#include "mrjunejune/inference_bridge.h"

#include <errno.h>
#include <pthread.h>
#include <signal.h>
#include <stdatomic.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>

#define INFERENCE_EVENT_MAX (1024 * 1024 + 4096)
#define INFERENCE_READY_TIMEOUT_MS 15000

struct Inference_Bridge {
  char *sidecar_path;
  char *copilot_cli_path;
  FILE *p_commands;
  FILE *p_events;
  pid_t child_pid;
  pthread_t reader_thread;
  pthread_mutex_t write_mutex;
  pthread_mutex_t ready_mutex;
  pthread_cond_t ready_condition;
  boolean write_mutex_initialized;
  boolean ready_mutex_initialized;
  boolean ready_condition_initialized;
  boolean reader_started;
  _Atomic boolean running;
  _Atomic boolean ready;
  Inference_Event_Handler handler;
  void *p_user_data;
};

static int64 Inference_Monotonic_Milliseconds(void)
{
  struct timespec now;
  clock_gettime(CLOCK_MONOTONIC, &now);
  return (int64)now.tv_sec * 1000 + now.tv_nsec / 1000000;
}

static const char *Inference_JSON_String(
    Dowa_JSON_Entry *object,
    const char *key)
{
  char *value = Dowa_JSON_Get_String(object, key);
  return value ? value : "";
}

static boolean Inference_JSON_Boolean(
    Dowa_JSON_Entry *object,
    const char *key)
{
  Dowa_JSON_Value *p_value = Dowa_JSON_Get(object, key);
  return p_value && p_value->type == DOWA_JSON_BOOL
      ? p_value->bool_val
      : FALSE;
}

static int64 Inference_JSON_Integer(
    Dowa_JSON_Entry *object,
    const char *key)
{
  Dowa_JSON_Value *p_value = Dowa_JSON_Get(object, key);
  return p_value && p_value->type == DOWA_JSON_NUMBER
      ? (int64)p_value->num_val
      : 0;
}

static void Inference_Bridge_Handle_Line(
    Inference_Bridge *p_bridge,
    const char *line,
    size_t length)
{
  Dowa_Arena *p_arena = Dowa_Arena_Create(length * 3 + 4096);
  if (!p_arena)
    return;
  Dowa_JSON_Value parsed = Dowa_JSON_Parse(line, (int32)length, p_arena);
  if (parsed.type != DOWA_JSON_OBJECT)
  {
    Dowa_Arena_Free(p_arena);
    return;
  }
  Dowa_JSON_Entry *object = parsed.object_val;
  const char *type = Inference_JSON_String(object, "type");
  if (strcmp(type, "ready") == 0)
  {
    atomic_store(&p_bridge->ready, TRUE);
    pthread_mutex_lock(&p_bridge->ready_mutex);
    pthread_cond_broadcast(&p_bridge->ready_condition);
    pthread_mutex_unlock(&p_bridge->ready_mutex);
  }

  Inference_Event event = {
    .type = type,
    .request_id = Inference_JSON_String(object, "request_id"),
    .conversation_id = Inference_JSON_String(object, "conversation_id"),
    .delta = Inference_JSON_String(object, "delta"),
    .content = Inference_JSON_String(object, "content"),
    .raw_json = line,
    .raw_json_length = length,
    .failed = Inference_JSON_Boolean(object, "failed"),
    .aborted = Inference_JSON_Boolean(object, "aborted"),
  };

  Dowa_JSON_Value *p_error = Dowa_JSON_Get(object, "error");
  if (p_error && p_error->type == DOWA_JSON_OBJECT)
  {
    Dowa_JSON_Entry *error = p_error->object_val;
    event.error_code = Inference_JSON_String(error, "code");
    event.error_message = Inference_JSON_String(error, "message");
  }
  else
  {
    event.error_code = "";
    event.error_message = "";
  }

