")],
        state.metrics.prometheus_output(),
    )
}

/// Validate request body size and parse JSON (EH01, EH05).
#[cfg(feature = "inference")]
#[allow(clippy::result_large_err)]
fn validate_and_parse<T: serde::de::DeserializeOwned>(
    body: &[u8],
    metrics: &super::types::ServerMetrics,
) -> std::result::Result<T, Response> {
    if body.len() > MAX_REQUEST_SIZE {
        metrics.record_client_error();
        return Err((
            StatusCode::PAYLOAD_TOO_LARGE,
            Json(ErrorResponse::new(
                "payload_too_large",
                "Request body too large",
            )),
        )
            .into_response());
    }
    serde_json::from_slice(body).map_err(|e| {
        metrics.record_client_error();
        (
            StatusCode::BAD_REQUEST,
            Json(ErrorResponse::new(
                "invalid_json",
                format!("Invalid JSON: {e}"),
            )),
        )
            .into_response()
    })
}

/// Handler: POST /predict (IC01-IC15)
#[cfg(feature = "inference")]
#[allow(clippy::disallowed_methods)] // serde_json::json!() macro uses infallible unwrap
async fn predict_handler(
    State(state): State<Arc<ServerState>>,
    body: axum::body::Bytes,
) -> impl IntoResponse {
    contract_pre_request_response_schema!();
    let start = Instant::now();

    let request: serde_json::Value = match validate_and_parse(&body, &state.metrics) {
        Ok(v) => v,
        Err(resp) => return resp,
    };

    if request.get("inputs").is_none() {
        state.metrics.record_client_error();
        return (
            StatusCode::BAD_REQUEST,
            Json(ErrorResponse::new(
                "missing_field",
                "Missing required field: inputs",
            )),
        )
            .into_response();
    }

    let duration_ms = start.elapsed().as_millis() as u64;
    state.metrics.record_request(true, 0, duration_ms);

    contract_post_request_response_schema!(&());
    (
        StatusCode::OK,
        Json(serde_json::json!({
            "outputs": {},
            "latency_ms": duration_ms
        })),
    )
        .into_response()
}

/// Handler: POST /generate with SSE streaming (SP01-SP10)
#[cfg(feature = "inference")]
async fn generate_handler(
    State(state): State<Arc<ServerState>>,
    body: axum::body::Bytes,
) -> Response {
    let start = Instant::now();

    let request: GenerateRequest = match validate_and_parse(&body, &state.metrics) {
        Ok(v) => v,
        Err(resp) => return resp,
    };

    if state.config.verbose {
        log_generate_request(&request);
    }

    if request.prompt.is_empty() {
        state.metrics.record_client_error();
        return (
            StatusCode::BAD_REQUEST,
            Json(ErrorResponse::new("empty_prompt", "Prompt cannot be empty")),
        )
            .into_response();
    }

    if request.stream {
        return generate_streaming(&state, start);
    }

    generate_non_streaming(&state, start)
}

/// Log a verbose generate request preview
#[cfg(feature = "inference")]
fn log_generate_request(request: &GenerateRequest) {
    let prompt_preview = if request.prompt.len() > 100 {
        format!("{}...", &request.prompt[..100])
    } else {
        request.prompt.clone()
    };
    eprintln!(
        "[VERBOSE] POST /generate: prompt={:?}, max_tokens={}, stream={}",
        prompt_preview, request.max_tokens, request.stream
    );
}

/// Build SSE streaming response for /generate (SP01-SP10)
#[cfg(feature = "inference")]
fn generate_streaming(state: &Arc<ServerState>, start: Instant) -> Response {
    use futures_util::stream;
    use std::convert::Infallible;

    let metrics = state.metrics.clone();

    let stream = stream::iter((0..3).map(move |i| {
        let event = if i < 2 {
            StreamEvent::token(&format!("token{}", i), i)
        } else {
            StreamEvent::done("stop", 2)
        };
        Ok::<_, Infallible>(
            axum::response::sse::Event::default()
                .event(&event.event)
                .data(&event.data),
        )
    }));

    let duration_ms = start.elapsed().as_millis() as u64;
    metrics.record_request(true, 2, duration_ms);

    if state.config.verbose {
        eprintln!(
            "[VERBOSE] POST /generate streaming: started, latency_ms={}",
            duration_ms
        );
    }

    Sse::new(stream)
        .keep_alive(axum::response::sse::KeepAlive::default())
        .into_response()
}

/// Build non-streaming response for /generate (LG03)
#[cfg(feature = "inference")]
fn generate_non_streaming(state: &Arc<ServerState>, start: Instant) -> Response {
    let duration_ms = start.elapsed().as_millis() as u64;
    state.metrics.record_request(true, 0, duration_ms);

    if state.config.verbose {
        eprintln!(
            "[VERBOSE] POST /generate response: tokens=0, latency_ms={}, finish_reason=stop",
            duration_ms
        );
    }

    (
        StatusCode::OK,
        Json(GenerateResponse {
            text: String::new(),
            tokens_generated: 0,
            finish_reason: "stop".to_string(),
            latency_ms: duration_ms,
        }),
    )
        .into_response()
}

/// Handler: POST /transcribe (audio transcription)
#[cfg(feature = "inference")]
async fn transcribe_handler(
    State(state): State<Arc<ServerState>>,
    body: axum::body::Bytes,
) -> impl IntoResponse {
    let start = Instant::now();

    if body.len() > MAX_REQUEST_SIZE {
        state.metrics.record_client_error();
        return (
            StatusCode::PAYLOAD_TOO_LARGE,
            Json(ErrorResponse::new(
                "payload_too_large",
                "Request body too large",
            )),
        )
            .into_response();
    }

    let duration_ms = start.elapsed().as_millis() as u64;
    state.metrics.record_request(true, 0, duration_ms);

