view src/number.c @ 4885:6772ce4d982b

Fix hash tables, #'member*, #'assoc*, #'eql compiler macros if bignums lisp/ChangeLog addition: 2010-01-24 Aidan Kehoe <kehoea@parhasard.net> Correct the semantics of #'member*, #'eql, #'assoc* in the presence of bignums; change the integerp byte code to fixnump semantics. * bytecomp.el (fixnump, integerp, byte-compile-integerp): Change the integerp byte code to fixnump; add a byte-compile method to integerp using fixnump and numberp and avoiding a funcall most of the time, since in the non-core contexts where integerp is used, it's mostly distinguishing between fixnums and things that are not numbers at all. * byte-optimize.el (side-effect-free-fns, byte-after-unbind-ops) (byte-compile-side-effect-and-error-free-ops): Replace the integerp bytecode with fixnump; add fixnump to the side-effect-free-fns. Add the other extended number type predicates to the list in passing. * obsolete.el (floatp-safe): Mark this as obsolete. * cl.el (eql): Go into more detail in the docstring here. Don't bother checking whether both arguments are numbers; one is enough, #'equal will fail correctly if they have distinct types. (subst): Replace a call to #'integerp (deciding whether to use #'memq or not) with one to #'fixnump. Delete most-positive-fixnum, most-negative-fixnum from this file; they're now always in C, so they can't be modified from Lisp. * cl-seq.el (member*, assoc*, rassoc*): Correct these functions in the presence of bignums. * cl-macs.el (cl-make-type-test): The type test for a fixnum is now fixnump. Ditch floatp-safe, use floatp instead. (eql): Correct this compiler macro in the presence of bignums. (assoc*): Correct this compiler macro in the presence of bignums. * simple.el (undo): Change #'integerp to #'fixnump here, since we use #'delq with the same value as ELT a few lines down. src/ChangeLog addition: 2010-01-24 Aidan Kehoe <kehoea@parhasard.net> Fix problems with #'eql, extended number types, and the hash table implementation; change the Bintegerp bytecode to fixnump semantics even on bignum builds, since #'integerp can have a fast implementation in terms of #'fixnump for most of its extant uses, but not vice-versa. * lisp.h: Always #include number.h; we want the macros provided in it, even if the various number types are not available. * number.h (NON_FIXNUM_NUMBER_P): New macro, giving 1 when its argument is of non-immediate number type. Equivalent to FLOATP if WITH_NUMBER_TYPES is not defined. * elhash.c (lisp_object_eql_equal, lisp_object_eql_hash): Use NON_FIXNUM_NUMBER_P in these functions, instead of FLOATP, giving more correct behaviour in the presence of the extended number types. * bytecode.c (Bfixnump, execute_optimized_program): Rename Bintegerp to Bfixnump; change its semantics to reflect the new name on builds with bignum support. * data.c (Ffixnump, Fintegerp, syms_of_data, vars_of_data): Always make #'fixnump available, even on non-BIGNUM builds; always implement #'integerp in this file, even on BIGNUM builds. Move most-positive-fixnum, most-negative-fixnum here from number.c, so they are Lisp constants even on builds without number types, and attempts to change or bind them error. Use the NUMBERP and INTEGERP macros even on builds without extended number types. * data.c (fixnum_char_or_marker_to_int): Rename this function from integer_char_or_marker_to_int, to better reflect the arguments it accepts. * number.c (Fevenp, Foddp, syms_of_number): Never provide #'integerp in this file. Remove #'oddp, #'evenp; their implementations are overridden by those in cl.el. * number.c (vars_of_number): most-positive-fixnum, most-negative-fixnum are no longer here. man/ChangeLog addition: 2010-01-23 Aidan Kehoe <kehoea@parhasard.net> Generally: be careful to say fixnum, not integer, when talking about fixed-precision integral types. I'm sure I've missed instances, both here and in the docstrings, but this is a decent start. * lispref/text.texi (Columns): Document where only fixnums, not integers generally, are accepted. (Registers): Remove some ancient char-int confoundance here. * lispref/strings.texi (Creating Strings, Creating Strings): Be more exact in describing where fixnums but not integers in general are accepted. (Creating Strings): Use a more contemporary example to illustrate how concat deals with lists including integers about #xFF. Delete some obsolete documentation on same. (Char Table Types): Document that only fixnums are accepted as values in syntax tables. * lispref/searching.texi (String Search, Search and Replace): Be exact in describing where fixnums but not integers in general are accepted. * lispref/range-tables.texi (Range Tables): Be exact in describing them; only fixnums are accepted to describe ranges. * lispref/os.texi (Killing XEmacs, User Identification) (Time of Day, Time Conversion): Be more exact about using fixnum where only fixed-precision integers are accepted. * lispref/objects.texi (Integer Type): Be more exact (and up-to-date) about the possible values for integers. Cross-reference to documentation of the bignum extension. (Equality Predicates): (Range Table Type): (Array Type): Use fixnum, not integer, to describe a fixed-precision integer. (Syntax Table Type): Correct some English syntax here. * lispref/numbers.texi (Numbers): Change the phrasing here to use fixnum to mean the fixed-precision integers normal in emacs. Document that our terminology deviates from that of Common Lisp, and that we're working on it. (Compatibility Issues): Reiterate the Common Lisp versus Emacs Lisp compatibility issues. (Comparison of Numbers, Arithmetic Operations): * lispref/commands.texi (Command Loop Info, Working With Events): * lispref/buffers.texi (Modification Time): Be more exact in describing where fixnums but not integers in general are accepted.
author Aidan Kehoe <kehoea@parhasard.net>
date Sun, 24 Jan 2010 15:21:27 +0000
parents 2fc0e2f18322
children 1e9078742fa7
line wrap: on
line source

