/*
** String library.
** Copyright (C) 2005-2017 Mike Pall. See Copyright Notice in luajit.h
**
** Major portions taken verbatim or adapted from the Lua interpreter.
** Copyright (C) 1994-2008 Lua.org, PUC-Rio. See Copyright Notice in lua.h
*/

#define lib_string_c
#define LUA_LIB

#include "lua.h"
#include "lauxlib.h"
#include "lualib.h"

#include "lj_obj.h"
#include "lj_gc.h"
#include "lj_err.h"
#include "lj_buf.h"
#include "lj_str.h"
#include "lj_tab.h"
#include "lj_meta.h"
#include "lj_state.h"
#include "lj_ff.h"
#include "lj_bcdump.h"
#include "lj_char.h"
#include "lj_strfmt.h"
#include "lj_lib.h"

/* ------------------------------------------------------------------------ */

#define LJLIB_MODULE_string

LJLIB_LUA(string_len) /*
  function(s)
    CHECK_str(s)
    return #s
  end
*/

LJLIB_ASM(string_byte)		LJLIB_REC(string_range 0)
{
  GCstr *s = lj_lib_checkstr(L, 1);
  int32_t len = (int32_t)s->len;
  int32_t start = lj_lib_optint(L, 2, 1);
  int32_t stop = lj_lib_optint(L, 3, start);
  int32_t n, i;
  const unsigned char *p;
  if (stop < 0) stop += len+1;
  if (start < 0) start += len+1;
  if (start <= 0) start = 1;
  if (stop > len) stop = len;
  if (start > stop) return FFH_RES(0);  /* Empty interval: return no results. */
  start--;
  n = stop - start;
  if ((uint32_t)n > LUAI_MAXCSTACK)
    lj_err_caller(L, LJ_ERR_STRSLC);
  lj_state_checkstack(L, (MSize)n);
  p = (const unsigned char *)strdata(s) + start;
  for (i = 0; i < n; i++)
    setintV(L->base + i-1-LJ_FR2, p[i]);
  return FFH_RES(n);
}

LJLIB_ASM(string_char)		LJLIB_REC(.)
{
  int i, nargs = (int)(L->top - L->base);
  char *buf = lj_buf_tmp(L, (MSize)nargs);
  for (i = 1; i <= nargs; i++) {
    int32_t k = lj_lib_checkint(L, i);
    if (!checku8(k))
      lj_err_arg(L, i, LJ_ERR_BADVAL);
    buf[i-1] = (char)k;
  }
  setstrV(L, L->base-1-LJ_FR2, lj_str_new(L, buf, (size_t)nargs));
  return FFH_RES(1);
}

LJLIB_ASM(string_sub)		LJLIB_REC(string_range 1)
{
  lj_lib_checkstr(L, 1);
  lj_lib_checkint(L, 2);
  setintV(L->base+2, lj_lib_optint(L, 3, -1));
  return FFH_RETRY;
}

LJLIB_CF(string_rep)		LJLIB_REC(.)
{
  GCstr *s = lj_lib_checkstr(L, 1);
  int32_t rep = lj_lib_checkint(L, 2);
  GCstr *sep = lj_lib_optstr(L, 3);
  SBuf *sb = lj_buf_tmp_(L);
  if (sep && rep > 1) {
    GCstr *s2 = lj_buf_cat2str(L, sep, s);
    lj_buf_reset(sb);
    lj_buf_putstr(sb, s);
    s = s2;
    rep--;
  }
  sb = lj_buf_putstr_rep(sb, s, rep);
  setstrV(L, L->top-1, lj_buf_str(L, sb));
  lj_gc_check(L);
  return 1;
}

LJLIB_ASM(string_reverse)  LJLIB_REC(string_op IRCALL_lj_buf_putstr_reverse)
{
  lj_lib_checkstr(L, 1);
  return FFH_RETRY;
}
LJLIB_ASM_(string_lower)  LJLIB_REC(string_op IRCALL_lj_buf_putstr_lower)
LJLIB_ASM_(string_upper)  LJLIB_REC(string_op IRCALL_lj_buf_putstr_upper)

/* ------------------------------------------------------------------------ */

static int writer_buf(lua_State *L, const void *p, size_t size, void *sb)
{
  lj_buf_putmem((SBuf *)sb, p, (MSize)size);
  UNUSED(L);
  return 0;
}

LJLIB_CF(string_dump)
{
  GCfunc *fn = lj_lib_checkfunc(L, 1);
  int strip = L->base+1 < L->top && tvistruecond(L->base+1);
  SBuf *sb = lj_buf_tmp_(L);  /* Assumes lj_bcwrite() doesn't use tmpbuf. */
  L->top = L->base+1;
  if (!isluafunc(fn) || lj_bcwrite(L, funcproto(fn), writer_buf, sb, strip))
    lj_err_caller(L, LJ_ERR_STRDUMP);
  setstrV(L, L->top-1, lj_buf_str(L, sb));
  lj_gc_check(L);
  return 1;
}

/* ------------------------------------------------------------------------ */

/* macro to `unsign' a character */
#define uchar(c)        ((unsigned char)(c))

#define CAP_UNFINISHED	(-1)
#define CAP_POSITION	(-2)

typedef struct MatchState {
  const char *src_init;  /* init of source string */
  const char *src_end;  /* end ('\0') of source string */
  const char *p_end;  /* end ('\0') of pattern */
  lua_State *L;
  int level;  /* total number of captures (finished or unfinished) */
  int depth;
  struct {
    const char *init;
    ptrdiff_t len;
  } capture[LUA_MAXCAPTURES];
} MatchState;