  Dowa_JSON_Value *p_usage = Dowa_JSON_Get(object, "usage");
  if (p_usage && p_usage->type == DOWA_JSON_OBJECT)
  {
    Dowa_JSON_Entry *usage = p_usage->object_val;
    event.input_tokens = Inference_JSON_Integer(usage, "input_tokens");
    event.output_tokens = Inference_JSON_Integer(usage, "output_tokens");
  }

  if (p_bridge->handler)
    p_bridge->handler(&event, p_bridge->p_user_data);
  Dowa_Arena_Free(p_arena);
}

static void *Inference_Bridge_Read_Events(void *p_context)
{
  Inference_Bridge *p_bridge = p_context;
  char *line = NULL;
  size_t capacity = 0;
  while (atomic_load(&p_bridge->running))
  {
    ssize_t amount = getline(&line, &capacity, p_bridge->p_events);
    if (amount < 0)
      break;
    if ((size_t)amount > INFERENCE_EVENT_MAX)
      continue;
    size_t length = (size_t)amount;
    while (length > 0 &&
           (line[length - 1] == '\n' || line[length - 1] == '\r'))
      length--;
    if (length > 0)
    {
      line[length] = '\0';
      Inference_Bridge_Handle_Line(p_bridge, line, length);
    }
  }
  free(line);
  atomic_store(&p_bridge->ready, FALSE);
  atomic_store(&p_bridge->running, FALSE);
  if (p_bridge->handler)
  {
    Inference_Event closed = {
      .type = "bridge.closed",
      .error_code = "sidecar_closed",
      .error_message = "Inference sidecar connection closed",
      .failed = TRUE,
    };
    p_bridge->handler(&closed, p_bridge->p_user_data);
  }
  pthread_mutex_lock(&p_bridge->ready_mutex);
  pthread_cond_broadcast(&p_bridge->ready_condition);
  pthread_mutex_unlock(&p_bridge->ready_mutex);
  return NULL;
}

static boolean Inference_Bridge_Write(
    Inference_Bridge *p_bridge,
    const char *payload)
{
  if (!p_bridge || !payload || !atomic_load(&p_bridge->running))
    return FALSE;
  pthread_mutex_lock(&p_bridge->write_mutex);
  boolean success =
      fputs(payload, p_bridge->p_commands) >= 0 &&
      fputc('\n', p_bridge->p_commands) != EOF &&
      fflush(p_bridge->p_commands) == 0;
  pthread_mutex_unlock(&p_bridge->write_mutex);
  return success;
}

static const char *Inference_Bridge_Profile_Name(
    Inference_Prompt_Profile profile)
{
  switch (profile)
  {
    case INFERENCE_PROMPT_PROFILE_PUBLIC_VISITOR:
      return "public_visitor";
    case INFERENCE_PROMPT_PROFILE_INVITED_FRIEND:
      return "invited_friend";
    case INFERENCE_PROMPT_PROFILE_JUNE_ADMIN:
      return "june_admin";
  }
  return NULL;
}