/* Numeric types for XEmacs.
   Copyright (C) 2004 Jerry James.

This file is part of XEmacs.

XEmacs is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2, or (at your option) any
later version.

XEmacs is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received a copy of the GNU General Public License
along with XEmacs; see the file COPYING.  If not, write to
the Free Software Foundation, Inc., 51 Franklin St - Fifth Floor,
Boston, MA 02111-1301, USA.  */

/* Synched up with: Not in FSF. */

#include <config.h>
#include <limits.h>
#include "lisp.h"

#ifdef HAVE_BIGFLOAT
#define USED_IF_BIGFLOAT(decl) decl
#else
#define USED_IF_BIGFLOAT(decl) UNUSED (decl)
#endif

Lisp_Object Qrationalp, Qfloatingp, Qrealp;
Lisp_Object Vdefault_float_precision;

static Lisp_Object Qunsupported_type;
static Lisp_Object Vbigfloat_max_prec;
static int number_initialized;

#ifdef HAVE_BIGNUM
bignum scratch_bignum, scratch_bignum2;
#endif
#ifdef HAVE_RATIO
ratio scratch_ratio, scratch_ratio2;
#endif
#ifdef HAVE_BIGFLOAT
bigfloat scratch_bigfloat, scratch_bigfloat2;
#endif

/********************************* Bignums **********************************/
#ifdef HAVE_BIGNUM
static void
bignum_print (Lisp_Object obj, Lisp_Object printcharfun,
	      int UNUSED (escapeflag))
{
  CIbyte *bstr = bignum_to_string (XBIGNUM_DATA (obj), 10);
  write_c_string (printcharfun, bstr);
  xfree (bstr, CIbyte *);
}

#ifdef NEW_GC
static void
bignum_finalize (void *header, int for_disksave)
{
  if (!for_disksave)
    {
      struct Lisp_Bignum *num = (struct Lisp_Bignum *) header;
      bignum_fini (num->data);
    }
}
#define BIGNUM_FINALIZE bignum_finalize
#else
#define BIGNUM_FINALIZE 0
#endif

static int
bignum_equal (Lisp_Object obj1, Lisp_Object obj2, int UNUSED (depth))
{
  return bignum_eql (XBIGNUM_DATA (obj1), XBIGNUM_DATA (obj2));
}

static Hashcode
bignum_hash (Lisp_Object obj, int UNUSED (depth))
{
  return bignum_hashcode (XBIGNUM_DATA (obj));
}

static void
bignum_convert (const void *object, void **data, Bytecount *size)
{
  CIbyte *bstr = bignum_to_string (*(bignum *)object, 10);
  *data = bstr;
  *size = strlen(bstr)+1;
}