#define L_ESC		'%'

static int check_capture(MatchState *ms, int l)
{
  l -= '1';
  if (l < 0 || l >= ms->level || ms->capture[l].len == CAP_UNFINISHED)
    lj_err_caller(ms->L, LJ_ERR_STRCAPI);
  return l;
}

static int capture_to_close(MatchState *ms)
{
  int level = ms->level;
  for (level--; level>=0; level--)
    if (ms->capture[level].len == CAP_UNFINISHED) return level;
  lj_err_caller(ms->L, LJ_ERR_STRPATC);
  return 0;  /* unreachable */
}

static const char *classend(MatchState *ms, const char *p)
{
  switch (*p++) {
  case L_ESC:
    if (p == ms->p_end)
      lj_err_caller(ms->L, LJ_ERR_STRPATE);
    return p+1;
  case '[':
    if (*p == '^') p++;
    do {  /* look for a `]' */
      if (p == ms->p_end)
	lj_err_caller(ms->L, LJ_ERR_STRPATM);
      if (*(p++) == L_ESC && p < ms->p_end)
	p++;  /* skip escapes (e.g. `%]') */
    } while (*p != ']');
    return p+1;
  default:
    return p;
  }
}

static const unsigned char match_class_map[32] = {
  0,LJ_CHAR_ALPHA,0,LJ_CHAR_CNTRL,LJ_CHAR_DIGIT,0,0,LJ_CHAR_GRAPH,0,0,0,0,
  LJ_CHAR_LOWER,0,0,0,LJ_CHAR_PUNCT,0,0,LJ_CHAR_SPACE,0,
  LJ_CHAR_UPPER,0,LJ_CHAR_ALNUM,LJ_CHAR_XDIGIT,0,0,0,0,0,0,0
};

static int match_class(int c, int cl)
{
  if ((cl & 0xc0) == 0x40) {
    int t = match_class_map[(cl&0x1f)];
    if (t) {
      t = lj_char_isa(c, t);
      return (cl & 0x20) ? t : !t;
    }
    /* deprecated option in Lua 5.2 */
    if (cl == 'z') return c == 0;
    if (cl == 'Z') return c != 0;
  }
  return (cl == c);
}

static int matchbracketclass(int c, const char *p, const char *ec)
{
  int sig = 1;
  if (*(p+1) == '^') {
    sig = 0;
    p++;  /* skip the `^' */
  }
  while (++p < ec) {
    if (*p == L_ESC) {
      p++;
      if (match_class(c, uchar(*p)))
	return sig;
    }
    else if ((*(p+1) == '-') && (p+2 < ec)) {
      p+=2;
      if (uchar(*(p-2)) <= c && c <= uchar(*p))
	return sig;
    }
    else if (uchar(*p) == c) return sig;
  }
  return !sig;
}

static int singlematch(MatchState *ms, const char *s, const char *p,
		       const char *ep) {
  if (s >= ms->src_end)
    return 0;
  else {
    int c = uchar(*s);
    switch (*p) {
    case '.': return 1;  /* matches any char */
    case L_ESC: return match_class(c, uchar(*(p+1)));
    case '[': return matchbracketclass(c, p, ep-1);
    default:  return (uchar(*p) == c);
    }
  }
}

static const char *match(MatchState *ms, const char *s, const char *p);

static const char *matchbalance(MatchState *ms, const char *s, const char *p)
{
  if (p >= ms->p_end - 1)
    lj_err_caller(ms->L, LJ_ERR_STRPATU);
  if (*s != *p) {
    return NULL;
  } else {
    int b = *p;
    int e = *(p+1);
    int cont = 1;
    while (++s < ms->src_end) {
      if (*s == e) {
	if (--cont == 0) return s+1;
      } else if (*s == b) {
	cont++;
      }
    }
  }
  return NULL;  /* string ends out of balance */
}

static const char *max_expand(MatchState *ms, const char *s,
			      const char *p, const char *ep)
{
  ptrdiff_t i = 0;  /* counts maximum expand for item */
  while (singlematch(ms, s + i, p, ep))
    i++;
  /* keeps trying to match with the maximum repetitions */
  while (i>=0) {
    const char *res = match(ms, (s+i), ep+1);
    if (res) return res;
    i--;  /* else didn't match; reduce 1 repetition to try again */
  }
  return NULL;
}

static const char *min_expand(MatchState *ms, const char *s,
			      const char *p, const char *ep)
{
  for (;;) {
    const char *res = match(ms, s, ep+1);
    if (res != NULL)
      return res;
    else if (singlematch(ms, s, p, ep))
      s++;  /* try with one more repetition */
    else
      return NULL;
  }
}

static const char *start_capture(MatchState *ms, const char *s,
				 const char *p, int what)
{
  const char *res;
  int level = ms->level;
  if (level >= LUA_MAXCAPTURES) lj_err_caller(ms->L, LJ_ERR_STRCAPN);
  ms->capture[level].init = s;
  ms->capture[level].len = what;
  ms->level = level+1;
  if ((res=match(ms, s, p)) == NULL)  /* match failed? */
    ms->level--;  /* undo capture */
  return res;
}

