lj_strscan.c

/*
** String scanning.
** Copyright (C) 2005-2013 Mike Pall. See Copyright Notice in luajit.h
*/

#include <math.h>

#define lj_strscan_c
#define LUA_CORE

#include "lj_obj.h"
#include "lj_char.h"
#include "lj_strscan.h"

/* -- Scanning numbers ---------------------------------------------------- */

/*
** Rationale for the builtin string to number conversion library:
**
** It removes a dependency on libc's strtod(), which is a true portability
** nightmare. Mainly due to the plethora of supported OS and toolchain
** combinations. Sadly, the various implementations
** a) are often buggy, incomplete (no hex floats) and/or imprecise,
** b) sometimes crash or hang on certain inputs,
** c) return non-standard NaNs that need to be filtered out, and
** d) fail if the locale-specific decimal separator is not a dot,
**    which can only be fixed with atrocious workarounds.
**
** Also, most of the strtod() implementations are hopelessly bloated,
** which is not just an I-cache hog, but a problem for static linkage
** on embedded systems, too.
**
** OTOH the builtin conversion function is very compact. Even though it
** does a lot more, like parsing long longs, octal or imaginary numbers
** and returning the result in different formats:
** a) It needs less than 3 KB (!) of machine code (on x64 with -Os),
** b) it doesn't perform any dynamic allocation and,
** c) it needs only around 600 bytes of stack space.
**
** The builtin function is faster than strtod() for typical inputs, e.g.
** "123", "1.5" or "1e6". Arguably, it's slower for very large exponents,
** which are not very common (this could be fixed, if needed).
**
** And most importantly, the builtin function is equally precise on all
** platforms. It correctly converts and rounds any input to a double.
** If this is not the case, please send a bug report -- but PLEASE verify
** that the implementation you're comparing to is not the culprit!
**
** The implementation quickly pre-scans the entire string first and
** handles simple integers on-the-fly. Otherwise, it dispatches to the
** base-specific parser. Hex and octal is straightforward.
**
** Decimal to binary conversion uses a fixed-length circular buffer in
** base 100. Some simple cases are handled directly. For other cases, the
** number in the buffer is up-scaled or down-scaled until the integer part
** is in the proper range. Then the integer part is rounded and converted
** to a double which is finally rescaled to the result. Denormals need
** special treatment to prevent incorrect 'double rounding'.
*/

/* Definitions for circular decimal digit buffer (base 100 = 2 digits/byte). */
#define STRSCAN_DIG	1024
#define STRSCAN_MAXDIG	800		/* 772 + extra are sufficient. */
#define STRSCAN_DDIG	(STRSCAN_DIG/2)
#define STRSCAN_DMASK	(STRSCAN_DDIG-1)

/* Helpers for circular buffer. */
#define DNEXT(a)	(((a)+1) & STRSCAN_DMASK)
#define DPREV(a)	(((a)-1) & STRSCAN_DMASK)
#define DLEN(lo, hi)	((int32_t)(((lo)-(hi)) & STRSCAN_DMASK))

#define casecmp(c, k)	(((c) | 0x20) == k)

/* Final conversion to double. */
static void strscan_double(uint64_t x, TValue *o, int32_t ex2, int32_t neg)
{
  double n;

  /* Avoid double rounding for denormals. */
  if (LJ_UNLIKELY(ex2 <= -1075 && x != 0)) {
    /* NYI: all of this generates way too much code on 32 bit CPUs. */
#if defined(__GNUC__) && LJ_64
    int32_t b = (int32_t)(__builtin_clzll(x)^63);
#else
    int32_t b = (x>>32) ? 32+(int32_t)lj_fls((uint32_t)(x>>32)) :
			  (int32_t)lj_fls((uint32_t)x);
#endif
    if ((int32_t)b + ex2 <= -1023 && (int32_t)b + ex2 >= -1075) {
      uint64_t rb = (uint64_t)1 << (-1075-ex2);
      if ((x & rb) && ((x & (rb+rb+rb-1)))) x += rb+rb;
      x = (x & ~(rb+rb-1));
    }
  }