static void
bignum_convfree (const void * UNUSED (object), void *data,
		 Bytecount UNUSED (size))
{
  xfree (data, void *);
}

static void *
bignum_deconvert (void *object, void *data, Bytecount UNUSED (size))
{
  bignum *b = (bignum *) object;
  bignum_init(*b);
  bignum_set_string(*b, (const char *) data, 10);
  return object;
}

static const struct opaque_convert_functions bignum_opc = {
  bignum_convert,
  bignum_convfree,
  bignum_deconvert
};

static const struct memory_description bignum_description[] = {
  { XD_OPAQUE_DATA_CONVERTIBLE, offsetof (Lisp_Bignum, data),
    0, { &bignum_opc }, XD_FLAG_NO_KKCC },
  { XD_END }
};

DEFINE_BASIC_LRECORD_IMPLEMENTATION ("bignum", bignum, 1, 0, bignum_print,
				     BIGNUM_FINALIZE, bignum_equal,
				     bignum_hash, bignum_description,
				     Lisp_Bignum);

#endif /* HAVE_BIGNUM */

Lisp_Object Qbignump;

DEFUN ("bignump", Fbignump, 1, 1, 0, /*
Return t if OBJECT is a bignum, nil otherwise.
*/
       (object))
{
  return BIGNUMP (object) ? Qt : Qnil;
}


/********************************** Ratios **********************************/
#ifdef HAVE_RATIO
static void
ratio_print (Lisp_Object obj, Lisp_Object printcharfun,
	     int UNUSED (escapeflag))
{
  CIbyte *rstr = ratio_to_string (XRATIO_DATA (obj), 10);
  write_c_string (printcharfun, rstr);
  xfree (rstr, CIbyte *);
}

#ifdef NEW_GC
static void
ratio_finalize (void *header, int for_disksave)
{
  if (!for_disksave)
    {
      struct Lisp_Ratio *num = (struct Lisp_Ratio *) header;
      ratio_fini (num->data);
    }
}
#define RATIO_FINALIZE ratio_finalize
#else
#define RATIO_FINALIZE 0
#endif

static int
ratio_equal (Lisp_Object obj1, Lisp_Object obj2, int UNUSED (depth))
{
  return ratio_eql (XRATIO_DATA (obj1), XRATIO_DATA (obj2));
}

static Hashcode
ratio_hash (Lisp_Object obj, int UNUSED (depth))
{
  return ratio_hashcode (XRATIO_DATA (obj));
}

static const struct memory_description ratio_description[] = {
  { XD_OPAQUE_PTR, offsetof (Lisp_Ratio, data) },
  { XD_END }
};

DEFINE_BASIC_LRECORD_IMPLEMENTATION ("ratio", ratio, 0, 0, ratio_print,
				     RATIO_FINALIZE, ratio_equal, ratio_hash,
				     ratio_description, Lisp_Ratio);

#endif /* HAVE_RATIO */

Lisp_Object Qratiop;

DEFUN ("ratiop", Fratiop, 1, 1, 0, /*
Return t if OBJECT is a ratio, nil otherwise.
*/
       (object))
{
  return RATIOP (object) ? Qt : Qnil;
}


/******************************** Rationals *********************************/
DEFUN ("rationalp", Frationalp, 1, 1, 0, /*
Return t if OBJECT is a rational, nil otherwise.
*/
       (object))
{
  return RATIONALP (object) ? Qt : Qnil;
}

DEFUN ("numerator", Fnumerator, 1, 1, 0, /*
Return the numerator of the canonical form of RATIONAL.
If RATIONAL is an integer, RATIONAL is returned.
*/
       (rational))
{
  CONCHECK_RATIONAL (rational);
#ifdef HAVE_RATIO
  return RATIOP (rational)
    ? make_bignum_bg (XRATIO_NUMERATOR (rational))
    : rational;
#else
  return rational;
#endif
}

DEFUN ("denominator", Fdenominator, 1, 1, 0, /*
Return the denominator of the canonical form of RATIONAL.
If RATIONAL is an integer, 1 is returned.
*/
       (rational))
{
  CONCHECK_RATIONAL (rational);
#ifdef HAVE_RATIO
  return RATIOP (rational)
    ? make_bignum_bg (XRATIO_DENOMINATOR (rational))
    : make_int (1);
#else
  return make_int (1);
#endif
}