static const char *end_capture(MatchState *ms, const char *s,
			       const char *p)
{
  int l = capture_to_close(ms);
  const char *res;
  ms->capture[l].len = s - ms->capture[l].init;  /* close capture */
  if ((res = match(ms, s, p)) == NULL)  /* match failed? */
    ms->capture[l].len = CAP_UNFINISHED;  /* undo capture */
  return res;
}

static const char *match_capture(MatchState *ms, const char *s, int l)
{
  size_t len;
  l = check_capture(ms, l);
  len = (size_t)ms->capture[l].len;
  if ((size_t)(ms->src_end-s) >= len &&
      memcmp(ms->capture[l].init, s, len) == 0)
    return s+len;
  else
    return NULL;
}

static const char *match(MatchState *ms, const char *s, const char *p)
{
  if (++ms->depth > LJ_MAX_XLEVEL)
    lj_err_caller(ms->L, LJ_ERR_STRPATX);
  init: /* using goto's to optimize tail recursion */
  if (p != ms->p_end) {  /* end of pattern? */
    switch (*p) {
    case '(':  /* start capture */
      if (*(p+1) == ')')  /* position capture? */
	s = start_capture(ms, s, p+2, CAP_POSITION);
      else
	s = start_capture(ms, s, p+1, CAP_UNFINISHED);
      break;
    case ')':  /* end capture */
      s = end_capture(ms, s, p+1);
      break;
    case L_ESC:
      switch (*(p+1)) {
      case 'b':  /* balanced string? */
	s = matchbalance(ms, s, p+2);
	if (s == NULL) break;
	p+=4;
	goto init;  /* else s = match(ms, s, p+4); */
      case 'f': {  /* frontier? */
	const char *ep; char previous;
	p += 2;
	if (*p != '[')
	  lj_err_caller(ms->L, LJ_ERR_STRPATB);
	ep = classend(ms, p);  /* points to what is next */
	previous = (s == ms->src_init) ? '\0' : *(s-1);
	if (matchbracketclass(uchar(previous), p, ep-1) ||
	  !matchbracketclass(uchar(*s), p, ep-1)) { s = NULL; break; }
	p=ep;
	goto init;  /* else s = match(ms, s, ep); */
	}
      default:
	if (lj_char_isdigit(uchar(*(p+1)))) {  /* capture results (%0-%9)? */
	  s = match_capture(ms, s, uchar(*(p+1)));
	  if (s == NULL) break;
	  p+=2;
	  goto init;  /* else s = match(ms, s, p+2) */
	}
	goto dflt;  /* case default */
      }
      break;
    case '$':
      /* is the `$' the last char in pattern? */
      if ((p+1) != ms->p_end) goto dflt;
      if (s != ms->src_end) s = NULL;  /* check end of string */
      break;
    default: dflt: {  /* it is a pattern item */
      const char *ep = classend(ms, p);  /* points to what is next */
      if (!singlematch(ms, s, p, ep)) {
        if (*ep == '*' || *ep == '?' || *ep == '-') {  /* accept empty? */
          p = ep + 1; goto init;  /* return match(ms, s, ep + 1); */
        }
        else  /* '+' or no suffix */
          s = NULL;  /* fail */
      }
      else {  /* matched once */
	switch (*ep) {
	case '?': {  /* optional */
	  const char *res;
	  if ((res=match(ms, s+1, ep+1)) != NULL)
	    s = res;
	  else {
	    p=ep+1;
	    goto init;  /* else return match(ms, s, ep + 1); */
	  }
	  break;
	}
	case '+':  /* 1 or more repetitions */
	  s++;  /* 1 match already done */
	  /* go through */
	case '*':  /* 0 or more repetitions */
	  s = max_expand(ms, s, p, ep);
	  break;
	case '-':  /* 0 or more repetitions (minimum) */
	  s = min_expand(ms, s, p, ep);
	  break;
	default:  /* no suffix */
	  s++; p = ep; goto init;  /* return match(ms, s + 1, ep); */
	}
      }
      break;
      }
    }
  }
  ms->depth--;
  return s;
}

static void push_onecapture(MatchState *ms, int i, const char *s, const char *e)
{
  if (i >= ms->level) {
    if (i == 0)  /* ms->level == 0, too */
      lua_pushlstring(ms->L, s, (size_t)(e - s));  /* add whole match */
    else
      lj_err_caller(ms->L, LJ_ERR_STRCAPI);
  } else {
    ptrdiff_t l = ms->capture[i].len;
    if (l == CAP_UNFINISHED) lj_err_caller(ms->L, LJ_ERR_STRCAPU);
    if (l == CAP_POSITION)
      lua_pushinteger(ms->L, ms->capture[i].init - ms->src_init + 1);
    else
      lua_pushlstring(ms->L, ms->capture[i].init, (size_t)l);
  }
}

static int push_captures(MatchState *ms, const char *s, const char *e)
{
  int i;
  int nlevels = (ms->level == 0 && s) ? 1 : ms->level;
  luaL_checkstack(ms->L, nlevels, "too many captures");
  for (i = 0; i < nlevels; i++)
    push_onecapture(ms, i, s, e);
  return nlevels;  /* number of strings pushed */
}

