use std::io::BufRead;
use std::path::Path;
use std::time::Duration;

use bytes::Bytes;
use codex_login::AuthMode;
use codex_login::CodexAuth;
use eventsource_stream::Eventsource;
use futures::prelude::*;
use reqwest::StatusCode;
use serde::Deserialize;
use serde::Serialize;
use serde_json::Value;
use tokio::sync::mpsc;
use tokio::time::timeout;
use tokio_util::io::ReaderStream;
use tracing::debug;
use tracing::trace;
use tracing::warn;
use uuid::Uuid;

use crate::chat_completions::AggregateStreamExt;
use crate::chat_completions::stream_chat_completions;
use crate::client_common::Prompt;
use crate::client_common::ResponseEvent;
use crate::client_common::ResponseStream;
use crate::client_common::ResponsesApiRequest;
use crate::client_common::create_reasoning_param_for_request;
use crate::config::Config;
use crate::config_types::ReasoningEffort as ReasoningEffortConfig;
use crate::config_types::ReasoningSummary as ReasoningSummaryConfig;
use crate::error::CodexErr;
use crate::error::Result;
use crate::flags::CODEX_RS_SSE_FIXTURE;
use crate::model_provider_info::ModelProviderInfo;
use crate::model_provider_info::WireApi;
use crate::models::ResponseItem;
use crate::openai_tools::create_tools_json_for_responses_api;
use crate::protocol::TokenUsage;
use crate::util::backoff;
use std::sync::Arc;

#[derive(Clone)]
pub struct ModelClient {
    config: Arc<Config>,
    auth: Option<CodexAuth>,
    client: reqwest::Client,
    provider: ModelProviderInfo,
    session_id: Uuid,
    effort: ReasoningEffortConfig,
    summary: ReasoningSummaryConfig,
}

impl ModelClient {
    pub fn new(
        config: Arc<Config>,
        auth: Option<CodexAuth>,
        provider: ModelProviderInfo,
        effort: ReasoningEffortConfig,
        summary: ReasoningSummaryConfig,
        session_id: Uuid,
    ) -> Self {
        Self {
            config,
            auth,
            client: reqwest::Client::new(),
            provider,
            session_id,
            effort,
            summary,
        }
    }

    /// Dispatches to either the Responses or Chat implementation depending on
    /// the provider config.  Public callers always invoke `stream()` – the
    /// specialised helpers are private to avoid accidental misuse.
    pub async fn stream(&self, prompt: &Prompt) -> Result<ResponseStream> {
        match self.provider.wire_api {
            WireApi::Responses => self.stream_responses(prompt).await,
            WireApi::Chat => {
                // Create the raw streaming connection first.
                let response_stream = stream_chat_completions(
                    prompt,
                    &self.config.model_family,
                    &self.client,
                    &self.provider,
                )
                .await?;

                // Wrap it with the aggregation adapter so callers see *only*
                // the final assistant message per turn (matching the
                // behaviour of the Responses API).
                let mut aggregated = if self.config.show_raw_agent_reasoning {
                    crate::chat_completions::AggregatedChatStream::streaming_mode(response_stream)
                } else {
                    response_stream.aggregate()
                };

                // Bridge the aggregated stream back into a standard
                // `ResponseStream` by forwarding events through a channel.
                let (tx, rx) = mpsc::channel::<Result<ResponseEvent>>(16);

                tokio::spawn(async move {
                    use futures::StreamExt;
                    while let Some(ev) = aggregated.next().await {
                        // Exit early if receiver hung up.
                        if tx.send(ev).await.is_err() {
                            break;
                        }
                    }
                });

