import { type Chat, type GeneratorController } from "@lmstudio/sdk";
import { configSchematics, CONFIG_DEFAULTS } from "./config";
import { ShartiLogger, type LogLevel } from "./utils/logger";
import { asBool, asNumber, asString } from "./utils/coerce";
import { formatBlock } from "./utils/block";
import { toOpenAIMessages, toOpenAITools, sanitizeToolMessageOrdering, deriveConversationKey } from "./convert/messages";
import { consumeStream } from "./convert/stream";
import { createOpenAI, buildRequestPayload, logDebugPayload, executeCompletions } from "./request";
import { extractProviderError, formatErrorMessage } from "./errors";
import { logGenerationStats, getSessionAccumulator } from "./stats";
import type { OpenAICompatMessageParam } from "./types";

/* -------------------------------------------------------------------------- */
/*                              Generation entrypoint                         */
/*                                                                            */
/* Thin orchestrator: wires validation, conversion, request building, abort   */
/* handling, retry and stats. Domain mechanics live in dedicated modules.     */
/* -------------------------------------------------------------------------- */

export async function generate(ctl: GeneratorController, history: Chat) {
  const config = ctl.getPluginConfig(configSchematics);
  const startTime = Date.now();

  // --- Logging -------------------------------------------------------------
  const logLevel = asString(config.get("logLevel"), CONFIG_DEFAULTS.logLevel) as LogLevel;
  const logger = new ShartiLogger(ctl, { logLevel });

  // --- Validate required connection fields ---------------------------------
  const baseUrl = asString(config.get("baseUrl"));
  const model = asString(config.get("model"));

  if (!baseUrl) {
    logger.error("Missing Base URL. Please set it in the plugin settings.");
    return;
  }
  if (!model) {
    logger.error("Missing Model name. Please set it in the plugin settings.");
    return;
  }

  // --- Remaining config ----------------------------------------------------
  const apiKey = asString(config.get("apiKey"));
  const maxRetries = asNumber(config.get("maxRetries"), CONFIG_DEFAULTS.maxRetries);

  // --- Assemble request inputs --------------------------------------------
  let messages: OpenAICompatMessageParam[] = await toOpenAIMessages(history);
  messages = sanitizeToolMessageOrdering(messages);
  const tools = toOpenAITools(ctl);
  const requestPayload = buildRequestPayload(config, messages, tools);

  // --- Observability -------------------------------------------------------
  const reqBlock = formatBlock("🚀 Request:", [
    ["baseUrl", baseUrl],
    ["model", model],
    ["messages", messages.length],
    ["tools", tools?.length ?? 0],
  ]);
  if (reqBlock) logger.info(reqBlock);
  logDebugPayload(logger, requestPayload);

  // --- Abort plumbing ------------------------------------------------------
  const abortController = new AbortController();
  const onAbort = (): void => {
    logger.warn("Generation aborted by user.");
    abortController.abort();
  };
  ctl.abortSignal.addEventListener("abort", onAbort, { once: true });

  // A single client is reused across retries (connection pooling preserved).
  const openai = createOpenAI(baseUrl, apiKey);
  // Persistent-per-conversation accumulator: survives the multi-round (tool handoff)
  // regenerations of a single turn, isolated per conversation thread and concurrency-safe.
  const turnStats = getSessionAccumulator(deriveConversationKey(history));

  try {
    for (let attempt = 0; attempt <= maxRetries; attempt++) {
      if (ctl.abortSignal.aborted) {
        logger.warn("Aborted during retry wait.");
        return;
      }

      let stream;
      try {
        stream = await executeCompletions(openai, requestPayload, abortController.signal);
      } catch (error: unknown) {
        if (ctl.abortSignal.aborted) {
          logger.warn("Generation aborted.");
          return;
        }

        const info = extractProviderError(error);

        // Transient 503: retry with capped exponential backoff (1s → 2s → 4s → 8s → 10s).
        if (info.status === 503 && attempt < maxRetries) {
          const delay = Math.min(1000 * Math.pow(2, attempt), 10000);
          logger.warn(`Server returned 503 (attempt ${attempt + 1}/${maxRetries + 1}). Retrying in ${delay}ms…`);
          await new Promise((resolve) => setTimeout(resolve, delay));
          continue;
        }

        const formatted = formatErrorMessage(info);
        logger.error(formatted);
        logger.debug(`Raw error details: ${JSON.stringify(info, null, 2)}`);
        return;
      }

      // Stream established — consume (mid-stream failures are reported, not retried).
      try {
        const stats = await consumeStream(stream, ctl, logger, startTime);
        logGenerationStats(logger, stats);
        turnStats.record(stats, history.getLength());
        turnStats.emitSummaryIfFinal(logger, stats.finishReason, asBool(config.get("showTurnSummaryInChat"), CONFIG_DEFAULTS.showTurnSummaryInChat));
      } catch (error: unknown) {
        if (ctl.abortSignal.aborted) {
          logger.warn("Generation aborted during streaming.");
          return;
        }
        const info = extractProviderError(error);
        const formatted = formatErrorMessage(info);
        logger.error(`Stream error: ${formatted}`);
        logger.debug(`Raw stream error details: ${JSON.stringify(info, null, 2)}`);
      }
      return;
    }
  } finally {
    ctl.abortSignal.removeEventListener("abort", onAbort);
  }
}

import { type Chat, type GeneratorController } from "@lmstudio/sdk";
import { configSchematics, CONFIG_DEFAULTS } from "./config";
import { ShartiLogger, type LogLevel } from "./utils/logger";
import { asBool, asNumber, asString } from "./utils/coerce";
import { formatBlock } from "./utils/block";
import { toOpenAIMessages, toOpenAITools, sanitizeToolMessageOrdering, deriveConversationKey } from "./convert/messages";
import { consumeStream } from "./convert/stream";
import { createOpenAI, buildRequestPayload, logDebugPayload, executeCompletions } from "./request";
import { extractProviderError, formatErrorMessage } from "./errors";
import { logGenerationStats, getSessionAccumulator } from "./stats";
import type { OpenAICompatMessageParam } from "./types";