    (
        StatusCode::OK,
        Json(TranscribeResponse {
            text: String::new(),
            language: "en".to_string(),
            duration_seconds: 0.0,
            latency_ms: duration_ms,
        }),
    )
        .into_response()
}

/// Handler: Method not allowed (EH04)
#[cfg(feature = "inference")]
async fn method_not_allowed(State(state): State<Arc<ServerState>>) -> impl IntoResponse {
    state.metrics.record_client_error();
    (
        StatusCode::METHOD_NOT_ALLOWED,
        Json(ErrorResponse::new(
            "method_not_allowed",
            "Method not allowed for this endpoint",
        )),
    )
}

/// Handler: 404 for unknown endpoints (EH03)
#[cfg(feature = "inference")]
async fn fallback_handler(State(state): State<Arc<ServerState>>) -> impl IntoResponse {
    contract_pre_error_envelope_preservation!();
    state.metrics.record_client_error();
    let result = ErrorResponse::new("not_found", "Endpoint not found");
    contract_post_error_envelope_preservation!(&result);
    (StatusCode::NOT_FOUND, Json(result))
}

/// Create the inference server router
///
/// This function creates an axum Router that can be used for both production
/// and testing. All endpoints implement APR-SPEC §4.15.8.3 REST API spec.
///
/// HELIX-IDEA-009 / FALSIFY-AUTH-001: when `auth_gate.is_enabled()`, every
/// route is gated by the `Authorization: Bearer <key>` middleware. Calling
/// with `AuthGate::disabled()` produces an unauthenticated router (used by
/// the `--auth-disabled` startup path and most tests).
#[cfg(feature = "inference")]
pub fn create_router(state: Arc<ServerState>) -> axum::Router {
    create_router_with_auth(state, super::auth::AuthGate::disabled())
}

/// Same as [`create_router`] but takes an explicit [`super::auth::AuthGate`].
/// Production startup code passes `AuthGate::from_env()`; tests use
/// `AuthGate::from_plain_key("...")` to drive the FALSIFY-AUTH-001 / 002
/// gates without going through env vars.
#[cfg(feature = "inference")]
pub fn create_router_with_auth(
    state: Arc<ServerState>,
    auth_gate: super::auth::AuthGate,
) -> axum::Router {
    let router = Router::new()
        .route("/", get(root_handler))
        .route("/health", get(health_handler))
        .route("/metrics", get(metrics_handler))
        .route("/predict", post(predict_handler))
        .route("/generate", post(generate_handler))
        .route("/transcribe", post(transcribe_handler))
        .route("/predict", get(method_not_allowed))
        .route("/generate", get(method_not_allowed))
        .route("/transcribe", get(method_not_allowed))
        .layer(middleware::from_fn(size_limit_middleware))
        .fallback(fallback_handler)
        .with_state(state);
    super::auth::layer(auth_gate, router)
}

// ============================================================================
// PMAT-125 B3: unit tests for the request-body validation seam.
// ============================================================================
#[cfg(all(test, feature = "inference"))]
mod routes_tests {
    use super::super::types::{GenerateRequest, ServerMetrics};
    use super::*;

    #[test]
    fn validate_and_parse_success_parses_typed_body() {
        let metrics = ServerMetrics::new();
        let body = br#"{"prompt": "hi", "max_tokens": 8}"#;
        let parsed: GenerateRequest =
            validate_and_parse(body, &metrics).expect("valid JSON parses");
        assert_eq!(parsed.prompt, "hi");
        assert_eq!(parsed.max_tokens, 8);
        // No client error recorded on success.
        assert_eq!(
            metrics
                .requests_client_error
                .load(std::sync::atomic::Ordering::Relaxed),
            0
        );
    }

    #[tokio::test]
    async fn validate_and_parse_oversized_body_is_413_and_counts_error() {
        let metrics = ServerMetrics::new();
        let big = vec![b'x'; MAX_REQUEST_SIZE + 1];
        let result: std::result::Result<serde_json::Value, _> = validate_and_parse(&big, &metrics);
        let resp = result.err().expect("oversized body must be rejected");
        assert_eq!(resp.status(), StatusCode::PAYLOAD_TOO_LARGE);
        assert_eq!(
            metrics
                .requests_client_error
                .load(std::sync::atomic::Ordering::Relaxed),
            1,
            "client error counter incremented"
        );
    }

    #[tokio::test]
    async fn validate_and_parse_invalid_json_is_400_and_counts_error() {
        let metrics = ServerMetrics::new();
        let body = b"{not valid json";
        let result: std::result::Result<GenerateRequest, _> = validate_and_parse(body, &metrics);
        let resp = result.err().expect("invalid JSON must be rejected");
        assert_eq!(resp.status(), StatusCode::BAD_REQUEST);
        assert_eq!(
            metrics
                .requests_client_error
                .load(std::sync::atomic::Ordering::Relaxed),
            1
        );
    }

    #[test]
    fn validate_and_parse_at_size_limit_boundary_still_parses() {
        // Exactly MAX_REQUEST_SIZE bytes (not > limit) is allowed through to
        // the JSON parser; padded whitespace keeps it valid JSON.
        let metrics = ServerMetrics::new();
        let mut body = br#"{"prompt":"x"}"#.to_vec();
        body.resize(MAX_REQUEST_SIZE, b' ');
        let parsed: GenerateRequest =
            validate_and_parse(&body, &metrics).expect("boundary-size valid JSON parses");
        assert_eq!(parsed.prompt, "x");
    }
}