                Ok(ResponseStream { rx_event: rx })
            }
        }
    }

    /// Implementation for the OpenAI *Responses* experimental API.
    async fn stream_responses(&self, prompt: &Prompt) -> Result<ResponseStream> {
        if let Some(path) = &*CODEX_RS_SSE_FIXTURE {
            // short circuit for tests
            warn!(path, "Streaming from fixture");
            return stream_from_fixture(path, self.provider.clone()).await;
        }

        let auth = self.auth.clone();

        let auth_mode = auth.as_ref().map(|a| a.mode);

        if self.config.model_family.family == "2025-08-06-model"
            && auth_mode != Some(AuthMode::ChatGPT)
        {
            return Err(CodexErr::UnexpectedStatus(
                StatusCode::BAD_REQUEST,
                "2025-08-06-model is only supported with ChatGPT auth, run `codex login status` to check your auth status and `codex login` to login with ChatGPT".to_string(),
            ));
        }

        let store = prompt.store && auth_mode != Some(AuthMode::ChatGPT);

        let full_instructions = prompt.get_full_instructions(&self.config.model_family);
        let tools_json = create_tools_json_for_responses_api(&prompt.tools)?;
        let reasoning = create_reasoning_param_for_request(
            &self.config.model_family,
            self.effort,
            self.summary,
        );

        // Request encrypted COT if we are not storing responses,
        // otherwise reasoning items will be referenced by ID
        let include: Vec<String> = if !store && reasoning.is_some() {
            vec!["reasoning.encrypted_content".to_string()]
        } else {
            vec![]
        };

        let input_with_instructions = prompt.get_formatted_input();

        let payload = ResponsesApiRequest {
            model: &self.config.model,
            instructions: &full_instructions,
            input: &input_with_instructions,
            tools: &tools_json,
            tool_choice: "auto",
            parallel_tool_calls: false,
            reasoning,
            store,
            stream: true,
            include,
        };

        let mut attempt = 0;
        let max_retries = self.provider.request_max_retries();

        trace!(
            "POST to {}: {}",
            self.provider.get_full_url(&auth),
            serde_json::to_string(&payload)?
        );

        loop {
            attempt += 1;

            let mut req_builder = self
                .provider
                .create_request_builder(&self.client, &auth)
                .await?;

            req_builder = req_builder
                .header("OpenAI-Beta", "responses=experimental")
                .header("session_id", self.session_id.to_string())
                .header(reqwest::header::ACCEPT, "text/event-stream")
                .json(&payload);

            if let Some(auth) = auth.as_ref()
                && auth.mode == AuthMode::ChatGPT
                && let Some(account_id) = auth.get_account_id().await
            {
                req_builder = req_builder.header("chatgpt-account-id", account_id);
            }

            let originator = self
                .config
                .internal_originator
                .as_deref()
                .unwrap_or("codex_cli_rs");
            req_builder = req_builder.header("originator", originator);

            let res = req_builder.send().await;
            if let Ok(resp) = &res {
                trace!(
                    "Response status: {}, request-id: {}",
                    resp.status(),
                    resp.headers()
                        .get("x-request-id")
                        .map(|v| v.to_str().unwrap_or_default())
                        .unwrap_or_default()
                );
            }

            match res {
                Ok(resp) if resp.status().is_success() => {
                    let (tx_event, rx_event) = mpsc::channel::<Result<ResponseEvent>>(1600);

                    // spawn task to process SSE
                    let stream = resp.bytes_stream().map_err(CodexErr::Reqwest);
                    tokio::spawn(process_sse(
                        stream,
                        tx_event,
                        self.provider.stream_idle_timeout(),
                    ));

                    return Ok(ResponseStream { rx_event });
                }
                Ok(res) => {
                    let status = res.status();
                    // The OpenAI Responses endpoint returns structured JSON bodies even for 4xx/5xx
                    // errors. When we bubble early with only the HTTP status the caller sees an opaque
                    // "unexpected status 400 Bad Request" which makes debugging nearly impossible.
                    // Instead, read (and include) the response text so higher layers and users see the
                    // exact error message (e.g. "Unknown parameter: 'input[0].metadata'"). The body is
                    // small and this branch only runs on error paths so the extra allocation is
                    // negligible.
                    if !(status == StatusCode::TOO_MANY_REQUESTS || status.is_server_error()) {
                        // Surface the error body to callers. Use `unwrap_or_default` per Clippy.
                        let body = res.text().await.unwrap_or_default();
                        return Err(CodexErr::UnexpectedStatus(status, body));
                    }

                    if attempt > max_retries {
                        return Err(CodexErr::RetryLimit(status));
                    }

