(********************************************************************)
(*                                                                  *)
(*  bigrat.s7i    Big rational number support library               *)
(*  Copyright (C) 2006 - 2009, 2011 - 2015  Thomas Mertes           *)
(*                2017 - 2019, 2023, 2024  Thomas Mertes            *)
(*                                                                  *)
(*  This file is part of the Seed7 Runtime Library.                 *)
(*                                                                  *)
(*  The Seed7 Runtime Library is free software; you can             *)
(*  redistribute it and/or modify it under the terms of the GNU     *)
(*  Lesser General Public License as published by the Free Software *)
(*  Foundation; either version 2.1 of the License, or (at your      *)
(*  option) any later version.                                      *)
(*                                                                  *)
(*  The Seed7 Runtime Library 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 Lesser General Public License for more    *)
(*  details.                                                        *)
(*                                                                  *)
(*  You should have received a copy of the GNU Lesser General       *)
(*  Public License along with this program; if not, write to the    *)
(*  Free Software Foundation, Inc., 51 Franklin Street,             *)
(*  Fifth Floor, Boston, MA  02110-1301, USA.                       *)
(*                                                                  *)
(********************************************************************)


include "bigint.s7i";


(**
 *  Rational numbers represented with [[bigint|bigInteger]] numerator and denominator.
 *  The values of the type ''bigRational'' are finite and periodical
 *  decimal numbers. BigRational literals do not exist. Although
 *  ''bigRational'' operations cannot overflow, it can happen that
 *  there is not enough memory to represent a ''bigRational'' value.
 *  In this case the exception MEMORY_ERROR is raised.
 *)
const type: bigRational is new object struct
    var bigInteger: numerator is 0_;
    var bigInteger: denominator is 1_;
  end struct;


const proc: normalize (inout bigRational: number) is func
  begin
    if number.denominator < 0_ then
      number.numerator := -number.numerator;
      number.denominator := -number.denominator;
    end if;
  end func;


const proc: reduce (inout bigRational: number) is func
  local
    var bigInteger: gcd is 0_;
  begin
    gcd := gcd(number.numerator, number.denominator);
    if gcd >= 2_ then
      number.numerator := number.numerator div gcd;
      number.denominator := number.denominator div gcd;
    end if;
  end func;


(**
 *  Create a ''bigRational'' number from its numerator and denominator.
 *  @return the created ''bigRational'' value.
 *)
const func bigRational: (in bigInteger: numerator) / (in bigInteger: denominator) is func
  result
    var bigRational: aRational is bigRational.value;
  begin
    aRational.numerator := numerator;
    aRational.denominator := denominator;
    normalize(aRational);
    reduce(aRational);
  end func;


(**
 *  Plus sign for ''bigRational'' numbers.
 *  @return its operand unchanged.
 *)
const func bigRational: + (in bigRational: number) is
  return number;


(**
 *  Minus sign, negate a ''bigRational'' number.
 *  @return the negated value of the number.
 *)
const func bigRational: - (in bigRational: number) is func
  result
    var bigRational: negatedNumber is bigRational.value;
  begin
    negatedNumber.numerator := -number.numerator;
    negatedNumber.denominator := number.denominator;
  end func;


(**
 *  Add two ''bigRational'' numbers.
 *  @return the sum of the two numbers.
 *)
const func bigRational: (in bigRational: summand1) + (in bigRational: summand2) is func
  result
    var bigRational: sum is bigRational.value;
  local
    var bigInteger: gcd_denominator is 0_;
  begin
    gcd_denominator := gcd(summand1.denominator, summand2.denominator);
    sum.numerator := (summand1.numerator * summand2.denominator +
        summand2.numerator * summand1.denominator) div gcd_denominator;
    sum.denominator := summand1.denominator div gcd_denominator * summand2.denominator;
    reduce(sum);
  end func;


(**
 *  Compute the subtraction of two ''bigRational'' numbers.
 *  @return the difference of the two numbers.
 *)
const func bigRational: (in bigRational: minuend) - (in bigRational: subtrahend) is func
  result
    var bigRational: difference is bigRational.value;
  local
    var bigInteger: gcd_denominator is 0_;
  begin
    gcd_denominator := gcd(minuend.denominator, subtrahend.denominator);
    difference.numerator := (minuend.numerator * subtrahend.denominator -
        subtrahend.numerator * minuend.denominator) div gcd_denominator;
    difference.denominator := minuend.denominator div gcd_denominator * subtrahend.denominator;
    reduce(difference);
  end func;