/******************************** Bigfloats *********************************/
#ifdef HAVE_BIGFLOAT
static void
bigfloat_print (Lisp_Object obj, Lisp_Object printcharfun,
		int UNUSED (escapeflag))
{
  CIbyte *fstr = bigfloat_to_string (XBIGFLOAT_DATA (obj), 10);
  write_c_string (printcharfun, fstr);
  xfree (fstr, CIbyte *);
}

#ifdef NEW_GC
static void
bigfloat_finalize (void *header, int for_disksave)
{
  if (!for_disksave)
    {
      struct Lisp_Bigfloat *num = (struct Lisp_Bigfloat *) header;
      bigfloat_fini (num->bf);
    }
}
#define BIGFLOAT_FINALIZE bigfloat_finalize
#else
#define BIGFLOAT_FINALIZE 0
#endif

static int
bigfloat_equal (Lisp_Object obj1, Lisp_Object obj2, int UNUSED (depth))
{
  return bigfloat_eql (XBIGFLOAT_DATA (obj1), XBIGFLOAT_DATA (obj2));
}

static Hashcode
bigfloat_hash (Lisp_Object obj, int UNUSED (depth))
{
  return bigfloat_hashcode (XBIGFLOAT_DATA (obj));
}

static const struct memory_description bigfloat_description[] = {
  { XD_OPAQUE_PTR, offsetof (Lisp_Bigfloat, bf) },
  { XD_END }
};

DEFINE_BASIC_LRECORD_IMPLEMENTATION ("bigfloat", bigfloat, 1, 0,
				     bigfloat_print, BIGFLOAT_FINALIZE,
				     bigfloat_equal, bigfloat_hash,
				     bigfloat_description, Lisp_Bigfloat);

#endif /* HAVE_BIGFLOAT */

Lisp_Object Qbigfloatp;

DEFUN ("bigfloatp", Fbigfloatp, 1, 1, 0, /*
Return t if OBJECT is a bigfloat, nil otherwise.
*/
       (object))
{
  return BIGFLOATP (object) ? Qt : Qnil;
}

DEFUN ("bigfloat-get-precision", Fbigfloat_get_precision, 1, 1, 0, /*
Return the precision of bigfloat F as an integer.
*/
       (f))
{
  CHECK_BIGFLOAT (f);
#ifdef HAVE_BIGNUM
  bignum_set_ulong (scratch_bignum, XBIGFLOAT_GET_PREC (f));
  return Fcanonicalize_number (make_bignum_bg (scratch_bignum));
#else
  return make_int ((int) XBIGFLOAT_GET_PREC (f));
#endif
}

DEFUN ("bigfloat-set-precision", Fbigfloat_set_precision, 2, 2, 0, /*
Set the precision of F, a bigfloat, to PRECISION, a nonnegative integer.
The new precision of F is returned.  Note that the return value may differ
from PRECISION if the underlying library is unable to support exactly
PRECISION bits of precision.
*/
       (f, precision))
{
  unsigned long prec;

  CHECK_BIGFLOAT (f);
  if (INTP (precision))
    {
      prec = (XINT (precision) <= 0) ? 1UL : (unsigned long) XINT (precision);
    }
#ifdef HAVE_BIGNUM
  else if (BIGNUMP (precision))
    {
      prec = bignum_fits_ulong_p (XBIGNUM_DATA (precision))
	? bignum_to_ulong (XBIGNUM_DATA (precision))
	: UINT_MAX;
    }
#endif
  else
    {
      dead_wrong_type_argument (Qintegerp, f);
      return Qnil;
    }

  XBIGFLOAT_SET_PREC (f, prec);
  return Fbigfloat_get_precision (f);
}

static int
default_float_precision_changed (Lisp_Object UNUSED (sym), Lisp_Object *val,
				 Lisp_Object UNUSED (in_object),
				 int UNUSED (flags))
{
  unsigned long prec;