                    // Pull out Retry‑After header if present.
                    let retry_after_secs = res
                        .headers()
                        .get(reqwest::header::RETRY_AFTER)
                        .and_then(|v| v.to_str().ok())
                        .and_then(|s| s.parse::<u64>().ok());

                    let delay = retry_after_secs
                        .map(|s| Duration::from_millis(s * 1_000))
                        .unwrap_or_else(|| backoff(attempt));
                    tokio::time::sleep(delay).await;
                }
                Err(e) => {
                    if attempt > max_retries {
                        return Err(e.into());
                    }
                    let delay = backoff(attempt);
                    tokio::time::sleep(delay).await;
                }
            }
        }
    }

    pub fn get_provider(&self) -> ModelProviderInfo {
        self.provider.clone()
    }
}

#[derive(Debug, Deserialize, Serialize)]
struct SseEvent {
    #[serde(rename = "type")]
    kind: String,
    response: Option<Value>,
    item: Option<Value>,
    delta: Option<String>,
}

#[derive(Debug, Deserialize)]
struct ResponseCreated {}

#[derive(Debug, Deserialize)]
struct ResponseCompleted {
    id: String,
    usage: Option<ResponseCompletedUsage>,
}

#[derive(Debug, Deserialize)]
struct ResponseCompletedUsage {
    input_tokens: u64,
    input_tokens_details: Option<ResponseCompletedInputTokensDetails>,
    output_tokens: u64,
    output_tokens_details: Option<ResponseCompletedOutputTokensDetails>,
    total_tokens: u64,
}

impl From<ResponseCompletedUsage> for TokenUsage {
    fn from(val: ResponseCompletedUsage) -> Self {
        TokenUsage {
            input_tokens: val.input_tokens,
            cached_input_tokens: val.input_tokens_details.map(|d| d.cached_tokens),
            output_tokens: val.output_tokens,
            reasoning_output_tokens: val.output_tokens_details.map(|d| d.reasoning_tokens),
            total_tokens: val.total_tokens,
        }
    }
}

#[derive(Debug, Deserialize)]
struct ResponseCompletedInputTokensDetails {
    cached_tokens: u64,
}

#[derive(Debug, Deserialize)]
struct ResponseCompletedOutputTokensDetails {
    reasoning_tokens: u64,
}

async fn process_sse<S>(
    stream: S,
    tx_event: mpsc::Sender<Result<ResponseEvent>>,
    idle_timeout: Duration,
) where
    S: Stream<Item = Result<Bytes>> + Unpin,
{
    let mut stream = stream.eventsource();

    // If the stream stays completely silent for an extended period treat it as disconnected.
    // The response id returned from the "complete" message.
    let mut response_completed: Option<ResponseCompleted> = None;

    loop {
        let sse = match timeout(idle_timeout, stream.next()).await {
            Ok(Some(Ok(sse))) => sse,
            Ok(Some(Err(e))) => {
                debug!("SSE Error: {e:#}");
                let event = CodexErr::Stream(e.to_string());
                let _ = tx_event.send(Err(event)).await;
                return;
            }
            Ok(None) => {
                match response_completed {
                    Some(ResponseCompleted {
                        id: response_id,
                        usage,
                    }) => {
                        let event = ResponseEvent::Completed {
                            response_id,
                            token_usage: usage.map(Into::into),
                        };
                        let _ = tx_event.send(Ok(event)).await;
                    }
                    None => {
                        let _ = tx_event
                            .send(Err(CodexErr::Stream(
                                "stream closed before response.completed".into(),
                            )))
                            .await;
                    }
                }
                return;
            }
            Err(_) => {
                let _ = tx_event
                    .send(Err(CodexErr::Stream("idle timeout waiting for SSE".into())))
                    .await;
                return;
            }
        };

        let event: SseEvent = match serde_json::from_str(&sse.data) {
            Ok(event) => event,
            Err(e) => {
                debug!("Failed to parse SSE event: {e}, data: {}", &sse.data);
                continue;
            }
        };