static int str_find_aux(lua_State *L, int find)
{
  GCstr *s = lj_lib_checkstr(L, 1);
  GCstr *p = lj_lib_checkstr(L, 2);
  int32_t start = lj_lib_optint(L, 3, 1);
  MSize st;
  if (start < 0) start += (int32_t)s->len; else start--;
  if (start < 0) start = 0;
  st = (MSize)start;
  if (st > s->len) {
#if LJ_52
    setnilV(L->top-1);
    return 1;
#else
    st = s->len;
#endif
  }
  if (find && ((L->base+3 < L->top && tvistruecond(L->base+3)) ||
	       !lj_str_haspattern(p))) {  /* Search for fixed string. */
    const char *q = lj_str_find(strdata(s)+st, strdata(p), s->len-st, p->len);
    if (q) {
      setintV(L->top-2, (int32_t)(q-strdata(s)) + 1);
      setintV(L->top-1, (int32_t)(q-strdata(s)) + (int32_t)p->len);
      return 2;
    }
  } else {  /* Search for pattern. */
    MatchState ms;
    const char *pstr = strdata(p);
    const char *sstr = strdata(s) + st;
    int anchor = 0;
    if (*pstr == '^') { pstr++; anchor = 1; }
    ms.L = L;
    ms.src_init = strdata(s);
    ms.src_end = strdata(s) + s->len;
    ms.p_end = strdata(p) + p->len;
    do {  /* Loop through string and try to match the pattern. */
      const char *q;
      ms.level = ms.depth = 0;
      q = match(&ms, sstr, pstr);
      if (q) {
	if (find) {
	  setintV(L->top++, (int32_t)(sstr-(strdata(s)-1)));
	  setintV(L->top++, (int32_t)(q-strdata(s)));
	  return push_captures(&ms, NULL, NULL) + 2;
	} else {
	  return push_captures(&ms, sstr, q);
	}
      }
    } while (sstr++ < ms.src_end && !anchor);
  }
  setnilV(L->top-1);  /* Not found. */
  return 1;
}

LJLIB_CF(string_find)		LJLIB_REC(.)
{
  return str_find_aux(L, 1);
}

LJLIB_CF(string_match)
{
  return str_find_aux(L, 0);
}

LJLIB_NOREG LJLIB_CF(string_gmatch_aux)
{
  GCstr *pat = strV(lj_lib_upvalue(L, 2));
  const char *p = strdata(pat);
  GCstr *str = strV(lj_lib_upvalue(L, 1));
  const char *s = strdata(str);
  TValue *tvpos = lj_lib_upvalue(L, 3);
  const char *src = s + tvpos->u32.lo;
  MatchState ms;
  ms.L = L;
  ms.src_init = s;
  ms.src_end = s + str->len;
  ms.p_end = p + pat->len;
  for (; src <= ms.src_end; src++) {
    const char *e;
    ms.level = ms.depth = 0;
    if ((e = match(&ms, src, p)) != NULL) {
      int32_t pos = (int32_t)(e - s);
      if (e == src) pos++;  /* Ensure progress for empty match. */
      tvpos->u32.lo = (uint32_t)pos;
      return push_captures(&ms, src, e);
    }
  }
  return 0;  /* not found */
}

LJLIB_CF(string_gmatch)
{
  lj_lib_checkstr(L, 1);
  lj_lib_checkstr(L, 2);
  L->top = L->base+3;
  (L->top-1)->u64 = 0;
  lj_lib_pushcc(L, lj_cf_string_gmatch_aux, FF_string_gmatch_aux, 3);
  return 1;
}

static void add_s(MatchState *ms, luaL_Buffer *b, const char *s, const char *e)
{
  size_t l, i;
  const char *news = lua_tolstring(ms->L, 3, &l);
  for (i = 0; i < l; i++) {
    if (news[i] != L_ESC) {
      luaL_addchar(b, news[i]);
    } else {
      i++;  /* skip ESC */
      if (!lj_char_isdigit(uchar(news[i]))) {
	luaL_addchar(b, news[i]);
      } else if (news[i] == '0') {
	luaL_addlstring(b, s, (size_t)(e - s));
      } else {
	push_onecapture(ms, news[i] - '1', s, e);
	luaL_addvalue(b);  /* add capture to accumulated result */
      }
    }
  }
}

static void add_value(MatchState *ms, luaL_Buffer *b,
		      const char *s, const char *e)
{
  lua_State *L = ms->L;
  switch (lua_type(L, 3)) {
    case LUA_TNUMBER:
    case LUA_TSTRING: {
      add_s(ms, b, s, e);
      return;
    }
    case LUA_TFUNCTION: {
      int n;
      lua_pushvalue(L, 3);
      n = push_captures(ms, s, e);
      lua_call(L, n, 1);
      break;
    }
    case LUA_TTABLE: {
      push_onecapture(ms, 0, s, e);
      lua_gettable(L, 3);
      break;
    }
  }
  if (!lua_toboolean(L, -1)) {  /* nil or false? */
    lua_pop(L, 1);
    lua_pushlstring(L, s, (size_t)(e - s));  /* keep original text */
  } else if (!lua_isstring(L, -1)) {
    lj_err_callerv(L, LJ_ERR_STRGSRV, luaL_typename(L, -1));
  }
  luaL_addvalue(b);  /* add result to accumulator */
}