static boolean Inference_Bridge_Command(
    Inference_Bridge *p_bridge,
    const char *command,
    const char *request_id,
    const char *conversation_id,
    const char *prompt,
    const char *prompt_profile,
    uint32 prompt_version,
    uint32 knowledge_version)
{
  if (!command || !request_id)
    return FALSE;
  size_t input_length =
      strlen(command) + strlen(request_id) +
      strlen(conversation_id ? conversation_id : "") +
      strlen(prompt ? prompt : "") +
      strlen(prompt_profile ? prompt_profile : "");
  if (input_length > (((size_t)-1) - 4096) / 12)
    return FALSE;
  Dowa_Arena *p_arena = Dowa_Arena_Create(input_length * 12 + 4096);
  if (!p_arena)
    return FALSE;
  char *escaped_command = Dowa_JSON_Escape_String(command, 0, p_arena);
  char *escaped_request = Dowa_JSON_Escape_String(request_id, 0, p_arena);
  char *escaped_conversation = Dowa_JSON_Escape_String(
      conversation_id ? conversation_id : "", 0, p_arena);
  char *escaped_prompt = prompt
      ? Dowa_JSON_Escape_String(prompt, 0, p_arena)
      : NULL;
  char *escaped_profile = prompt_profile
      ? Dowa_JSON_Escape_String(prompt_profile, 0, p_arena)
      : NULL;
  if (!escaped_command || !escaped_request || !escaped_conversation ||
      (prompt && !escaped_prompt) ||
      (prompt_profile && !escaped_profile))
  {
    Dowa_Arena_Free(p_arena);
    return FALSE;
  }
  size_t capacity =
      strlen(escaped_command) + strlen(escaped_request) +
      strlen(escaped_conversation) +
      (escaped_prompt ? strlen(escaped_prompt) : 0) +
      (escaped_profile ? strlen(escaped_profile) : 0) + 256;
  char *payload = Dowa_Arena_Allocate(p_arena, capacity);
  if (!payload)
  {
    Dowa_Arena_Free(p_arena);
    return FALSE;
  }
  if (escaped_prompt && escaped_profile)
  {
    snprintf(
        payload,
        capacity,
        "{\"command\":\"%s\",\"request_id\":\"%s\","
        "\"conversation_id\":\"%s\",\"prompt\":\"%s\","
        "\"prompt_profile\":\"%s\",\"prompt_version\":%u,"
        "\"knowledge_version\":%u}",
        escaped_command,
        escaped_request,
        escaped_conversation,
        escaped_prompt,
        escaped_profile,
        prompt_version,
        knowledge_version);
  }
  else
  {
    snprintf(
        payload,
        capacity,
        "{\"command\":\"%s\",\"request_id\":\"%s\","
        "\"conversation_id\":\"%s\"}",
        escaped_command,
        escaped_request,
        escaped_conversation);
  }
  boolean success = Inference_Bridge_Write(p_bridge, payload);
  Dowa_Arena_Free(p_arena);
  return success;
}

Inference_Bridge *Inference_Bridge_Create(
    const char *sidecar_path,
    const char *copilot_cli_path,
    Inference_Event_Handler handler,
    void *p_user_data)
{
  if (!sidecar_path || !copilot_cli_path)
    return NULL;
  Inference_Bridge *p_bridge = calloc(1, sizeof(*p_bridge));
  if (!p_bridge)
    return NULL;
  p_bridge->sidecar_path = strdup(sidecar_path);
  p_bridge->copilot_cli_path = strdup(copilot_cli_path);
  p_bridge->handler = handler;
  p_bridge->p_user_data = p_user_data;
  p_bridge->child_pid = -1;
  if (!p_bridge->sidecar_path || !p_bridge->copilot_cli_path)
  {
    Inference_Bridge_Destroy(p_bridge);
    return NULL;
  }
  if (pthread_mutex_init(&p_bridge->write_mutex, NULL) != 0)
  {
    Inference_Bridge_Destroy(p_bridge);
    return NULL;
  }
  p_bridge->write_mutex_initialized = TRUE;
  if (pthread_mutex_init(&p_bridge->ready_mutex, NULL) != 0)
  {
    Inference_Bridge_Destroy(p_bridge);
    return NULL;
  }
  p_bridge->ready_mutex_initialized = TRUE;
  if (pthread_cond_init(&p_bridge->ready_condition, NULL) != 0)
  {
    Inference_Bridge_Destroy(p_bridge);
    return NULL;
  }
  p_bridge->ready_condition_initialized = TRUE;
  return p_bridge;
}

static void Inference_Bridge_Stop_Process(Inference_Bridge *p_bridge)
{
  if (!p_bridge)
    return;
  if (atomic_load(&p_bridge->running) && p_bridge->p_commands)
    Inference_Bridge_Command(
        p_bridge, "shutdown", "server-shutdown", "", NULL, NULL, 0, 0);
  if (p_bridge->p_commands)
  {
    fclose(p_bridge->p_commands);
    p_bridge->p_commands = NULL;
  }