(**
 *  Multiply two ''bigRational'' numbers.
 *  @return the product of the two numbers.
 *)
const func bigRational: (in bigRational: factor1) * (in bigRational: factor2) is func
  result
    var bigRational: product is bigRational.value;
  local
    var bigInteger: gcd1 is 0_;
    var bigInteger: gcd2 is 0_;
  begin
    gcd1 := gcd(factor1.numerator, factor2.denominator);
    gcd2 := gcd(factor2.numerator, factor1.denominator);
    product.numerator := (factor1.numerator div gcd1) * (factor2.numerator div gcd2);
    product.denominator := (factor1.denominator div gcd2) * (factor2.denominator div gcd1);
  end func;


(**
 *  Compute the division of two ''bigRational'' numbers.
 *  @return the quotient of the division.
 *  @exception NUMERIC_ERROR If a division by zero occurs.
 *)
const func bigRational: (in bigRational: dividend) / (in bigRational: divisor) is func
  result
    var bigRational: quotient is bigRational.value;
  local
    var bigInteger: gcd1 is 0_;
    var bigInteger: gcd2 is 0_;
  begin
    gcd1 := gcd(dividend.numerator, divisor.numerator);
    gcd2 := gcd(divisor.denominator, dividend.denominator);
    quotient.numerator := (dividend.numerator div gcd1) * (divisor.denominator div gcd2);
    quotient.denominator := (dividend.denominator div gcd2) * (divisor.numerator div gcd1);
    normalize(quotient);
  end func;


(**
 *  Increment a ''bigRational'' number by a delta.
 *)
const proc: (inout bigRational: number) +:= (in bigRational: delta) is func
  local
    var bigInteger: gcd_denominator is 0_;
  begin
    gcd_denominator := gcd(number.denominator, delta.denominator);
    number.numerator := (number.numerator * delta.denominator +
        delta.numerator * number.denominator) div gcd_denominator;
    number.denominator *:= delta.denominator div gcd_denominator;
    reduce(number);
  end func;


(**
 *  Decrement a ''bigRational'' number by a delta.
 *)
const proc: (inout bigRational: number) -:= (in bigRational: delta) is func
  local
    var bigInteger: gcd_denominator is 0_;
  begin
    gcd_denominator := gcd(number.denominator, delta.denominator);
    number.numerator := (number.numerator * delta.denominator -
        delta.numerator * number.denominator) div gcd_denominator;
    number.denominator *:= delta.denominator div gcd_denominator;
    reduce(number);
  end func;


(**
 *  Multiply a ''bigRational'' number by a factor and assign the result back to number.
 *)
const proc: (inout bigRational: number) *:= (in bigRational: factor) is func
  begin
    number.numerator *:= factor.numerator;
    number.denominator *:= factor.denominator;
    reduce(number);
  end func;


(**
 *  Divide a ''bigRational'' number by a divisor and assign the result back to number.
 *)
const proc: (inout bigRational: number) /:= (in bigRational: divisor) is func
  begin
    number.numerator *:= divisor.denominator;
    number.denominator *:= divisor.numerator;
    normalize(number);
    reduce(number);
  end func;


(**
 *  Compute the exponentiation of a ''bigRational'' base with an integer exponent.
 *  @return the result of the exponentiation.
 *)
const func bigRational: (in bigRational: base) ** (in integer: exponent) is func
  result
    var bigRational: power is bigRational.value;
  begin
    if exponent >= 0 then
      power.numerator := base.numerator ** exponent;
      power.denominator := base.denominator ** exponent;
    else
      power.numerator := base.denominator ** (-exponent);
      power.denominator := base.numerator ** (-exponent);
      normalize(power);
    end if;
  end func;


(**
 *  Check if two ''bigRational'' numbers are equal.
 *  @return TRUE if both numbers are equal,
 *          FALSE otherwise.
 *)
const func boolean: (in bigRational: number1) = (in bigRational: number2) is
  return number1.numerator   = number2.numerator and
         number1.denominator = number2.denominator;


(**
 *  Check if two ''bigRational'' numbers are not equal.
 *  @return FALSE if both numbers are equal,
 *          TRUE otherwise.
 *)
const func boolean: (in bigRational: number1) <> (in bigRational: number2) is
  return number1.numerator   <> number2.numerator or
         number1.denominator <> number2.denominator;