  CONCHECK_INTEGER (*val);
#ifdef HAVE_BIGFLOAT
  if (INTP (*val))
    prec = XINT (*val);
  else
    {
      if (!bignum_fits_ulong_p (XBIGNUM_DATA (*val)))
	args_out_of_range_3 (*val, Qzero, Vbigfloat_max_prec);
      prec = bignum_to_ulong (XBIGNUM_DATA (*val));
    }
  if (prec != 0UL)
    bigfloat_set_default_prec (prec);
#endif
  return 0;
}


/********************************* Floating *********************************/
Lisp_Object
make_floating (double d)
{
#ifdef HAVE_BIGFLOAT
  if (ZEROP (Vdefault_float_precision))
#endif
    return make_float (d);
#ifdef HAVE_BIGFLOAT
  else
    return make_bigfloat (d, 0UL);
#endif
}

DEFUN ("floatingp", Ffloatingp, 1, 1, 0, /*
Return t if OBJECT is a floating point number of any kind, nil otherwise.
*/
       (object))
{
  return FLOATINGP (object) ? Qt : Qnil;
}


/********************************** Reals ***********************************/
DEFUN ("realp", Frealp, 1, 1, 0, /*
Return t if OBJECT is a real, nil otherwise.
*/
       (object))
{
  return REALP (object) ? Qt : Qnil;
}


/********************************* Numbers **********************************/
DEFUN ("canonicalize-number", Fcanonicalize_number, 1, 1, 0, /*
Return the canonical form of NUMBER.
*/
       (number))
{
  /* The tests should go in order from larger, more expressive, or more
     complex types to smaller, less expressive, or simpler types so that a
     number can cascade all the way down to the simplest type if
     appropriate. */
#ifdef HAVE_RATIO
  if (RATIOP (number) &&
      bignum_fits_long_p (XRATIO_DENOMINATOR (number)) &&
      bignum_to_long (XRATIO_DENOMINATOR (number)) == 1L)
    number = make_bignum_bg (XRATIO_NUMERATOR (number));
#endif
#ifdef HAVE_BIGNUM
  if (BIGNUMP (number) && bignum_fits_emacs_int_p (XBIGNUM_DATA (number)))
    {
      EMACS_INT n = bignum_to_emacs_int (XBIGNUM_DATA (number));
      if (NUMBER_FITS_IN_AN_EMACS_INT (n))
	number = make_int (n);
    }
#endif
  return number;
}

enum number_type
get_number_type (Lisp_Object arg)
{
  if (INTP (arg))
    return FIXNUM_T;
#ifdef HAVE_BIGNUM
  if (BIGNUMP (arg))
    return BIGNUM_T;
#endif
#ifdef HAVE_RATIO
  if (RATIOP (arg))
    return RATIO_T;
#endif
  if (FLOATP (arg))
    return FLOAT_T;
#ifdef HAVE_BIGFLOAT
  if (BIGFLOATP (arg))
    return BIGFLOAT_T;
#endif
  /* Catch unintentional bad uses of this function */
  ABORT ();
  /* NOTREACHED */
  return FIXNUM_T;
}

/* Convert NUMBER to type TYPE.  If TYPE is BIGFLOAT_T then use the indicated
   PRECISION; otherwise, PRECISION is ignored. */
static Lisp_Object
internal_coerce_number (Lisp_Object number, enum number_type type,
#ifdef HAVE_BIGFLOAT
			unsigned long precision
#else
			unsigned long UNUSED (precision)
#endif
			)
{
  enum number_type current_type;

  if (CHARP (number))
    number = make_int (XCHAR (number));
  else if (MARKERP (number))
    number = make_int (marker_position (number));