        trace!(?event, "SSE event");
        match event.kind.as_str() {
            // Individual output item finalised. Forward immediately so the
            // rest of the agent can stream assistant text/functions *live*
            // instead of waiting for the final `response.completed` envelope.
            //
            // IMPORTANT: We used to ignore these events and forward the
            // duplicated `output` array embedded in the `response.completed`
            // payload.  That produced two concrete issues:
            //   1. No real‑time streaming – the user only saw output after the
            //      entire turn had finished, which broke the "typing" UX and
            //      made long‑running turns look stalled.
            //   2. Duplicate `function_call_output` items – both the
            //      individual *and* the completed array were forwarded, which
            //      confused the backend and triggered 400
            //      "previous_response_not_found" errors because the duplicated
            //      IDs did not match the incremental turn chain.
            //
            // The fix is to forward the incremental events *as they come* and
            // drop the duplicated list inside `response.completed`.
            "response.output_item.done" => {
                let Some(item_val) = event.item else { continue };
                let Ok(item) = serde_json::from_value::<ResponseItem>(item_val) else {
                    debug!("failed to parse ResponseItem from output_item.done");
                    continue;
                };

                let event = ResponseEvent::OutputItemDone(item);
                if tx_event.send(Ok(event)).await.is_err() {
                    return;
                }
            }
            "response.output_text.delta" => {
                if let Some(delta) = event.delta {
                    let event = ResponseEvent::OutputTextDelta(delta);
                    if tx_event.send(Ok(event)).await.is_err() {
                        return;
                    }
                }
            }
            "response.reasoning_summary_text.delta" => {
                if let Some(delta) = event.delta {
                    let event = ResponseEvent::ReasoningSummaryDelta(delta);
                    if tx_event.send(Ok(event)).await.is_err() {
                        return;
                    }
                }
            }
            "response.reasoning_text.delta" => {
                if let Some(delta) = event.delta {
                    let event = ResponseEvent::ReasoningContentDelta(delta);
                    if tx_event.send(Ok(event)).await.is_err() {
                        return;
                    }
                }
            }
            "response.created" => {
                if event.response.is_some() {
                    let _ = tx_event.send(Ok(ResponseEvent::Created {})).await;
                }
            }
            "response.failed" => {
                if let Some(resp_val) = event.response {
                    let error = resp_val
                        .get("error")
                        .and_then(|v| v.get("message"))
                        .and_then(|v| v.as_str())
                        .unwrap_or("response.failed event received");

                    let _ = tx_event
                        .send(Err(CodexErr::Stream(error.to_string())))
                        .await;
                }
            }
            // Final response completed – includes array of output items & id
            "response.completed" => {
                if let Some(resp_val) = event.response {
                    match serde_json::from_value::<ResponseCompleted>(resp_val) {
                        Ok(r) => {
                            response_completed = Some(r);
                        }
                        Err(e) => {
                            debug!("failed to parse ResponseCompleted: {e}");
                            continue;
                        }
                    };
                };
            }
            "response.content_part.done"
            | "response.function_call_arguments.delta"
            | "response.in_progress"
            | "response.output_item.added"
            | "response.output_text.done"
            | "response.reasoning_summary_part.added"
            | "response.reasoning_summary_text.done" => {
                // Currently, we ignore these events, but we handle them
                // separately to skip the logging message in the `other` case.
            }
            other => debug!(other, "sse event"),
        }
    }
}

/// used in tests to stream from a text SSE file
async fn stream_from_fixture(
    path: impl AsRef<Path>,
    provider: ModelProviderInfo,
) -> Result<ResponseStream> {
    let (tx_event, rx_event) = mpsc::channel::<Result<ResponseEvent>>(1600);
    let f = std::fs::File::open(path.as_ref())?;
    let lines = std::io::BufReader::new(f).lines();

    // insert \n\n after each line for proper SSE parsing
    let mut content = String::new();
    for line in lines {
        content.push_str(&line?);
        content.push_str("\n\n");
    }

    let rdr = std::io::Cursor::new(content);
    let stream = ReaderStream::new(rdr).map_err(CodexErr::Io);
    tokio::spawn(process_sse(
        stream,
        tx_event,
        provider.stream_idle_timeout(),
    ));
    Ok(ResponseStream { rx_event })
}