(**
 *  Check if number1 is less than number2.
 *  @return TRUE if number1 is less than number2,
 *          FALSE otherwise.
 *)
const func boolean: (in bigRational: number1) < (in bigRational: number2) is
  return number1.numerator * number2.denominator <
         number2.numerator * number1.denominator;


(**
 *  Check if number1 is greater than number2.
 *  @return TRUE if number1 is greater than number2,
 *          FALSE otherwise.
 *)
const func boolean: (in bigRational: number1) > (in bigRational: number2) is
  return number1.numerator * number2.denominator >
         number2.numerator * number1.denominator;


(**
 *  Check if number1 is less than or equal to number2.
 *  @return TRUE if number1 is less than or equal to number2,
 *          FALSE otherwise.
 *)
const func boolean: (in bigRational: number1) <= (in bigRational: number2) is
  return number1.numerator * number2.denominator <=
         number2.numerator * number1.denominator;


(**
 *  Check if number1 is greater than or equal to number2.
 *  @return TRUE if number1 is greater than or equal to number2,
 *          FALSE otherwise.
 *)
const func boolean: (in bigRational: number1) >= (in bigRational: number2) is
  return number1.numerator * number2.denominator >=
         number2.numerator * number1.denominator;


(**
 *  Compare two ''bigRational'' numbers.
 *  @return -1, 0 or 1 if the first argument is considered to be
 *          respectively less than, equal to, or greater than the
 *          second.
 *)
const func integer: compare (in bigRational: number1, in bigRational: number2) is
  return compare(number1.numerator * number2.denominator,
                 number2.numerator * number1.denominator);


(**
 *  Compute the hash value of a ''bigRational'' number.
 *  @return the hash value.
 *)
const func integer: hashCode (in bigRational: number) is
  return hashCode(number.numerator) mod 16#40000000 + hashCode(number.denominator) mod 16#40000000;


(**
 *  Return the conversion of a [[bigint|bigInteger]] to a ''bigRational''.
 *)
const func bigRational: rat (in bigInteger: number) is func
  result
    var bigRational: aRational is bigRational.value;
  begin
    aRational.numerator := number;
    aRational.denominator := 1_;
  end func;


(**
 *  Return the conversion of an integer to a ''bigRational''.
 *)
const func bigRational: bigRational (in integer: number) is func
  result
    var bigRational: aRational is bigRational.value;
  begin
    aRational.numerator := bigInteger(number);
    aRational.denominator := 1_;
  end func;


(**
 *  Return the conversion of an integer to a ''bigRational''.
 *)
const func bigRational: (attr bigRational) conv (in integer: number) is func
  result
    var bigRational: aRational is bigRational.value;
  begin
    aRational.numerator := bigInteger(number);
    aRational.denominator := 1_;
  end func;


(**
 *  Return the conversion of a bigInteger to a ''bigRational''.
 *)
const func bigRational: bigRational (in bigInteger: number) is func
  result
    var bigRational: aRational is bigRational.value;
  begin
    aRational.numerator := number;
    aRational.denominator := 1_;
  end func;


(**
 *  Return the conversion of a bigInteger to a ''bigRational''.
 *)
const func bigRational: (attr bigRational) conv (in bigInteger: number) is func
  result
    var bigRational: aRational is bigRational.value;
  begin
    aRational.numerator := number;
    aRational.denominator := 1_;
  end func;


(**
 *  Compute the absolute value of a ''bigRational'' number.
 *  @return the absolute value.
 *)
const func bigRational: abs (in bigRational: number) is func
  result
    var bigRational: absoluteValue is bigRational.value;
  begin
    absoluteValue.numerator := abs(number.numerator);
    absoluteValue.denominator := number.denominator;
  end func;


(**
 *  Return a ''bigRational'' number truncated towards negative infinity.
 *)
const func bigInteger: floor (in bigRational: number) is
  return number.numerator mdiv number.denominator;


(**
 *  Return a ''bigRational'' number rounded up towards positive infinity.
 *)
const func bigInteger: ceil (in bigRational: number) is
  return -(number.numerator mdiv -number.denominator);


(**
 *  Return a ''bigRational'' number truncated towards zero.
 *)
const func bigInteger: trunc (in bigRational: number) is
  return number.numerator div number.denominator;


(**
 *  Round a ''bigRational'' number to the nearest [[bigint|bigInteger]].
 *  Halfway cases are rounded away from zero.
 *  @return the rounded value.
 *)
const func bigInteger: round (in bigRational: number) is func
  result
    var bigInteger: int_val is 0_;
  begin
    if number.numerator >= 0_ then
      int_val := (2_ * number.numerator + number.denominator) div (2_ * number.denominator);
    else
      int_val := (2_ * number.numerator - number.denominator) div (2_ * number.denominator);
    end if;
  end func;