  /* Note that CHECK_NUMBER ensures that NUMBER is a supported type.  Hence,
     we ABORT() in the #else sections below, because it shouldn't be possible
     to arrive there. */
  CHECK_NUMBER (number);
  current_type = get_number_type (number);
  switch (current_type)
    {
    case FIXNUM_T:
      switch (type)
	{
	case FIXNUM_T:
	  return number;
	case BIGNUM_T:
#ifdef HAVE_BIGNUM
	  return make_bignum (XREALINT (number));
#else
	  ABORT ();
#endif /* HAVE_BIGNUM */
	case RATIO_T:
#ifdef HAVE_RATIO
	  return make_ratio (XREALINT (number), 1UL);
#else
	  ABORT ();
#endif /* HAVE_RATIO */
	case FLOAT_T:
	  return make_float (XREALINT (number));
	case BIGFLOAT_T:
#ifdef HAVE_BIGFLOAT
	  return make_bigfloat (XREALINT (number), precision);
#else
	  ABORT ();
#endif /* HAVE_BIGFLOAT */
	}
    case BIGNUM_T:
#ifdef HAVE_BIGNUM
      switch (type)
	{
	case FIXNUM_T:
	  return make_int (bignum_to_long (XBIGNUM_DATA (number)));
	case BIGNUM_T:
	  return number;
	case RATIO_T:
#ifdef HAVE_RATIO
	  bignum_set_long (scratch_bignum, 1L);
	  return make_ratio_bg (XBIGNUM_DATA (number), scratch_bignum);
#else
	  ABORT ();
#endif /* HAVE_RATIO */
	case FLOAT_T:
	  return make_float (bignum_to_double (XBIGNUM_DATA (number)));
	case BIGFLOAT_T:
#ifdef HAVE_BIGFLOAT
	  {
	    Lisp_Object temp;
	    temp = make_bigfloat (0.0, precision);
	    bigfloat_set_bignum (XBIGFLOAT_DATA (temp), XBIGNUM_DATA (number));
	    return temp;
	  }
#else
	  ABORT ();
#endif /* HAVE_BIGFLOAT */
	}
#else
      ABORT ();
#endif /* HAVE_BIGNUM */
    case RATIO_T:
#ifdef HAVE_RATIO
      switch (type)
	{
	case FIXNUM_T:
	  bignum_div (scratch_bignum, XRATIO_NUMERATOR (number),
		      XRATIO_DENOMINATOR (number));
	  return make_int (bignum_to_long (scratch_bignum));
	case BIGNUM_T:
	  bignum_div (scratch_bignum, XRATIO_NUMERATOR (number),
		      XRATIO_DENOMINATOR (number));
	  return make_bignum_bg (scratch_bignum);
	case RATIO_T:
	  return number;
	case FLOAT_T:
	  return make_float (ratio_to_double (XRATIO_DATA (number)));
	case BIGFLOAT_T:
#ifdef HAVE_BIGFLOAT
	  {
	    Lisp_Object temp;
	    temp = make_bigfloat (0.0, precision);
	    bigfloat_set_ratio (XBIGFLOAT_DATA (temp), XRATIO_DATA (number));
	    return temp;
	  }
#else
	  ABORT ();
#endif /* HAVE_BIGFLOAT */
	}
#else
      ABORT ();
#endif /* HAVE_RATIO */
    case FLOAT_T:
      switch (type)
	{
	case FIXNUM_T:
	  return Ftruncate (number, Qnil);
	case BIGNUM_T:
#ifdef HAVE_BIGNUM
	  bignum_set_double (scratch_bignum, XFLOAT_DATA (number));
	  return make_bignum_bg (scratch_bignum);
#else
	  ABORT ();
#endif /* HAVE_BIGNUM */
	case RATIO_T:
#ifdef HAVE_RATIO
	  ratio_set_double (scratch_ratio, XFLOAT_DATA (number));
	  return make_ratio_rt (scratch_ratio);
#else
	  ABORT ();
#endif /* HAVE_RATIO */
	case FLOAT_T:
	  return number;
	case BIGFLOAT_T:
#ifdef HAVE_BIGFLOAT
	  bigfloat_set_prec (scratch_bigfloat, precision);
	  bigfloat_set_double (scratch_bigfloat, XFLOAT_DATA (number));
	  return make_bigfloat_bf (scratch_bigfloat);
#else
	  ABORT ();
#endif /* HAVE_BIGFLOAT */
	}
    case BIGFLOAT_T:
#ifdef HAVE_BIGFLOAT
      switch (type)
	{
	case FIXNUM_T:
	  return make_int (bigfloat_to_long (XBIGFLOAT_DATA (number)));
	case BIGNUM_T:
#ifdef HAVE_BIGNUM
	  bignum_set_bigfloat (scratch_bignum, XBIGFLOAT_DATA (number));
	  return make_bignum_bg (scratch_bignum);
#else
	  ABORT ();
#endif /* HAVE_BIGNUM */
	case RATIO_T:
#ifdef HAVE_RATIO
	  ratio_set_bigfloat (scratch_ratio, XBIGFLOAT_DATA (number));
	  return make_ratio_rt (scratch_ratio);
#else
	  ABORT ();
#endif
	case FLOAT_T:
	  return make_float (bigfloat_to_double (XBIGFLOAT_DATA (number)));
	case BIGFLOAT_T:
	  /* FIXME: Do we need to change the precision? */
	  return number;
	}
#else
      ABORT ();
#endif /* HAVE_BIGFLOAT */
    }
  ABORT ();
  /* NOTREACHED */
  return Qzero;
}