#[cfg(test)]
mod tests {
    #![allow(clippy::expect_used, clippy::unwrap_used)]

    use super::*;
    use serde_json::json;
    use tokio::sync::mpsc;
    use tokio_test::io::Builder as IoBuilder;
    use tokio_util::io::ReaderStream;

    // ────────────────────────────
    // Helpers
    // ────────────────────────────

    /// Runs the SSE parser on pre-chunked byte slices and returns every event
    /// (including any final `Err` from a stream-closure check).
    async fn collect_events(
        chunks: &[&[u8]],
        provider: ModelProviderInfo,
    ) -> Vec<Result<ResponseEvent>> {
        let mut builder = IoBuilder::new();
        for chunk in chunks {
            builder.read(chunk);
        }

        let reader = builder.build();
        let stream = ReaderStream::new(reader).map_err(CodexErr::Io);
        let (tx, mut rx) = mpsc::channel::<Result<ResponseEvent>>(16);
        tokio::spawn(process_sse(stream, tx, provider.stream_idle_timeout()));

        let mut events = Vec::new();
        while let Some(ev) = rx.recv().await {
            events.push(ev);
        }
        events
    }

    /// Builds an in-memory SSE stream from JSON fixtures and returns only the
    /// successfully parsed events (panics on internal channel errors).
    async fn run_sse(
        events: Vec<serde_json::Value>,
        provider: ModelProviderInfo,
    ) -> Vec<ResponseEvent> {
        let mut body = String::new();
        for e in events {
            let kind = e
                .get("type")
                .and_then(|v| v.as_str())
                .expect("fixture event missing type");
            if e.as_object().map(|o| o.len() == 1).unwrap_or(false) {
                body.push_str(&format!("event: {kind}\n\n"));
            } else {
                body.push_str(&format!("event: {kind}\ndata: {e}\n\n"));
            }
        }

        let (tx, mut rx) = mpsc::channel::<Result<ResponseEvent>>(8);
        let stream = ReaderStream::new(std::io::Cursor::new(body)).map_err(CodexErr::Io);
        tokio::spawn(process_sse(stream, tx, provider.stream_idle_timeout()));

        let mut out = Vec::new();
        while let Some(ev) = rx.recv().await {
            out.push(ev.expect("channel closed"));
        }
        out
    }

    // ────────────────────────────
    // Tests from `implement-test-for-responses-api-sse-parser`
    // ────────────────────────────

    #[tokio::test]
    async fn parses_items_and_completed() {
        let item1 = json!({
            "type": "response.output_item.done",
            "item": {
                "type": "message",
                "role": "assistant",
                "content": [{"type": "output_text", "text": "Hello"}]
            }
        })
        .to_string();

        let item2 = json!({
            "type": "response.output_item.done",
            "item": {
                "type": "message",
                "role": "assistant",
                "content": [{"type": "output_text", "text": "World"}]
            }
        })
        .to_string();

        let completed = json!({
            "type": "response.completed",
            "response": { "id": "resp1" }
        })
        .to_string();

        let sse1 = format!("event: response.output_item.done\ndata: {item1}\n\n");
        let sse2 = format!("event: response.output_item.done\ndata: {item2}\n\n");
        let sse3 = format!("event: response.completed\ndata: {completed}\n\n");

        let provider = ModelProviderInfo {
            name: "test".to_string(),
            base_url: Some("https://test.com".to_string()),
            env_key: Some("TEST_API_KEY".to_string()),
            env_key_instructions: None,
            wire_api: WireApi::Responses,
            query_params: None,
            http_headers: None,
            env_http_headers: None,
            request_max_retries: Some(0),
            stream_max_retries: Some(0),
            stream_idle_timeout_ms: Some(1000),
            requires_openai_auth: false,
        };

        let events = collect_events(
            &[sse1.as_bytes(), sse2.as_bytes(), sse3.as_bytes()],
            provider,
        )
        .await;

        assert_eq!(events.len(), 3);

        matches!(
            &events[0],
            Ok(ResponseEvent::OutputItemDone(ResponseItem::Message { role, .. }))
                if role == "assistant"
        );

        matches!(
            &events[1],
            Ok(ResponseEvent::OutputItemDone(ResponseItem::Message { role, .. }))
                if role == "assistant"
        );

        match &events[2] {
            Ok(ResponseEvent::Completed {
                response_id,
                token_usage,
            }) => {
                assert_eq!(response_id, "resp1");
                assert!(token_usage.is_none());
            }
            other => panic!("unexpected third event: {other:?}"),
        }
    }

