// G.EXE - ask a Gradio Space one question, over plain HTTP, from DOS.
//
//   G <question words...>
//   environment: GRADIO_HOST (default tencent-hy3.hf.space)
//
// It speaks the documented Gradio HTTP API itself:
//   1) POST /gradio_api/call/chat   {"data":[...9 parameters...]}   -> {"event_id":"..."}
//   2) GET  /gradio_api/call/chat/<event_id>                        -> server-sent events
//      ("generating" frames while the model streams, then one "complete" frame)
// DOS has no TLS, so the page's bridge (gradio-bridge.js) forwards these requests over HTTPS.
#include "mqtt.h"

#define DEFAULT_HOST "tencent-hy3.hf.space"
#define SYSTEM_PROMPT "Answer in plain ASCII English in at most 40 words."
#define LINE_MAX_LEN 2000
// The emulator's TCP sends 1460-byte segments back to back, ignores the receive window and never
// retransmits, so a segment that does not fit in our buffer is lost for good and the connection hangs.
// Keep the receive buffer several segments large and drain it quickly.
#define RECV_BUF 8192

static char line[LINE_MAX_LEN];
static int  lineLen = 0;
static char finalData[LINE_MAX_LEN];
static char evt[16];
static int  haveFinal = 0, sawError = 0, frames = 0;
static char statusLine[48];
static int  statusDone = 0;

static char resp[1024];
static int  respLen = 0;

static void handleLine( void ) {
  if ( !statusDone ) {                               // first line of an HTTP response
    strncpy( statusLine, line, sizeof( statusLine ) - 1 );
    statusDone = 1;
    return;
  }
  if ( strncmp( line, "event: ", 7 ) == 0 ) {
    strncpy( evt, line + 7, sizeof( evt ) - 1 );
    evt[sizeof( evt ) - 1] = 0;
    if ( strcmp( evt, "generating" ) == 0 ) { frames++; printf( "." ); fflush( stdout ); }
    if ( strcmp( evt, "error" ) == 0 ) sawError = 1;
  }
  else if ( strncmp( line, "data: ", 6 ) == 0 && strcmp( evt, "complete" ) == 0 ) {
    strcpy( finalData, line + 6 );
    haveFinal = 1;
  }
}

static void feed( const uint8_t *p, int n, int sse ) {
  for ( int i = 0; i < n; i++ ) {
    if ( !sse ) { if ( respLen < (int)sizeof( resp ) - 1 ) resp[respLen++] = (char)p[i]; continue; }
    char c = (char)p[i];
    if ( c == '\n' ) { line[lineLen] = 0; handleLine( ); lineLen = 0; }
    else if ( c != '\r' && lineLen < LINE_MAX_LEN - 1 ) line[lineLen++] = c;
  }
}

static uint16_t portCounter = 0;

// One HTTP request/response on a fresh TCP connection. The response is read until the server closes.
static int httpExchange( IpAddr_t addr, const char *req, int sse, uint32_t timeoutSec ) {
  TcpSocket *s = TcpSocketMgr::getSocket( );
  if ( s->setRecvBuffer( RECV_BUF ) ) { printf( "TCP: no memory for the receive buffer\n" ); TcpSocketMgr::freeSocket( s ); return -1; }
  portCounter += 17;
  if ( s->connect( (uint16_t)( pseudoPort( ) + portCounter ), addr, 80, 10000 ) != 0 ) {
    TcpSocketMgr::freeSocket( s );
    printf( "TCP: connect to port 80 failed\n" );
    return -1;
  }
  lineLen = 0; respLen = 0; statusDone = 0; statusLine[0] = 0; evt[0] = 0;
  s->send( (uint8_t *)req, (uint16_t)strlen( req ) );

  static uint8_t buf[1024];
  clockTicks_t start = TIMER_GET_CURRENT( );
  int closed = 0, rc = 0;
  while ( !CtrlBreakDetected ) {
    Utils::processPackets( );
    int16_t r = s->recv( buf, sizeof( buf ) );
    if ( r > 0 ) { feed( buf, r, sse ); continue; }
    if ( closed ) break;
    if ( s->isRemoteClosed( ) ) { closed = 1; continue; }   // one more pass to drain the buffer
    if ( Timer_diff( start, TIMER_GET_CURRENT( ) ) > TIMER_SECS_TO_TICKS( timeoutSec ) ) {
      printf( "\nHTTP: timed out after %lu s\n", (unsigned long)timeoutSec ); rc = -2; break;
    }
  }
  s->close( );
  TcpSocketMgr::freeSocket( s );
  if ( sse && lineLen > 0 ) { line[lineLen] = 0; handleLine( ); }
  return rc;
}