/* -------------------------------------------------------------------------- */
/*                              Generation entrypoint                         */
/*                                                                            */
/* Thin orchestrator: wires validation, conversion, request building, abort   */
/* handling, retry and stats. Domain mechanics live in dedicated modules.     */
/* -------------------------------------------------------------------------- */

export async function generate(ctl: GeneratorController, history: Chat) {
  const config = ctl.getPluginConfig(configSchematics);
  const startTime = Date.now();

  // --- Logging -------------------------------------------------------------
  const logLevel = asString(config.get("logLevel"), CONFIG_DEFAULTS.logLevel) as LogLevel;
  const logger = new ShartiLogger(ctl, { logLevel });

  // --- Validate required connection fields ---------------------------------
  const baseUrl = asString(config.get("baseUrl"));
  const model = asString(config.get("model"));

  if (!baseUrl) {
    logger.error("Missing Base URL. Please set it in the plugin settings.");
    return;
  }
  if (!model) {
    logger.error("Missing Model name. Please set it in the plugin settings.");
    return;
  }

  // --- Remaining config ----------------------------------------------------
  const apiKey = asString(config.get("apiKey"));
  const maxRetries = asNumber(config.get("maxRetries"), CONFIG_DEFAULTS.maxRetries);

  // --- Assemble request inputs --------------------------------------------
  let messages: OpenAICompatMessageParam[] = await toOpenAIMessages(history);
  messages = sanitizeToolMessageOrdering(messages);
  const tools = toOpenAITools(ctl);
  const requestPayload = buildRequestPayload(config, messages, tools);

  // --- Observability -------------------------------------------------------
  const reqBlock = formatBlock("🚀 Request:", [
    ["baseUrl", baseUrl],
    ["model", model],
    ["messages", messages.length],
    ["tools", tools?.length ?? 0],
  ]);
  if (reqBlock) logger.info(reqBlock);
  logDebugPayload(logger, requestPayload);

  // --- Abort plumbing ------------------------------------------------------
  const abortController = new AbortController();
  const onAbort = (): void => {
    logger.warn("Generation aborted by user.");
    abortController.abort();
  };
  ctl.abortSignal.addEventListener("abort", onAbort, { once: true });

  // A single client is reused across retries (connection pooling preserved).
  const openai = createOpenAI(baseUrl, apiKey);
  // Persistent-per-conversation accumulator: survives the multi-round (tool handoff)
  // regenerations of a single turn, isolated per conversation thread and concurrency-safe.
  const turnStats = getSessionAccumulator(deriveConversationKey(history));

  try {
    for (let attempt = 0; attempt <= maxRetries; attempt++) {
      if (ctl.abortSignal.aborted) {
        logger.warn("Aborted during retry wait.");
        return;
      }

      let stream;
      try {
        stream = await executeCompletions(openai, requestPayload, abortController.signal);
      } catch (error: unknown) {
        if (ctl.abortSignal.aborted) {
          logger.warn("Generation aborted.");
          return;
        }

        const info = extractProviderError(error);

        // Transient 503: retry with capped exponential backoff (1s → 2s → 4s → 8s → 10s).
        if (info.status === 503 && attempt < maxRetries) {
          const delay = Math.min(1000 * Math.pow(2, attempt), 10000);
          logger.warn(`Server returned 503 (attempt ${attempt + 1}/${maxRetries + 1}). Retrying in ${delay}ms…`);
          await new Promise((resolve) => setTimeout(resolve, delay));
          continue;
        }

        const formatted = formatErrorMessage(info);
        logger.error(formatted);
        logger.debug(`Raw error details: ${JSON.stringify(info, null, 2)}`);
        return;
      }

      // Stream established — consume (mid-stream failures are reported, not retried).
      try {
        const stats = await consumeStream(stream, ctl, logger, startTime);
        logGenerationStats(logger, stats);
        turnStats.record(stats, history.getLength());
        turnStats.emitSummaryIfFinal(logger, stats.finishReason, asBool(config.get("showTurnSummaryInChat"), CONFIG_DEFAULTS.showTurnSummaryInChat));
      } catch (error: unknown) {
        if (ctl.abortSignal.aborted) {
          logger.warn("Generation aborted during streaming.");
          return;
        }
        const info = extractProviderError(error);
        const formatted = formatErrorMessage(info);
        logger.error(`Stream error: ${formatted}`);
        logger.debug(`Raw stream error details: ${JSON.stringify(info, null, 2)}`);
      }
      return;
    }
  } finally {
    ctl.abortSignal.removeEventListener("abort", onAbort);
  }
}