(**
 *  Round a ''bigRational'' number with a decimal ''precision''.
 *  Halfway cases are rounded away from zero.
 *  @return the rounded value.
 *)
const func bigRational: round10 (in bigRational: number, in integer: precision) is func
  result
    var bigRational: rounded is bigRational.value;
  begin
    if precision < 0 then
      rounded.numerator := (abs(number.numerator) div 10_ ** pred(-precision) +
                           number.denominator * 5_) div (number.denominator * 10_) *
                           10_ ** (-precision);
      rounded.denominator := 1_;
    else
      rounded.numerator := (abs(number.numerator) * 10_ ** succ(precision) +
                           number.denominator * 5_) div (number.denominator * 10_);
      rounded.denominator := 10_ ** precision;
    end if;
    if number.numerator < 0_ then
      rounded.numerator := -rounded.numerator;
    end if;
  end func;


const type: BIG_REMAINDER_HASH_TYPE is hash [bigInteger] integer;


(**
 *  Convert a ''bigRational'' number to a [[string]].
 *  The number is converted to a [[string]] with a decimal representation
 *  (e.g.: "1.25"). The representation has a decimal point and at
 *  least one digit before and after the decimal point. Negative
 *  numbers are preceded by a minus sign (e.g.: "-1.25"). The
 *  decimal number can have repeating decimals, which are enclosed
 *  in parentheses ("e.g.: "0.(3)"). The repeating decimals will
 *  not start before the decimal point.
 *  @return the string result of the conversion.
 *)
const func string: str (in bigRational: number) is func
  result
    var string: stri is "";
  local
    var BIG_REMAINDER_HASH_TYPE: remainderHash is BIG_REMAINDER_HASH_TYPE.value;
    var quotRem: quotAndRem is quotRem.value;
    var integer: pos is 0;
  begin
    if number.denominator = 0_ then
      if number.numerator > 0_ then
        stri := "Infinity";
      elsif number.numerator = 0_ then
        stri := "NaN";
      else
        stri := "-Infinity";
      end if;
    else
      quotAndRem := abs(number.numerator) divRem number.denominator;
      stri := str(quotAndRem.quotient);
      stri &:= ".";
      if quotAndRem.remainder = 0_ then
        stri &:= "0";
      else
        repeat
          remainderHash @:= [quotAndRem.remainder] length(stri);
          quotAndRem.remainder *:= 10_;
          quotAndRem := quotAndRem.remainder divRem number.denominator;
          stri &:= str(quotAndRem.quotient);
        until quotAndRem.remainder = 0_ or quotAndRem.remainder in remainderHash;
        if quotAndRem.remainder <> 0_ then
          pos := remainderHash[quotAndRem.remainder];
          stri := stri[.. pos] & "(" & stri[succ(pos) ..] & ")";
        end if;
      end if;
      if number.numerator < 0_ then
        stri := "-" & stri;
      end if;
    end if;
  end func;


(**
 *  Convert a ''bigRational'' number to a ''bigRational'' literal.
 *  @return the ''bigRational'' literal.
 *)
const func string: literal (in bigRational: number) is
    return "bigRational(" & literal(str(number)) & ")";


(**
 *  Convert a ''bigRational'' to a [[string]] with a fraction.
 *   fraction(bigRational("0.(3)"))  returns  "1/3"
 *  @return the string with the corrresponding fraction.
 *)
const func string: fraction (in bigRational: number) is
    return str(number.numerator) & "/" & str(number.denominator);