/* This function promotes its arguments as necessary to make them both the
   same type.  It destructively modifies its arguments to do so.  Characters
   and markers are ALWAYS converted to integers. */
enum number_type
promote_args (Lisp_Object *arg1, Lisp_Object *arg2)
{
  enum number_type type1, type2;

  if (CHARP (*arg1))
    *arg1 = make_int (XCHAR (*arg1));
  else if (MARKERP (*arg1))
    *arg1 = make_int (marker_position (*arg1));
  if (CHARP (*arg2))
    *arg2 = make_int (XCHAR (*arg2));
  else if (MARKERP (*arg2))
    *arg2 = make_int (marker_position (*arg2));

  CHECK_NUMBER (*arg1);
  CHECK_NUMBER (*arg2);

  type1 = get_number_type (*arg1);
  type2 = get_number_type (*arg2);

  if (type1 < type2)
    {
      *arg1 = internal_coerce_number (*arg1, type2,
#ifdef HAVE_BIGFLOAT
				      type2 == BIGFLOAT_T
				      ? XBIGFLOAT_GET_PREC (*arg2) :
#endif
				      0UL);
      return type2;
    }

  if (type2 < type1)
    {
      *arg2 = internal_coerce_number (*arg2, type1,
#ifdef HAVE_BIGFLOAT
				      type1 == BIGFLOAT_T
				      ? XBIGFLOAT_GET_PREC (*arg1) :
#endif
				      0UL);
      return type1;
    }

  /* No conversion necessary */
  return type1;
}

DEFUN ("coerce-number", Fcoerce_number, 2, 3, 0, /*
Convert NUMBER to the indicated type, possibly losing information.
Do not call this function.  Use `coerce' instead.

TYPE is one of the symbols `fixnum', `integer', `ratio', `float', or
`bigfloat'.  Not all of these types may be supported.

PRECISION is the number of bits of precision to use when converting to
bigfloat; it is ignored otherwise.  If nil, the default precision is used.

Note that some conversions lose information.  No error is signaled in such
cases; the information is silently lost.
*/
       (number, type, USED_IF_BIGFLOAT (precision)))
{
  CHECK_SYMBOL (type);
  if (EQ (type, Qfixnum))
    return internal_coerce_number (number, FIXNUM_T, 0UL);
  else if (EQ (type, Qinteger))
    {
      /* If bignums are available, we always convert to one first, then
	 downgrade to a fixnum if possible. */
#ifdef HAVE_BIGNUM
      return Fcanonicalize_number
	(internal_coerce_number (number, BIGNUM_T, 0UL));
#else
      return internal_coerce_number (number, FIXNUM_T, 0UL);
#endif
    }
#ifdef HAVE_RATIO
  else if (EQ (type, Qratio))
    return internal_coerce_number (number, RATIO_T, 0UL);
#endif
  else if (EQ (type, Qfloat))
    return internal_coerce_number (number, FLOAT_T, 0UL);
#ifdef HAVE_BIGFLOAT
  else if (EQ (type, Qbigfloat))
    {
      unsigned long prec;

      if (NILP (precision))
	prec = bigfloat_get_default_prec ();
      else
	{
	  CHECK_INTEGER (precision);
#ifdef HAVE_BIGNUM
	  if (INTP (precision))
#endif /* HAVE_BIGNUM */
	    prec = (unsigned long) XREALINT (precision);
#ifdef HAVE_BIGNUM
	  else
	    {
	      if (!bignum_fits_ulong_p (XBIGNUM_DATA (precision)))
		args_out_of_range (precision, Vbigfloat_max_prec);
	      prec = bignum_to_ulong (XBIGNUM_DATA (precision));
	    }
#endif /* HAVE_BIGNUM */
	}
      return internal_coerce_number (number, BIGFLOAT_T, prec);
    }
#endif /* HAVE_BIGFLOAT */