LJLIB_CF(string_gsub)
{
  size_t srcl, lp;
  const char *src = luaL_checklstring(L, 1, &srcl);
  const char *p = luaL_checklstring(L, 2, &lp);
  int  tr = lua_type(L, 3);
  int max_s = luaL_optint(L, 4, (int)(srcl+1));
  int anchor = (*p == '^');
  int n = 0;
  MatchState ms;
  luaL_Buffer b;
  if (!(tr == LUA_TNUMBER || tr == LUA_TSTRING ||
	tr == LUA_TFUNCTION || tr == LUA_TTABLE))
    lj_err_arg(L, 3, LJ_ERR_NOSFT);
  luaL_buffinit(L, &b);
  if (anchor) {
    p++; lp--;  /* skip anchor character */
  }
  ms.L = L;
  ms.src_init = src;
  ms.src_end = src+srcl;
  ms.p_end = p + lp;
  while (n < max_s) {
    const char *e;
    ms.level = ms.depth = 0;
    e = match(&ms, src, p);
    if (e) {
      n++;
      add_value(&ms, &b, src, e);
    }
    if (e && e>src) /* non empty match? */
      src = e;  /* skip it */
    else if (src < ms.src_end)
      luaL_addchar(&b, *src++);
    else
      break;
    if (anchor)
      break;
  }
  luaL_addlstring(&b, src, (size_t)(ms.src_end-src));
  luaL_pushresult(&b);
  lua_pushinteger(L, n);  /* number of substitutions */
  return 2;
}

/* ------------------------------------------------------------------------ */

/* Emulate tostring() inline. */
static GCstr *string_fmt_tostring(lua_State *L, int arg, int retry)
{
  TValue *o = L->base+arg-1;
  cTValue *mo;
  lua_assert(o < L->top);  /* Caller already checks for existence. */
  if (LJ_LIKELY(tvisstr(o)))
    return strV(o);
  if (retry != 2 && !tvisnil(mo = lj_meta_lookup(L, o, MM_tostring))) {
    copyTV(L, L->top++, mo);
    copyTV(L, L->top++, o);
    lua_call(L, 1, 1);
    copyTV(L, L->base+arg-1, --L->top);
    return NULL;  /* Buffer may be overwritten, retry. */
  }
  return lj_strfmt_obj(L, o);
}

LJLIB_CF(string_format)		LJLIB_REC(.)
{
  int arg, top = (int)(L->top - L->base);
  GCstr *fmt;
  SBuf *sb;
  FormatState fs;
  SFormat sf;
  int retry = 0;
again:
  arg = 1;
  sb = lj_buf_tmp_(L);
  fmt = lj_lib_checkstr(L, arg);
  lj_strfmt_init(&fs, strdata(fmt), fmt->len);
  while ((sf = lj_strfmt_parse(&fs)) != STRFMT_EOF) {
    if (sf == STRFMT_LIT) {
      lj_buf_putmem(sb, fs.str, fs.len);
    } else if (sf == STRFMT_ERR) {
      lj_err_callerv(L, LJ_ERR_STRFMT, strdata(lj_str_new(L, fs.str, fs.len)));
    } else {
      if (++arg > top)
	luaL_argerror(L, arg, lj_obj_typename[0]);
      switch (STRFMT_TYPE(sf)) {
      case STRFMT_INT:
	if (tvisint(L->base+arg-1)) {
	  int32_t k = intV(L->base+arg-1);
	  if (sf == STRFMT_INT)
	    lj_strfmt_putint(sb, k);  /* Shortcut for plain %d. */
	  else
	    lj_strfmt_putfxint(sb, sf, k);
	} else {
	  lj_strfmt_putfnum_int(sb, sf, lj_lib_checknum(L, arg));
	}
	break;
      case STRFMT_UINT:
	if (tvisint(L->base+arg-1))
	  lj_strfmt_putfxint(sb, sf, intV(L->base+arg-1));
	else
	  lj_strfmt_putfnum_uint(sb, sf, lj_lib_checknum(L, arg));
	break;
      case STRFMT_NUM:
	lj_strfmt_putfnum(sb, sf, lj_lib_checknum(L, arg));
	break;
      case STRFMT_STR: {
	GCstr *str = string_fmt_tostring(L, arg, retry);
	if (str == NULL)
	  retry = 1;
	else if ((sf & STRFMT_T_QUOTED))
	  lj_strfmt_putquoted(sb, str);  /* No formatting. */
	else
	  lj_strfmt_putfstr(sb, sf, str);
	break;
	}
      case STRFMT_CHAR:
	lj_strfmt_putfchar(sb, sf, lj_lib_checkint(L, arg));
	break;
      case STRFMT_PTR:  /* No formatting. */
	lj_strfmt_putptr(sb, lj_obj_ptr(L->base+arg-1));
	break;
      default:
	lua_assert(0);
	break;
      }
    }
  }
  if (retry++ == 1) goto again;
  setstrV(L, L->top-1, lj_buf_str(L, sb));
  lj_gc_check(L);
  return 1;
}

/* ------------------------------------------------------------------------ */

/*
** Some sizes are better limited to fit in 'int', but must also fit in
** 'size_t'. (We assume that 'lua_Integer' cannot be smaller than 'int'.)
*/
#define MAXSIZE  \
	(sizeof(size_t) < sizeof(int) ? SIZE_MAX : (size_t)(INT_MAX))

/*
** {======================================================
** PACK/UNPACK
** =======================================================
*/


/* value used for padding */
#if !defined(LUAL_PACKPADBYTE)
#define LUAL_PACKPADBYTE		0x00
#endif