(**
 *  Convert a ''bigRational'' number to a [[string]] in decimal fixed point notation.
 *  The number is rounded to the specified number of digits (''precision'').
 *  Halfway cases are rounded away from zero. Except for a ''precision'' of
 *  zero the representation has a decimal point and at least one digit
 *  before and after the decimal point. Negative numbers are preceded by
 *  a minus sign (e.g.: "-1.25"). If all digits in the result are 0 a
 *  possible negative sign is omitted.
 *   1_/64_ digits 7     returns "0.0156250"
 *   1_/64_ digits 4     returns "0.0156"
 *   1_/64_ digits 2     returns "0.02"
 *   355_/113_ digits 6  returns "3.141593"
 *   22_/7_ digits 0     returns "3"
 *   -1_/2_ digits 1     returns "-1"
 *   1_/0_ digits 5      returns "Infinity"
 *   -1_/0_ digits 6     returns "-Infinity"
 *   0_/0_ digits 7      returns "NaN"
 *   -1_/2048_ digits 3  returns "0.000"
 *  @param precision Number of digits after the decimal point.
 *         If the ''precision'' is zero the decimal point is omitted.
 *  @return the string result of the conversion.
 *  @exception RANGE_ERROR If the ''precision'' is negative.
 *)
const func string: (in bigRational: number) digits (in integer: precision) is func
  result
    var string: stri is "";
  local
    var bigInteger: mantissa is 0_;
  begin
    if precision < 0 then
      raise RANGE_ERROR;
    elsif number.denominator = 0_ then
      if number.numerator > 0_ then
        stri := "Infinity";
      elsif number.numerator = 0_ then
        stri := "NaN";
      else
        stri := "-Infinity";
      end if;
    else
      mantissa := (abs(number.numerator) * 10_ ** succ(precision) +
                  number.denominator * 5_) div (number.denominator * 10_);
      stri := str(mantissa);
      if precision >= length(stri) then
        stri := "0" mult (precision - length(stri) + 1) & stri;
      end if;
      if precision <> 0 then
        stri := stri[ .. length(stri) - precision] & "." &
            stri[length(stri) - precision + 1 .. ];
      end if;
      if number.numerator < 0_ and mantissa <> 0_ then
        stri := "-" & stri;
      end if;
    end if;
  end func;


const func integer: decimalExponent (in bigRational: number) is func
  result
    var integer: exponent is 0;
  begin
    if abs(number.numerator) >= number.denominator then
      exponent := ord(log10(abs(number.numerator) div number.denominator));
    else
      exponent := -ord(log10(number.denominator div abs(number.numerator))) - 1;
    end if;
  end func;


(**
 *  Convert a ''bigRational'' number to a [[string]] in scientific notation.
 *  Scientific notation uses a decimal significand and a decimal exponent.
 *  The significand has an optional sign and exactly one digit before the
 *  decimal point. The fractional part of the significand is rounded
 *  to the specified number of digits (''precision''). Halfway cases are
 *  rounded away from zero. The fractional part is followed by the
 *  letter e and an exponent, which is always signed. The value zero is
 *  never written with a negative sign.
 *   1_/64_ sci 4     returns "1.5625e-2"
 *   1_/64_ sci 3     returns "1.563e-2"
 *   1_/64_ sci 2     returns "1.56e-2"
 *   355_/113_ sci 6  returns "3.141593e+0"
 *   22_/7_ sci 0     returns "3e+0"
 *   -1_/2_ sci 1     returns "-5.0e-1"
 *   1_/0_ sci 5      returns "Infinity"
 *   -1_/0_ sci 6     returns "-Infinity"
 *   0_/0_ sci 7      returns "NaN"
 *   -1_/2048_ sci 3  returns "-4.883e-4"
 *   -0_/1_ sci 2     returns "0.00e+0"
 *  @param precision Number of digits after the decimal point.
 *         If the ''precision'' is zero the decimal point is omitted.
 *  @return the string result of the conversion.
 *  @exception RANGE_ERROR If the ''precision'' is negative.
 *)
const func string: (in bigRational: number) sci (in integer: precision) is func
  result
    var string: stri is "";
  local
    var integer: exponent is 0;
    var bigInteger: mantissa is 0_;
  begin
    if precision < 0 then
      raise RANGE_ERROR;
    elsif number.denominator = 0_ then
      if number.numerator > 0_ then
        stri := "Infinity";
      elsif number.numerator = 0_ then
        stri := "NaN";
      else
        stri := "-Infinity";
      end if;
    elsif number.numerator = 0_ then
      if precision = 0 then
        stri := "0e+0";
      else
        stri := "0." & "0" mult precision & "e+0";
      end if;
    else
      exponent := decimalExponent(number);
      if succ(precision) >= exponent then
        mantissa := (abs(number.numerator) * 10_ ** succ(precision - exponent) +
                    number.denominator * 5_) div (number.denominator * 10_);
      else
        mantissa := (abs(number.numerator) div 10_ ** pred(exponent - precision) +
                    number.denominator * 5_) div (number.denominator * 10_);
      end if;
      stri := str(mantissa);
      if length(stri) > succ(precision) then
        # Rounding up increased the number of digits.
        incr(exponent);
        stri := stri[.. succ(precision)];
      end if;
      if precision <> 0 then
        stri := stri[1 fixLen 1] & "." & stri[2 .. ];
      end if;
      if exponent >= 0 then
        stri &:= "e+" <& exponent;
      else
        stri &:= "e" <& exponent;
      end if;
      if number.numerator < 0_ then
        stri := "-" & stri;
      end if;
    end if;
  end func;