  if (p_bridge->child_pid > 0)
  {
    int status = 0;
    pid_t result = 0;
    for (int attempt = 0; attempt < 20; attempt++)
    {
      result = waitpid(p_bridge->child_pid, &status, WNOHANG);
      if (result != 0)
        break;
      usleep(50000);
    }
    if (result == 0)
    {
      kill(p_bridge->child_pid, SIGTERM);
      waitpid(p_bridge->child_pid, &status, 0);
    }
    p_bridge->child_pid = -1;
  }
  atomic_store(&p_bridge->running, FALSE);
  atomic_store(&p_bridge->ready, FALSE);
  if (p_bridge->reader_started)
  {
    pthread_join(p_bridge->reader_thread, NULL);
    p_bridge->reader_started = FALSE;
  }
  if (p_bridge->p_events)
  {
    fclose(p_bridge->p_events);
    p_bridge->p_events = NULL;
  }
}

boolean Inference_Bridge_Start(Inference_Bridge *p_bridge)
{
  if (!p_bridge || atomic_load(&p_bridge->running))
    return FALSE;
  int commands[2];
  int events[2];
  if (pipe(commands) != 0)
    return FALSE;
  if (pipe(events) != 0)
  {
    close(commands[0]);
    close(commands[1]);
    return FALSE;
  }

  pid_t child = fork();
  if (child < 0)
  {
    close(commands[0]);
    close(commands[1]);
    close(events[0]);
    close(events[1]);
    return FALSE;
  }
  if (child == 0)
  {
    dup2(commands[0], STDIN_FILENO);
    dup2(events[1], STDOUT_FILENO);
    close(commands[0]);
    close(commands[1]);
    close(events[0]);
    close(events[1]);
    execl(
        p_bridge->sidecar_path,
        p_bridge->sidecar_path,
        p_bridge->copilot_cli_path,
        NULL);
    _exit(127);
  }

  close(commands[0]);
  close(events[1]);
  p_bridge->p_commands = fdopen(commands[1], "w");
  p_bridge->p_events = fdopen(events[0], "r");
  if (!p_bridge->p_commands || !p_bridge->p_events)
  {
    if (p_bridge->p_commands)
      fclose(p_bridge->p_commands);
    else
      close(commands[1]);
    if (p_bridge->p_events)
      fclose(p_bridge->p_events);
    else
      close(events[0]);
    kill(child, SIGTERM);
    waitpid(child, NULL, 0);
    return FALSE;
  }
  setvbuf(p_bridge->p_commands, NULL, _IOLBF, 0);
  p_bridge->child_pid = child;
  atomic_store(&p_bridge->running, TRUE);
  atomic_store(&p_bridge->ready, FALSE);
  if (pthread_create(
          &p_bridge->reader_thread,
          NULL,
          Inference_Bridge_Read_Events,
          p_bridge) != 0)
  {
    atomic_store(&p_bridge->running, FALSE);
    fclose(p_bridge->p_commands);
    fclose(p_bridge->p_events);
    kill(child, SIGTERM);
    waitpid(child, NULL, 0);
    p_bridge->p_commands = NULL;
    p_bridge->p_events = NULL;
    p_bridge->child_pid = -1;
    return FALSE;
  }
  p_bridge->reader_started = TRUE;