/* maximum size for the binary representation of an integer */
#define MAXINTSIZE	16

/* number of bits in a character */
#define NB	CHAR_BIT

/* mask for one character (NB 1's) */
#define MC	((1 << NB) - 1)

/* size of a lua_Integer */
#define SZINT	((int)sizeof(lua_Integer))


/* dummy structure to get native alignment requirements */
struct cD {
  char c;
  union { double d; void *p; lua_Integer i; lua_Number n; } u;
};

#define MAXALIGN	(offsetof(struct cD, u))


/*
** Union for serializing floats
*/
typedef union Ftypes {
  float f;
  double d;
  lua_Number n;
  char buff[5 * sizeof(lua_Number)];  /* enough for any float type */
} Ftypes;


/*
** information to pack/unpack stuff
*/
typedef struct Header {
  lua_State *L;
  int islittle;
  int maxalign;
} Header;


/*
** options for pack/unpack
*/
typedef enum KOption {
  Kint,		/* signed integers */
  Kuint,	/* unsigned integers */
  Kfloat,	/* floating-point numbers */
  Kchar,	/* fixed-length strings */
  Kstring,	/* strings with prefixed length */
  Kzstr,	/* zero-terminated strings */
  Kpadding,	/* padding */
  Kpaddalign,	/* padding for alignment */
  Knop		/* no-op (configuration or spaces) */
} KOption;


/*
** Read an integer numeral from string 'fmt' or return 'df' if
** there is no numeral
*/
static int digit (int c) { return '0' <= c && c <= '9'; }

static int getnum (const char **fmt, int df) {
  if (!digit(**fmt))  /* no number? */
    return df;  /* return default value */
  else {
    int a = 0;
    do {
      a = a*10 + (*((*fmt)++) - '0');
    } while (digit(**fmt) && a <= ((int)MAXSIZE - 9)/10);
    return a;
  }
}


/*
** Read an integer numeral and raises an error if it is larger
** than the maximum size for integers.
*/
static int getnumlimit (Header *h, const char **fmt, int df) {
  int sz = getnum(fmt, df);
  if (sz > MAXINTSIZE || sz <= 0)
    return luaL_error(h->L, "integral size (%d) out of limits [1,%d]",
                            sz, MAXINTSIZE);
  return sz;
}


/*
** Initialize Header
*/
static void initheader (lua_State *L, Header *h) {
  h->L = L;
  h->islittle = LJ_LE;
  h->maxalign = 1;
}


/*
** Read and classify next option. 'size' is filled with option's size.
*/
static KOption getoption (Header *h, const char **fmt, int *size) {
  int opt = *((*fmt)++);
  *size = 0;  /* default */
  switch (opt) {
    case 'b': *size = sizeof(char); return Kint;
    case 'B': *size = sizeof(char); return Kuint;
    case 'h': *size = sizeof(short); return Kint;
    case 'H': *size = sizeof(short); return Kuint;
    case 'l': *size = sizeof(long); return Kint;
    case 'L': *size = sizeof(long); return Kuint;
    case 'j': *size = sizeof(lua_Integer); return Kint;
    case 'J': *size = sizeof(lua_Integer); return Kuint;
    case 'T': *size = sizeof(size_t); return Kuint;
    case 'f': *size = sizeof(float); return Kfloat;
    case 'd': *size = sizeof(double); return Kfloat;
    case 'n': *size = sizeof(lua_Number); return Kfloat;
    case 'i': *size = getnumlimit(h, fmt, sizeof(int)); return Kint;
    case 'I': *size = getnumlimit(h, fmt, sizeof(int)); return Kuint;
    case 's': *size = getnumlimit(h, fmt, sizeof(size_t)); return Kstring;
    case 'c':
      *size = getnum(fmt, -1);
      if (*size == -1)
        luaL_error(h->L, "missing size for format option 'c'");
      return Kchar;
    case 'z': return Kzstr;
    case 'x': *size = 1; return Kpadding;
    case 'X': return Kpaddalign;
    case ' ': break;
    case '<': h->islittle = 1; break;
    case '>': h->islittle = 0; break;
    case '=': h->islittle = LJ_LE; break;
    case '!': h->maxalign = getnumlimit(h, fmt, MAXALIGN); break;
    default: luaL_error(h->L, "invalid format option '%c'", opt);
  }
  return Knop;
}


/*
** Read, classify, and fill other details about the next option.
** 'psize' is filled with option's size, 'notoalign' with its
** alignment requirements.
** Local variable 'size' gets the size to be aligned. (Kpadal option
** always gets its full alignment, other options are limited by
** the maximum alignment ('maxalign'). Kchar option needs no alignment
** despite its size.
*/
static KOption getdetails (Header *h, size_t totalsize,
                           const char **fmt, int *psize, int *ntoalign) {
  KOption opt = getoption(h, fmt, psize);
  int align = *psize;  /* usually, alignment follows size */
  if (opt == Kpaddalign) {  /* 'X' gets alignment from following option */
    if (**fmt == '\0' || getoption(h, fmt, &align) == Kchar || align == 0)
      luaL_argerror(h->L, 1, "invalid next option for option 'X'");
  }
  if (align <= 1 || opt == Kchar)  /* need no alignment? */
    *ntoalign = 0;
  else {
    if (align > h->maxalign)  /* enforce maximum alignment */
      align = h->maxalign;
    if ((align & (align - 1)) != 0)  /* is 'align' not a power of 2? */
      luaL_argerror(h->L, 1, "format asks for alignment not power of 2");
    *ntoalign = (align - (int)(totalsize & (align - 1))) & (align - 1);
  }
  return opt;
}