  /* Convert to double using a signed int64_t conversion, then rescale. */
  lua_assert((int64_t)x >= 0);
  n = (double)(int64_t)x;
  if (neg) n = -n;
  if (ex2) n = ldexp(n, ex2);
  o->n = n;
}

/* Parse hexadecimal number. */
static StrScanFmt strscan_hex(const uint8_t *p, TValue *o,
			      StrScanFmt fmt, uint32_t opt,
			      int32_t ex2, int32_t neg, uint32_t dig)
{
  uint64_t x = 0;
  uint32_t i;

  /* Scan hex digits. */
  for (i = dig > 16 ? 16 : dig ; i; i--, p++) {
    uint32_t d = (*p != '.' ? *p : *++p); if (d > '9') d += 9;
    x = (x << 4) + (d & 15);
  }

  /* Summarize rounding-effect of excess digits. */
  for (i = 16; i < dig; i++, p++)
    x |= ((*p != '.' ? *p : *++p) != '0'), ex2 += 4;

  /* Format-specific handling. */
  switch (fmt) {
  case STRSCAN_INT:
    if (!(opt & STRSCAN_OPT_TONUM) && x < 0x80000000u+neg) {
      o->i = neg ? -(int32_t)x : (int32_t)x;
      return STRSCAN_INT;  /* Fast path for 32 bit integers. */
    }
    if (!(opt & STRSCAN_OPT_C)) { fmt = STRSCAN_NUM; break; }
    /* fallthrough */
  case STRSCAN_U32:
    if (dig > 8) return STRSCAN_ERROR;
    o->i = neg ? -(int32_t)x : (int32_t)x;
    return STRSCAN_U32;
  case STRSCAN_I64:
  case STRSCAN_U64:
    if (dig > 16) return STRSCAN_ERROR;
    o->u64 = neg ? (uint64_t)-(int64_t)x : x;
    return fmt;
  default:
    break;
  }

  /* Reduce range then convert to double. */
  if ((x & U64x(c0000000,0000000))) { x = (x >> 2) | (x & 3); ex2 += 2; }
  strscan_double(x, o, ex2, neg);
  return fmt;
}

/* Parse octal number. */
static StrScanFmt strscan_oct(const uint8_t *p, TValue *o,
			      StrScanFmt fmt, int32_t neg, uint32_t dig)
{
  uint64_t x = 0;

  /* Scan octal digits. */
  if (dig > 22 || (dig == 22 && *p > '1')) return STRSCAN_ERROR;
  while (dig-- > 0) {
    if (!(*p >= '0' && *p <= '7')) return STRSCAN_ERROR;
    x = (x << 3) + (*p++ & 7);
  }

  /* Format-specific handling. */
  switch (fmt) {
  case STRSCAN_INT:
    if (x >= 0x80000000u+neg) fmt = STRSCAN_U32;
    /* fallthrough */
  case STRSCAN_U32:
    if ((x >> 32)) return STRSCAN_ERROR;
    o->i = neg ? -(int32_t)x : (int32_t)x;
    break;
  default:
  case STRSCAN_I64:
  case STRSCAN_U64:
    o->u64 = neg ? (uint64_t)-(int64_t)x : x;
    break;
  }
  return fmt;
}

/* Parse decimal number. */
static StrScanFmt strscan_dec(const uint8_t *p, TValue *o,
			      StrScanFmt fmt, uint32_t opt,
			      int32_t ex10, int32_t neg, uint32_t dig)
{
  uint8_t xi[STRSCAN_DDIG], *xip = xi;

  if (dig) {
    uint32_t i = dig;
    if (i > STRSCAN_MAXDIG) {
      ex10 += (int32_t)(i - STRSCAN_MAXDIG);
      i = STRSCAN_MAXDIG;
    }
    /* Scan unaligned leading digit. */
    if (((ex10^i) & 1))
      *xip++ = ((*p != '.' ? *p : *++p) & 15), i--, p++;
    /* Scan aligned double-digits. */
    for ( ; i > 1; i -= 2) {
      uint32_t d = 10 * ((*p != '.' ? *p : *++p) & 15); p++;
      *xip++ = d + ((*p != '.' ? *p : *++p) & 15); p++;
    }
    /* Scan and realign trailing digit. */
    if (i) *xip++ = 10 * ((*p != '.' ? *p : *++p) & 15), ex10--, p++;