  int64 deadline =
      Inference_Monotonic_Milliseconds() + INFERENCE_READY_TIMEOUT_MS;
  pthread_mutex_lock(&p_bridge->ready_mutex);
  while (atomic_load(&p_bridge->running) &&
         !atomic_load(&p_bridge->ready))
  {
    int64 remaining = deadline - Inference_Monotonic_Milliseconds();
    if (remaining <= 0)
      break;
    struct timespec timeout;
    clock_gettime(CLOCK_REALTIME, &timeout);
    timeout.tv_sec += remaining / 1000;
    timeout.tv_nsec += (remaining % 1000) * 1000000;
    if (timeout.tv_nsec >= 1000000000)
    {
      timeout.tv_sec++;
      timeout.tv_nsec -= 1000000000;
    }
    pthread_cond_timedwait(
        &p_bridge->ready_condition, &p_bridge->ready_mutex, &timeout);
  }
  boolean ready = atomic_load(&p_bridge->ready);
  pthread_mutex_unlock(&p_bridge->ready_mutex);
  if (!ready)
  {
    Inference_Bridge_Stop_Process(p_bridge);
    return FALSE;
  }
  return TRUE;
}

boolean Inference_Bridge_Is_Ready(const Inference_Bridge *p_bridge)
{
  return p_bridge &&
      atomic_load(&p_bridge->running) &&
      atomic_load(&p_bridge->ready);
}

boolean Inference_Bridge_Start_Turn(
    Inference_Bridge *p_bridge,
    const char *request_id,
    const char *conversation_id,
    const char *prompt,
    Inference_Prompt_Profile prompt_profile,
    uint32 prompt_version,
    uint32 knowledge_version,
    const Inference_Bridge_History_Message *p_history,
    uint32 history_count)
{
  const char *profile_name = Inference_Bridge_Profile_Name(prompt_profile);
  if (!profile_name || prompt_version == 0 || knowledge_version == 0)
    return FALSE;
  if (history_count > INFERENCE_BRIDGE_HISTORY_MAX)
    return FALSE;
  /* Validate every history entry before touching the wire. */
  for (uint32 i = 0; i < history_count; i++)
  {
    const Inference_Bridge_History_Message *m = &p_history[i];
    if (!m->role || !m->content)
      return FALSE;
    if (strcmp(m->role, "user") != 0 && strcmp(m->role, "assistant") != 0)
      return FALSE;
  }

  if (!request_id)
    return FALSE;

  /* Compute raw byte total to size the arena (12x expansion factor). */
  size_t input_length =
      strlen("turn.start") + strlen(request_id) +
      strlen(conversation_id ? conversation_id : "") +
      strlen(prompt ? prompt : "") +
      strlen(profile_name);
  for (uint32 i = 0; i < history_count; i++)
  {
    input_length += strlen(p_history[i].role) + strlen(p_history[i].content);
  }
  if (input_length > (((size_t)-1) - 4096) / 12)
    return FALSE;

  Dowa_Arena *p_arena = Dowa_Arena_Create(input_length * 12 + 4096);
  if (!p_arena)
    return FALSE;

  char *escaped_command = Dowa_JSON_Escape_String("turn.start", 0, p_arena);
  char *escaped_request = Dowa_JSON_Escape_String(request_id, 0, p_arena);
  char *escaped_conversation = Dowa_JSON_Escape_String(
      conversation_id ? conversation_id : "", 0, p_arena);
  char *escaped_prompt = prompt
      ? Dowa_JSON_Escape_String(prompt, 0, p_arena)
      : NULL;
  char *escaped_profile = Dowa_JSON_Escape_String(profile_name, 0, p_arena);

  if (!escaped_command || !escaped_request || !escaped_conversation ||
      (prompt && !escaped_prompt) || !escaped_profile)
  {
    Dowa_Arena_Free(p_arena);
    return FALSE;
  }

  /* Escape history roles and contents. */
  char *escaped_hist_role[INFERENCE_BRIDGE_HISTORY_MAX];
  char *escaped_hist_content[INFERENCE_BRIDGE_HISTORY_MAX];
  for (uint32 i = 0; i < history_count; i++)
  {
    escaped_hist_role[i] =
        Dowa_JSON_Escape_String(p_history[i].role, 0, p_arena);
    escaped_hist_content[i] =
        Dowa_JSON_Escape_String(p_history[i].content, 0, p_arena);
    if (!escaped_hist_role[i] || !escaped_hist_content[i])
    {
      Dowa_Arena_Free(p_arena);
      return FALSE;
    }
  }