  Fsignal (Qunsupported_type, type);
  /* NOTREACHED */
  return Qnil;
}


void
syms_of_number (void)
{
#ifdef HAVE_BIGNUM
  INIT_LRECORD_IMPLEMENTATION (bignum);
#endif
#ifdef HAVE_RATIO
  INIT_LRECORD_IMPLEMENTATION (ratio);
#endif
#ifdef HAVE_BIGFLOAT
  INIT_LRECORD_IMPLEMENTATION (bigfloat);
#endif

  /* Type predicates */
  DEFSYMBOL (Qrationalp);
  DEFSYMBOL (Qfloatingp);
  DEFSYMBOL (Qrealp);
  DEFSYMBOL (Qbignump);
  DEFSYMBOL (Qratiop);
  DEFSYMBOL (Qbigfloatp);

  /* Functions */
  DEFSUBR (Fbignump);
  DEFSUBR (Fratiop);
  DEFSUBR (Frationalp);
  DEFSUBR (Fnumerator);
  DEFSUBR (Fdenominator);
  DEFSUBR (Fbigfloatp);
  DEFSUBR (Fbigfloat_get_precision);
  DEFSUBR (Fbigfloat_set_precision);
  DEFSUBR (Ffloatingp);
  DEFSUBR (Frealp);
  DEFSUBR (Fcanonicalize_number);
  DEFSUBR (Fcoerce_number);

  /* Errors */
  DEFERROR_STANDARD (Qunsupported_type, Qwrong_type_argument);
}

void
vars_of_number (void)
{
  /* These variables are Lisp variables rather than number variables so that
     we can put bignums in them. */
  DEFVAR_LISP_MAGIC ("default-float-precision", &Vdefault_float_precision, /*
The default floating-point precision for newly created floating point values.
This should be 0 to create Lisp float types, or an unsigned integer no greater
than `bigfloat-maximum-precision' to create Lisp bigfloat types with the
indicated precision.
*/ default_float_precision_changed);
  Vdefault_float_precision = make_int (0);

  DEFVAR_CONST_LISP ("bigfloat-maximum-precision", &Vbigfloat_max_prec /*
The maximum number of bits of precision a bigfloat can have.
This is determined by the underlying library used to implement bigfloats.
*/);

#ifdef HAVE_BIGFLOAT
  /* Don't create a bignum here.  Otherwise, we lose with NEW_GC + pdump.
     See reinit_vars_of_number(). */
  Vbigfloat_max_prec = make_int (EMACS_INT_MAX);
#else
  Vbigfloat_max_prec = make_int (0);
#endif /* HAVE_BIGFLOAT */

  Fprovide (intern ("number-types"));
#ifdef HAVE_BIGNUM
  Fprovide (intern ("bignum"));
#endif
#ifdef HAVE_RATIO
  Fprovide (intern ("ratio"));
#endif
#ifdef HAVE_BIGFLOAT
  Fprovide (intern ("bigfloat"));
#endif
}

void
reinit_vars_of_number (void)
{
#if defined(HAVE_BIGFLOAT) && defined(HAVE_BIGNUM)
  Vbigfloat_max_prec = make_bignum (0L);
  bignum_set_ulong (XBIGNUM_DATA (Vbigfloat_max_prec), ULONG_MAX);
#endif
}

void
init_number (void)
{
  if (!number_initialized)
    {
      number_initialized = 1;

#ifdef WITH_GMP
      init_number_gmp ();
#endif
#ifdef WITH_MP
      init_number_mp ();
#endif

#ifdef HAVE_BIGNUM
      bignum_init (scratch_bignum);
      bignum_init (scratch_bignum2);
#endif

#ifdef HAVE_RATIO
      ratio_init (scratch_ratio);
      ratio_init (scratch_ratio2);
#endif

#ifdef HAVE_BIGFLOAT
      bigfloat_init (scratch_bigfloat);
      bigfloat_init (scratch_bigfloat2);
#endif

#ifndef PDUMP
      reinit_vars_of_number ();
#endif
    }
}