    /* Summarize rounding-effect of excess digits. */
    if (dig > STRSCAN_MAXDIG) {
      do {
	if ((*p != '.' ? *p : *++p) != '0') { xip[-1] |= 1; break; }
	p++;
      } while (--dig > STRSCAN_MAXDIG);
      dig = STRSCAN_MAXDIG;
    } else {  /* Simplify exponent. */
      while (ex10 > 0 && dig <= 18) *xip++ = 0, ex10 -= 2, dig += 2;
    }
  } else {  /* Only got zeros. */
    ex10 = 0;
    xi[0] = 0;
  }

  /* Fast path for numbers in integer format (but handles e.g. 1e6, too). */
  if (dig <= 20 && ex10 == 0) {
    uint8_t *xis;
    uint64_t x = xi[0];
    double n;
    for (xis = xi+1; xis < xip; xis++) x = x * 100 + *xis;
    if (!(dig == 20 && (xi[0] > 18 || (int64_t)x >= 0))) {  /* No overflow? */
      /* Format-specific handling. */
      switch (fmt) {
      case STRSCAN_INT:
	if (!(opt & STRSCAN_OPT_TONUM) && x < 0x80000000u+neg) {
	  o->i = neg ? -(int32_t)x : (int32_t)x;
	  return STRSCAN_INT;  /* Fast path for 32 bit integers. */
	}
	if (!(opt & STRSCAN_OPT_C)) { fmt = STRSCAN_NUM; goto plainnumber; }
	/* fallthrough */
      case STRSCAN_U32:
	if ((x >> 32) != 0) return STRSCAN_ERROR;
	o->i = neg ? -(int32_t)x : (int32_t)x;
	return STRSCAN_U32;
      case STRSCAN_I64:
      case STRSCAN_U64:
	o->u64 = neg ? (uint64_t)-(int64_t)x : x;
	return fmt;
      default:
      plainnumber:  /* Fast path for plain numbers < 2^63. */
	if ((int64_t)x < 0) break;
	n = (double)(int64_t)x;
	if (neg) n = -n;
	o->n = n;
	return fmt;
      }
    }
  }

  /* Slow non-integer path. */
  if (fmt == STRSCAN_INT) {
    if ((opt & STRSCAN_OPT_C)) return STRSCAN_ERROR;
    fmt = STRSCAN_NUM;
  } else if (fmt > STRSCAN_INT) {
    return STRSCAN_ERROR;
  }
  {
    uint32_t hi = 0, lo = (uint32_t)(xip-xi);
    int32_t ex2 = 0, idig = (int32_t)lo + (ex10 >> 1);

    lua_assert(lo > 0 && (ex10 & 1) == 0);

    /* Handle simple overflow/underflow. */
    if (idig > 310/2) { if (neg) setminfV(o); else setpinfV(o); return fmt; }
    else if (idig < -326/2) { o->n = neg ? -0.0 : 0.0; return fmt; }

    /* Scale up until we have at least 17 or 18 integer part digits. */
    while (idig < 9 && idig < DLEN(lo, hi)) {
      uint32_t i, cy = 0;
      ex2 -= 6;
      for (i = DPREV(lo); ; i = DPREV(i)) {
	uint32_t d = (xi[i] << 6) + cy;
	cy = (((d >> 2) * 5243) >> 17); d = d - cy * 100;  /* Div/mod 100. */
	xi[i] = (uint8_t)d;
	if (i == hi) break;
	if (d == 0 && i == DPREV(lo)) lo = i;
      }
      if (cy) {
	hi = DPREV(hi);
	if (xi[DPREV(lo)] == 0) lo = DPREV(lo);
	else if (hi == lo) { lo = DPREV(lo); xi[DPREV(lo)] |= xi[lo]; }
	xi[hi] = (uint8_t)cy; idig++;
      }
    }

    /* Scale down until no more than 17 or 18 integer part digits remain. */
    while (idig > 9) {
      uint32_t i, cy = 0;
      ex2 += 6;
      for (i = hi; i != lo; i = DNEXT(i)) {
	cy += xi[i];
	xi[i] = (cy >> 6);
	cy = 100 * (cy & 0x3f);
	if (xi[i] == 0 && i == hi) hi = DNEXT(hi), idig--;
      }
      while (cy) {
	if (hi == lo) { xi[DPREV(lo)] |= 1; break; }
	xi[lo] = (cy >> 6); lo = DNEXT(lo);
	cy = 100 * (cy & 0x3f);
      }
    }