  /* Compute payload capacity. */
  size_t capacity =
      strlen(escaped_command) + strlen(escaped_request) +
      strlen(escaped_conversation) +
      (escaped_prompt ? strlen(escaped_prompt) : 0) +
      strlen(escaped_profile) + 256;
  for (uint32 i = 0; i < history_count; i++)
  {
    capacity +=
        strlen(escaped_hist_role[i]) + strlen(escaped_hist_content[i]) + 32;
  }

  char *payload = Dowa_Arena_Allocate(p_arena, capacity);
  if (!payload)
  {
    Dowa_Arena_Free(p_arena);
    return FALSE;
  }

  /* Write the base turn.start fields. */
  int written = snprintf(
      payload,
      capacity,
      "{\"command\":\"%s\",\"request_id\":\"%s\","
      "\"conversation_id\":\"%s\",\"prompt\":\"%s\","
      "\"prompt_profile\":\"%s\",\"prompt_version\":%u,"
      "\"knowledge_version\":%u,\"history\":[",
      escaped_command,
      escaped_request,
      escaped_conversation,
      escaped_prompt ? escaped_prompt : "",
      escaped_profile,
      prompt_version,
      knowledge_version);
  if (written <= 0 || (size_t)written >= capacity)
  {
    Dowa_Arena_Free(p_arena);
    return FALSE;
  }
  size_t offset = (size_t)written;

  /* Append history entries. */
  for (uint32 i = 0; i < history_count; i++)
  {
    int entry_written = snprintf(
        payload + offset,
        capacity - offset,
        "%s{\"role\":\"%s\",\"content\":\"%s\"}",
        i == 0 ? "" : ",",
        escaped_hist_role[i],
        escaped_hist_content[i]);
    if (entry_written <= 0 || offset + (size_t)entry_written >= capacity)
    {
      Dowa_Arena_Free(p_arena);
      return FALSE;
    }
    offset += (size_t)entry_written;
  }

  /* Close the history array and the outer object. */
  if (offset + 2 >= capacity)
  {
    Dowa_Arena_Free(p_arena);
    return FALSE;
  }
  payload[offset++] = ']';
  payload[offset++] = '}';
  payload[offset]   = '\0';

  boolean success = Inference_Bridge_Write(p_bridge, payload);
  Dowa_Arena_Free(p_arena);
  return success;
}

boolean Inference_Bridge_Abort_Turn(
    Inference_Bridge *p_bridge,
    const char *request_id,
    const char *conversation_id)
{
  return Inference_Bridge_Command(
      p_bridge, "turn.abort", request_id, conversation_id, NULL, NULL, 0, 0);
}

boolean Inference_Bridge_Delete_Conversation(
    Inference_Bridge *p_bridge,
    const char *request_id,
    const char *conversation_id)
{
  return Inference_Bridge_Command(
      p_bridge,
      "conversation.delete",
      request_id,
      conversation_id,
      NULL,
      NULL,
      0,
      0);
}

void Inference_Bridge_Destroy(Inference_Bridge *p_bridge)
{
  if (!p_bridge)
    return;
  Inference_Bridge_Stop_Process(p_bridge);
  if (p_bridge->ready_condition_initialized)
    pthread_cond_destroy(&p_bridge->ready_condition);
  if (p_bridge->ready_mutex_initialized)
    pthread_mutex_destroy(&p_bridge->ready_mutex);
  if (p_bridge->write_mutex_initialized)
    pthread_mutex_destroy(&p_bridge->write_mutex);
  free(p_bridge->sidecar_path);
  free(p_bridge->copilot_cli_path);
  free(p_bridge);
}