(**
 *  Convert a [[string]] to a ''bigRational'' number.
 *  The [[string]] must contain a fraction (e.g.: "3/5") or a decimal number
 *  (e.g.: "1.25"). In a fraction numerator and denominator are separated
 *  with a slash (/). A decimal number can have repeating decimals,
 *  which are enclosed in parentheses ("e.g.: "0.(3)"). The repeating
 *  decimals are not allowed to start before the decimal point.
 *   bigRational("3/5")         returns   3_ /   5_
 *   bigRational("1.25")        returns   5_ /   4_
 *   bigRational("0.(3)")       returns   1_ /   3_
 *   bigRational("1.23(45)")    returns 679_ / 550_
 *   bigRational("3.(142857)")  returns  22_ /   7_
 *   bigRational("0.(846153)")  returns  11_ /  13_
 *  @return the ''bigRational'' result of the conversion.
 *  @exception RANGE_ERROR If stri contains not a valid ''bigRational'' value.
 *)
const func bigRational: bigRational (in var string: stri) is func
  result
    var bigRational: aRational is bigRational.value;
  local
    var boolean: negative is FALSE;
    var string: fraction is "";
    var string: period is "";
  begin
    if stri <> "" then
      if stri[1] = '-' then
        stri := stri[2 ..];
        negative := TRUE;
      elsif stri[1] = '+' then
        stri := stri[2 ..];
      end if;
      aRational.numerator := bigInteger(getint(stri));
      if stri[1] = '/' then
        stri := stri[2 ..];
        aRational.denominator := bigInteger(getint(stri));
        reduce(aRational);
      elsif stri[1] = '.' then
        stri := stri[2 ..];
        if startsWith(stri, "(") then
          stri := stri[2 ..];
          period := getint(stri);
          aRational.denominator := 1_;
          aRational +:= bigInteger(period) / pred(10_ ** length(period));
          if stri[1] = ')' then
            stri := stri[2 ..];
          end if;
        else
          fraction := getint(stri);
          aRational.denominator := 10_ ** length(fraction);
          aRational.numerator *:= aRational.denominator;
          aRational.numerator +:= bigInteger(fraction);
          if startsWith(stri, "(") then
            stri := stri[2 ..];
            period := getint(stri);
            aRational +:= bigInteger(period) / (pred(10_ ** length(period)) * aRational.denominator);
            if stri[1] = ')' then
              stri := stri[2 ..];
            end if;
          end if;
        end if;
        reduce(aRational);
      end if;
      if stri <> "" then
        raise RANGE_ERROR;
      elsif negative then
        aRational.numerator := -aRational.numerator;
      end if;
    else
      raise RANGE_ERROR;
    end if;
  end func;


(**
 *  Convert a [[string]] to a ''bigRational'' number.
 *  The [[string]] must contain a fraction (e.g.: "3/5") or a decimal number
 *  (e.g.: "1.25"). In a fraction numerator and denominator are separated
 *  with a slash (/). A decimal number can have repeating decimals,
 *  which are enclosed in parentheses ("e.g.: "0.(3)"). The repeating
 *  decimals are not allowed to start before the decimal point.
 *   bigRational parse "3/5"         returns   3_ /   5_
 *   bigRational parse "1.25"        returns   5_ /   4_
 *   bigRational parse "0.(3)"       returns   1_ /   3_
 *   bigRational parse "1.23(45)"    returns 679_ / 550_
 *   bigRational parse "3.(142857)"  returns  22_ /   7_
 *   bigRational parse "0.(846153)"  returns  11_ /  13_
 *  @return the ''bigRational'' result of the conversion.
 *  @exception RANGE_ERROR If stri contains not a valid ''bigRational'' value.
 *)
const func bigRational: (attr bigRational) parse (in string: stri) is
    return bigRational(stri);


enable_io(bigRational);
DECLARE_TERNARY(bigRational);
DECLARE_MIN_MAX(bigRational);


# Allows 'array bigRational' everywhere without extra type definition.
const type: _bigRationalArray is array bigRational;