    /* Collect integer part digits and convert to rescaled double. */
    {
      uint64_t x = xi[hi];
      uint32_t i;
      for (i = DNEXT(hi); --idig > 0 && i != lo; i = DNEXT(i))
	x = x * 100 + xi[i];
      if (i == lo) {
	while (--idig >= 0) x = x * 100;
      } else {  /* Gather round bit from remaining digits. */
	x <<= 1; ex2--;
	do {
	  if (xi[i]) { x |= 1; break; }
	  i = DNEXT(i);
	} while (i != lo);
      }
      strscan_double(x, o, ex2, neg);
    }
  }
  return fmt;
}

/* Scan string containing a number. Returns format. Returns value in o. */
StrScanFmt lj_strscan_scan(const uint8_t *p, TValue *o, uint32_t opt)
{
  int32_t neg = 0;

  /* Remove leading space, parse sign and non-numbers. */
  if (LJ_UNLIKELY(!lj_char_isdigit(*p))) {
    while (lj_char_isspace(*p)) p++;
    if (*p == '+' || *p == '-') neg = (*p++ == '-');
    if (LJ_UNLIKELY(*p >= 'A')) {  /* Parse "inf", "infinity" or "nan". */
      TValue tmp;
      setnanV(&tmp);
      if (casecmp(p[0],'i') && casecmp(p[1],'n') && casecmp(p[2],'f')) {
	if (neg) setminfV(&tmp); else setpinfV(&tmp);
	p += 3;
	if (casecmp(p[0],'i') && casecmp(p[1],'n') && casecmp(p[2],'i') &&
	    casecmp(p[3],'t') && casecmp(p[4],'y')) p += 5;
      } else if (casecmp(p[0],'n') && casecmp(p[1],'a') && casecmp(p[2],'n')) {
	p += 3;
      }
      while (lj_char_isspace(*p)) p++;
      if (*p) return STRSCAN_ERROR;
      o->u64 = tmp.u64;
      return STRSCAN_NUM;
    }
  }

  /* Parse regular number. */
  {
    StrScanFmt fmt = STRSCAN_INT;
    int cmask = LJ_CHAR_DIGIT;
    int base = (opt & STRSCAN_OPT_C) && *p == '0' ? 0 : 10;
    const uint8_t *sp, *dp = NULL;
    uint32_t dig = 0, hasdig = 0, x = 0;
    int32_t ex = 0;

    /* Determine base and skip leading zeros. */
    if (LJ_UNLIKELY(*p <= '0')) {
      if (*p == '0' && casecmp(p[1], 'x'))
	base = 16, cmask = LJ_CHAR_XDIGIT, p += 2;
      for ( ; ; p++) {
	if (*p == '0') {
	  hasdig = 1;
	} else if (*p == '.') {
	  if (dp) return STRSCAN_ERROR;
	  dp = p;
	} else {
	  break;
	}
      }
    }

    /* Preliminary digit and decimal point scan. */
    for (sp = p; ; p++) {
      if (LJ_LIKELY(lj_char_isa(*p, cmask))) {
	x = x * 10 + (*p & 15);  /* For fast path below. */
	dig++;
      } else if (*p == '.') {
	if (dp) return STRSCAN_ERROR;
	dp = p;
      } else {
	break;
      }
    }
    if (!(hasdig | dig)) return STRSCAN_ERROR;

    /* Handle decimal point. */
    if (dp) {
      fmt = STRSCAN_NUM;
      if (dig) {
	ex = (int32_t)(dp-(p-1)); dp = p-1;
	while (ex < 0 && *dp-- == '0') ex++, dig--;  /* Skip trailing zeros. */
	if (base == 16) ex *= 4;
      }
    }