/*
** Pack integer 'n' with 'size' bytes and 'islittle' endianness.
** The final 'if' handles the case when 'size' is larger than
** the size of a Lua integer, correcting the extra sign-extension
** bytes if necessary (by default they would be zeros).
*/
static void packint (SBuf *sb, lua_Unsigned n,
                     int islittle, int size, int neg) {
  char *buff = lj_buf_more(sb, size);
  int i;
  buff[islittle ? 0 : size - 1] = (char)(n & MC);  /* first byte */
  for (i = 1; i < size; i++) {
    n >>= NB;
    buff[islittle ? i : size - 1 - i] = (char)(n & MC);
  }
  if (neg && size > SZINT) {  /* negative number need sign extension? */
    for (i = SZINT; i < size; i++)  /* correct extra bytes */
      buff[islittle ? i : size - 1 - i] = (char)MC;
  }
  setsbufP(sb, buff + size);
}


/*
** Copy 'size' bytes from 'src' to 'dest', correcting endianness if
** given 'islittle' is different from native endianness.
*/
static void copywithendian (volatile char *dest, volatile const char *src,
                            int size, int islittle) {
  if (islittle == LJ_LE) {
    while (size-- != 0)
      *(dest++) = *(src++);
  }
  else {
    dest += size - 1;
    while (size-- != 0)
      *(dest--) = *(src++);
  }
}


LJLIB_CF(string_pack)
{
  SBuf *sb;
  Header h;
  const char *fmt = luaL_checkstring(L, 1);  /* format string */
  int arg = 1;  /* current argument to pack */
  size_t totalsize = 0;  /* accumulate total size of result */
  initheader(L, &h);
  sb = lj_buf_tmp_(L);
  while (*fmt != '\0') {
    int size, ntoalign;
    KOption opt = getdetails(&h, totalsize, &fmt, &size, &ntoalign);
    totalsize += ntoalign + size;
    while (ntoalign-- > 0)
     lj_buf_putchar(sb, LUAL_PACKPADBYTE);  /* fill alignment */
    arg++;
    switch (opt) {
      case Kint: {  /* signed integers */
        lua_Integer n = luaL_checkinteger(L, arg);
        if (size < SZINT) {  /* need overflow check? */
          lua_Integer lim = (lua_Integer)1 << ((size * NB) - 1);
          luaL_argcheck(L, -lim <= n && n < lim, arg, "integer overflow");
        }
        packint(sb, (lua_Unsigned)n, h.islittle, size, (n < 0));
        break;
      }
      case Kuint: {  /* unsigned integers */
        lua_Integer n = luaL_checkinteger(L, arg);
        if (size < SZINT)  /* need overflow check? */
          luaL_argcheck(L, (lua_Unsigned)n < ((lua_Unsigned)1 << (size * NB)),
                           arg, "unsigned overflow");
        packint(sb, (lua_Unsigned)n, h.islittle, size, 0);
        break;
      }
      case Kfloat: {  /* floating-point options */
        volatile Ftypes u;
        char *buff = lj_buf_more(sb, size);
        lua_Number n = luaL_checknumber(L, arg);  /* get argument */
        if (size == sizeof(u.f)) u.f = (float)n;  /* copy it into 'u' */
        else if (size == sizeof(u.d)) u.d = (double)n;
        else u.n = n;
        /* move 'u' to final result, correcting endianness if needed */
        copywithendian(buff, u.buff, size, h.islittle);
        setsbufP(sb, buff + size);
        break;
      }
      case Kchar: {  /* fixed-size string */
        size_t len;
        GCstr *s = lj_lib_checkstr(L, arg);
        luaL_argcheck(L, s->len <= (size_t)size, arg,
                         "string longer than given size");
        lj_buf_putstr(sb, s);  /* add string */
        len = s->len;
        while (len++ < (size_t)size)  /* pad extra space */
          lj_buf_putchar(sb, LUAL_PACKPADBYTE);
        break;
      }
      case Kstring: {  /* strings with length count */
        GCstr *s = lj_lib_checkstr(L, arg);
        luaL_argcheck(L, size >= (int)sizeof(size_t) ||
                         s->len < ((size_t)1 << (size * NB)),
                         arg, "string length does not fit in given size");
        packint(sb, (lua_Unsigned)s->len, h.islittle, size, 0);  /* pack length */
        lj_buf_putstr(sb, s);
        totalsize += s->len;
        break;
      }
      case Kzstr: {  /* zero-terminated string */
        GCstr *s = lj_lib_checkstr(L, arg);
        luaL_argcheck(L, strlen(strdata(s)) == s->len, arg, "string contains zeros");
        lj_buf_putstr(sb, s);
        lj_buf_putchar(sb, '\0');  /* add zero at the end */
        totalsize += s->len + 1;
        break;
      }
      case Kpadding: lj_buf_putchar(sb, LUAL_PACKPADBYTE);  /* FALLTHROUGH */
      case Kpaddalign: case Knop:
        arg--;  /* undo increment */
        break;
    }
  }
  setstrV(L, L->top-1, lj_buf_str(L, sb));
  lj_gc_check(L);
  return 1;
}