    #[tokio::test]
    async fn error_when_missing_completed() {
        let item1 = json!({
            "type": "response.output_item.done",
            "item": {
                "type": "message",
                "role": "assistant",
                "content": [{"type": "output_text", "text": "Hello"}]
            }
        })
        .to_string();

        let sse1 = format!("event: response.output_item.done\ndata: {item1}\n\n");
        let provider = ModelProviderInfo {
            name: "test".to_string(),
            base_url: Some("https://test.com".to_string()),
            env_key: Some("TEST_API_KEY".to_string()),
            env_key_instructions: None,
            wire_api: WireApi::Responses,
            query_params: None,
            http_headers: None,
            env_http_headers: None,
            request_max_retries: Some(0),
            stream_max_retries: Some(0),
            stream_idle_timeout_ms: Some(1000),
            requires_openai_auth: false,
        };

        let events = collect_events(&[sse1.as_bytes()], provider).await;

        assert_eq!(events.len(), 2);

        matches!(events[0], Ok(ResponseEvent::OutputItemDone(_)));

        match &events[1] {
            Err(CodexErr::Stream(msg)) => {
                assert_eq!(msg, "stream closed before response.completed")
            }
            other => panic!("unexpected second event: {other:?}"),
        }
    }

    // ────────────────────────────
    // Table-driven test from `main`
    // ────────────────────────────

    /// Verifies that the adapter produces the right `ResponseEvent` for a
    /// variety of incoming `type` values.
    #[tokio::test]
    async fn table_driven_event_kinds() {
        struct TestCase {
            name: &'static str,
            event: serde_json::Value,
            expect_first: fn(&ResponseEvent) -> bool,
            expected_len: usize,
        }

        fn is_created(ev: &ResponseEvent) -> bool {
            matches!(ev, ResponseEvent::Created)
        }
        fn is_output(ev: &ResponseEvent) -> bool {
            matches!(ev, ResponseEvent::OutputItemDone(_))
        }
        fn is_completed(ev: &ResponseEvent) -> bool {
            matches!(ev, ResponseEvent::Completed { .. })
        }

        let completed = json!({
            "type": "response.completed",
            "response": {
                "id": "c",
                "usage": {
                    "input_tokens": 0,
                    "input_tokens_details": null,
                    "output_tokens": 0,
                    "output_tokens_details": null,
                    "total_tokens": 0
                },
                "output": []
            }
        });

        let cases = vec![
            TestCase {
                name: "created",
                event: json!({"type": "response.created", "response": {}}),
                expect_first: is_created,
                expected_len: 2,
            },
            TestCase {
                name: "output_item.done",
                event: json!({
                    "type": "response.output_item.done",
                    "item": {
                        "type": "message",
                        "role": "assistant",
                        "content": [
                            {"type": "output_text", "text": "hi"}
                        ]
                    }
                }),
                expect_first: is_output,
                expected_len: 2,
            },
            TestCase {
                name: "unknown",
                event: json!({"type": "response.new_tool_event"}),
                expect_first: is_completed,
                expected_len: 1,
            },
        ];

        for case in cases {
            let mut evs = vec![case.event];
            evs.push(completed.clone());

            let provider = ModelProviderInfo {
                name: "test".to_string(),
                base_url: Some("https://test.com".to_string()),
                env_key: Some("TEST_API_KEY".to_string()),
                env_key_instructions: None,
                wire_api: WireApi::Responses,
                query_params: None,
                http_headers: None,
                env_http_headers: None,
                request_max_retries: Some(0),
                stream_max_retries: Some(0),
                stream_idle_timeout_ms: Some(1000),
                requires_openai_auth: false,
            };

            let out = run_sse(evs, provider).await;
            assert_eq!(out.len(), case.expected_len, "case {}", case.name);
            assert!(
                (case.expect_first)(&out[0]),
                "first event mismatch in case {}",
                case.name
            );
        }
    }
}