    /* Parse exponent. */
    if (casecmp(*p, (uint32_t)(base == 16 ? 'p' : 'e'))) {
      uint32_t xx;
      int negx = 0;
      fmt = STRSCAN_NUM; p++;
      if (*p == '+' || *p == '-') negx = (*p++ == '-');
      if (!lj_char_isdigit(*p)) return STRSCAN_ERROR;
      xx = (*p++ & 15);
      while (lj_char_isdigit(*p)) {
	if (xx < 65536) xx = xx * 10 + (*p & 15);
	p++;
      }
      ex += negx ? -(int32_t)xx : (int32_t)xx;
    }

    /* Parse suffix. */
    if (*p) {
      /* I (IMAG), U (U32), LL (I64), ULL/LLU (U64), L (long), UL/LU (ulong). */
      /* NYI: f (float). Not needed until cp_number() handles non-integers. */
      if (casecmp(*p, 'i')) {
	if (!(opt & STRSCAN_OPT_IMAG)) return STRSCAN_ERROR;
	p++; fmt = STRSCAN_IMAG;
      } else if (fmt == STRSCAN_INT) {
	if (casecmp(*p, 'u')) p++, fmt = STRSCAN_U32;
	if (casecmp(*p, 'l')) {
	  p++;
	  if (casecmp(*p, 'l')) p++, fmt += STRSCAN_I64 - STRSCAN_INT;
	  else if (!(opt & STRSCAN_OPT_C)) return STRSCAN_ERROR;
	  else if (sizeof(long) == 8) fmt += STRSCAN_I64 - STRSCAN_INT;
	}
	if (casecmp(*p, 'u') && (fmt == STRSCAN_INT || fmt == STRSCAN_I64))
	  p++, fmt += STRSCAN_U32 - STRSCAN_INT;
	if ((fmt == STRSCAN_U32 && !(opt & STRSCAN_OPT_C)) ||
	    (fmt >= STRSCAN_I64 && !(opt & STRSCAN_OPT_LL)))
	  return STRSCAN_ERROR;
      }
      while (lj_char_isspace(*p)) p++;
      if (*p) return STRSCAN_ERROR;
    }

    /* Fast path for decimal 32 bit integers. */
    if (fmt == STRSCAN_INT && base == 10 &&
	(dig < 10 || (dig == 10 && *sp <= '2' && x < 0x80000000u+neg))) {
      int32_t y = neg ? -(int32_t)x : (int32_t)x;
      if ((opt & STRSCAN_OPT_TONUM)) {
	o->n = (double)y;
	return STRSCAN_NUM;
      } else {
	o->i = y;
	return STRSCAN_INT;
      }
    }

    /* Dispatch to base-specific parser. */
    if (base == 0 && !(fmt == STRSCAN_NUM || fmt == STRSCAN_IMAG))
      return strscan_oct(sp, o, fmt, neg, dig);
    if (base == 16)
      fmt = strscan_hex(sp, o, fmt, opt, ex, neg, dig);
    else
      fmt = strscan_dec(sp, o, fmt, opt, ex, neg, dig);

    /* Try to convert number to integer, if requested. */
    if (fmt == STRSCAN_NUM && (opt & STRSCAN_OPT_TOINT)) {
      double n = o->n;
      int32_t i = lj_num2int(n);
      if (n == (lua_Number)i) { o->i = i; return STRSCAN_INT; }
    }
    return fmt;
  }
}

int LJ_FASTCALL lj_strscan_num(GCstr *str, TValue *o)
{
  StrScanFmt fmt = lj_strscan_scan((const uint8_t *)strdata(str), o,
				   STRSCAN_OPT_TONUM);
  lua_assert(fmt == STRSCAN_ERROR || fmt == STRSCAN_NUM);
  return (fmt != STRSCAN_ERROR);
}

#if LJ_DUALNUM
int LJ_FASTCALL lj_strscan_number(GCstr *str, TValue *o)
{
  StrScanFmt fmt = lj_strscan_scan((const uint8_t *)strdata(str), o,
				   STRSCAN_OPT_TOINT);
  lua_assert(fmt == STRSCAN_ERROR || fmt == STRSCAN_NUM || fmt == STRSCAN_INT);
  if (fmt == STRSCAN_INT) setitype(o, LJ_TISNUM);
  return (fmt != STRSCAN_ERROR);
}
#endif

#undef DNEXT
#undef DPREV
#undef DLEN