// Decode the first JSON string that follows `[["` in the SSE data line and print it word wrapped.
static void printAnswer( const char *data ) {
  const char *p = strstr( data, "[[\"" );
  if ( !p ) { printf( "A> (no text in reply: %.60s)\n", data ); return; }
  p += 3;
  char out[1200]; int n = 0;
  while ( *p && n < (int)sizeof( out ) - 2 ) {
    if ( *p == '"' ) break;
    if ( *p == '\\' ) {
      p++;
      if ( *p == 'n' ) out[n++] = ' ';
      else if ( *p == 'u' ) {
        unsigned v = 0; int k;
        for ( k = 1; k <= 4 && p[k]; k++ ) { char h = p[k]; v = v * 16 + ( h <= '9' ? h - '0' : ( h | 32 ) - 'a' + 10 ); }
        out[n++] = ( v >= 32 && v < 127 ) ? (char)v : '?';
        p += 4;
      }
      else out[n++] = *p;
      if ( *p ) p++;
      continue;
    }
    out[n++] = *p++;
  }
  out[n] = 0;
  printf( "A> " );
  int col = 3;
  for ( char *w = strtok( out, " " ); w; w = strtok( NULL, " " ) ) {
    int l = (int)strlen( w );
    if ( col + l + 1 > 78 ) { printf( "\n   " ); col = 3; }
    printf( "%s ", w ); col += l + 1;
  }
  printf( "\n" );
}

int main( int argc, char *argv[] ) {
  char question[160]; question[0] = 0;
  for ( int i = 1; i < argc; i++ ) {
    if ( strlen( question ) + strlen( argv[i] ) + 2 > sizeof( question ) ) break;
    if ( i > 1 ) strcat( question, " " );
    strcat( question, argv[i] );
  }
  for ( char *q = question; *q; q++ ) if ( *q < 32 || *q > 126 || *q == '"' || *q == '\\' ) *q = ' ';
  if ( !question[0] ) { printf( "Usage: G <question>\n   env GRADIO_HOST (default %s)\n", DEFAULT_HOST ); return 1; }

  const char *host = getenv( "GRADIO_HOST" );
  if ( !host || !*host ) host = DEFAULT_HOST;

  printf( "G.EXE - asking Gradio Space %s\nQ> %s\n", host, question );
  if ( netInit( ) ) { printf( "TCP/IP init failed (is the packet driver loaded?)\n" ); return 2; }

  IpAddr_t addr;
  printf( "DNS: resolving %s ...\n", host );
  if ( resolveHost( host, addr ) ) { printf( "DNS: failed\n" ); netShutdown( 3 ); }
  printf( "DNS: %d.%d.%d.%d\n", addr[0], addr[1], addr[2], addr[3] );

  clockTicks_t t0 = TIMER_GET_CURRENT( );
  char body[400], req[700], now[16];
  sprintf( body, "{\"data\":[\"%s\",\"%s\",[],\"no_think\",null,0,0,null,\"\"]}", question, SYSTEM_PROMPT );
  sprintf( req, "POST /gradio_api/call/chat HTTP/1.0\r\nHost: %s\r\nContent-Type: application/json\r\nContent-Length: %d\r\nConnection: close\r\n\r\n%s",
           host, (int)strlen( body ), body );
  stamp( now );
  printf( "[%s] POST /gradio_api/call/chat (%d bytes)\n", now, (int)strlen( body ) );
  if ( httpExchange( addr, req, 0, 30 ) ) netShutdown( 4 );
  resp[respLen] = 0;
  printf( "HTTP: %.40s\n", resp );

  char *e = strstr( resp, "\"event_id\":\"" );
  if ( !e ) { printf( "No event_id in reply:\n%.300s\n", resp ); netShutdown( 5 ); }
  e += 12;
  char id[64]; int k = 0;
  while ( *e && *e != '"' && k < 62 ) id[k++] = *e++;
  id[k] = 0;
  printf( "event_id: %s\n", id );

  sprintf( req, "GET /gradio_api/call/chat/%s HTTP/1.0\r\nHost: %s\r\nAccept: text/event-stream\r\nConnection: close\r\n\r\n", id, host );
  stamp( now );
  printf( "[%s] GET  /gradio_api/call/chat/<event_id>  (waiting for the model) ", now );
  fflush( stdout );
  if ( httpExchange( addr, req, 1, 150 ) ) netShutdown( 6 );
  printf( "\n" );

  stamp( now );
  if ( !haveFinal ) { printf( "[%s] No answer received (%s, events: %s)\n", now, statusLine, sawError ? "error" : "none" ); netShutdown( 7 ); }
  unsigned long secs = ( TIMER_GET_CURRENT( ) - t0 ) / 18;
  printf( "[%s] %d streaming frames, done in %lu s\n", now, frames, secs );
  printAnswer( finalData );
  netShutdown( 0 );
  return 0;
}