LJLIB_CF(string_packsize)
{
  Header h;
  const char *fmt = luaL_checkstring(L, 1);  /* format string */
  size_t totalsize = 0;  /* accumulate total size of result */
  initheader(L, &h);
  while (*fmt != '\0') {
    int size, ntoalign;
    KOption opt = getdetails(&h, totalsize, &fmt, &size, &ntoalign);
    size += ntoalign;  /* total space used by option */
    luaL_argcheck(L, totalsize <= MAXSIZE - size, 1,
                     "format result too large");
    totalsize += size;
    switch (opt) {
      case Kstring:  /* strings with length count */
      case Kzstr:    /* zero-terminated string */
        luaL_argerror(L, 1, "variable-length format");
        /* call never return, but to avoid warnings: *//* FALLTHROUGH */
      default:  break;
    }
  }
  lua_pushinteger(L, (lua_Integer)totalsize);
  return 1;
}


/*
** Unpack an integer with 'size' bytes and 'islittle' endianness.
** If size is smaller than the size of a Lua integer and integer
** is signed, must do sign extension (propagating the sign to the
** higher bits); if size is larger than the size of a Lua integer,
** it must check the unread bytes to see whether they do not cause an
** overflow.
*/
static lua_Integer unpackint (lua_State *L, const char *str,
                              int islittle, int size, int issigned) {
  lua_Unsigned res = 0;
  int i;
  int limit = (size  <= SZINT) ? size : SZINT;
  for (i = limit - 1; i >= 0; i--) {
    res <<= NB;
    res |= (lua_Unsigned)(unsigned char)str[islittle ? i : size - 1 - i];
  }
  if (size < SZINT) {  /* real size smaller than lua_Integer? */
    if (issigned) {  /* needs sign extension? */
      lua_Unsigned mask = (lua_Unsigned)1 << (size*NB - 1);
      res = ((res ^ mask) - mask);  /* do sign extension */
    }
  }
  else if (size > SZINT) {  /* must check unread bytes */
    int mask = (!issigned || (lua_Integer)res >= 0) ? 0 : MC;
    for (i = limit; i < size; i++) {
      if ((unsigned char)str[islittle ? i : size - 1 - i] != mask)
        luaL_error(L, "%d-byte integer does not fit into Lua Integer", size);
    }
  }
  return (lua_Integer)res;
}


LJLIB_CF(string_unpack)
{
  Header h;
  const char *fmt = luaL_checkstring(L, 1);
  size_t ld;
  const char *data = luaL_checklstring(L, 2, &ld);
  int n = 0;  /* number of results */
  size_t pos = lj_lib_optint(L, 3, 1);
  if (pos < 0) pos += ld; else pos--;
  if (pos < 0) pos = 0;
  luaL_argcheck(L, pos <= ld, 3, "initial position out of string");
  initheader(L, &h);
  while (*fmt != '\0') {
    int size, ntoalign;
    KOption opt = getdetails(&h, pos, &fmt, &size, &ntoalign);
    if ((size_t)ntoalign + size > ~pos || pos + ntoalign + size > ld)
      luaL_argerror(L, 2, "data string too short");
    pos += ntoalign;  /* skip alignment */
    /* stack space for item + next position */
    luaL_checkstack(L, 2, "too many results");
    n++;
    switch (opt) {
      case Kint:
      case Kuint: {
        lua_Integer res = unpackint(L, data + pos, h.islittle, size,
                                       (opt == Kint));
        lua_pushinteger(L, res);
        break;
      }
      case Kfloat: {
        volatile Ftypes u;
        lua_Number num;
        copywithendian(u.buff, data + pos, size, h.islittle);
        if (size == sizeof(u.f)) num = (lua_Number)u.f;
        else if (size == sizeof(u.d)) num = (lua_Number)u.d;
        else num = u.n;
        lua_pushnumber(L, num);
        break;
      }
      case Kchar: {
        lua_pushlstring(L, data + pos, size);
        break;
      }
      case Kstring: {
        size_t len = (size_t)unpackint(L, data + pos, h.islittle, size, 0);
        luaL_argcheck(L, pos + len + size <= ld, 2, "data string too short");
        lua_pushlstring(L, data + pos + size, len);
        pos += len;  /* skip string */
        break;
      }
      case Kzstr: {
        size_t len = (int)strlen(data + pos);
        lua_pushlstring(L, data + pos, len);
        pos += len + 1;  /* skip string plus final '\0' */
        break;
      }
      case Kpaddalign: case Kpadding: case Knop:
        n--;  /* undo increment */
        break;
    }
    pos += size;
  }
  lua_pushinteger(L, pos + 1);  /* next position */
  return n + 1;
}

/* ------------------------------------------------------------------------ */

#include "lj_libdef.h"

LUALIB_API int luaopen_string(lua_State *L)
{
  GCtab *mt;
  global_State *g;
  LJ_LIB_REG(L, LUA_STRLIBNAME, string);
  mt = lj_tab_new(L, 0, 1);
  /* NOBARRIER: basemt is a GC root. */
  g = G(L);
  setgcref(basemt_it(g, LJ_TSTR), obj2gco(mt));
  settabV(L, lj_tab_setstr(L, mt, mmname_str(g, MM_index)), tabV(L->top-1));
  mt->nomm = (uint8_t)(~(1u<<MM_index));
  return 1;
}