Project

General

Profile

Bug #8826

Updated by mrkn (Kenta Murata) about 3 years ago

BigDecimal's #div and #quo method behave differently despite the #div documentation says: "See BigDecimal#quo" 

 #div returns Fixnum if there is no precision argument for #div (this parameter is not documented): 

 ``` 
 2.0.0-p247 :018 > BigDecimal(5).div(5).class 
  => Fixnum  
 2.0.0-p247 :031 > BigDecimal(5).div(5.1).class 
  => Fixnum 
 ``` 

 #div returns Fixnum even for a Float argument: 

 ``` 
 2.0.0-p247 :118 > BigDecimal(5).div(5.01) 
  => 0  
 ``` 

  

 It returns Fixnum even if every argument is BigDecimal: 

 ``` 
 2.0.0-p247 :043 > BigDecimal(5).div(BigDecimal(5.1,5)).class 
  => Fixnum 
 ``` 

 When provided the precision argument, #div returns BigDecimal: 

 ``` 
 2.0.0-p247 :036 > BigDecimal(5).div(5,8).class 
  => BigDecimal  
 2.0.0-p247 :131 > BigDecimal(5).div(BigDecimal(5.1,5),8).class 
  => BigDecimal  
 ``` 

  

 But first argument cannot be Float along with precision: 

 ``` 
 2.0.0-p247 :032 > BigDecimal(5).div(5.1,8).class 
 ArgumentError: Float can't be coerced into BigDecimal without a precision 
	 from (irb):32:in `div' 
	 from (irb):32 
	 from /home/karatedog/.rvm/rubies/ruby-2.0.0-p247/bin/irb:13:in `<main>' 
 ``` 

 Whereas #quo does not accept a precision argument and returns BigDecimal (hence no configurable precision here, although the documentation says that #quo applies round operation): 

 ``` 
 2.0.0-p247 :121 > BigDecimal(5).quo(5.01) 
  => #<BigDecimal:8bacf68,'0.9980039920 1596806387 225549E0',27(45)>  
 ``` 

  

 Circumventing the precision with class method does not work on #quo, it's like the limit is maxed: 

 ``` 
 2.0.0-p247 :135 > BigDecimal::limit(5) 
  => 5  
 2.0.0-p247 :136 > BigDecimal(1).quo(3) 
  => #<BigDecimal:899c778,'0.33333E0',9(36)>  
 2.0.0-p247 :080 > BigDecimal::limit(50) 
  => 5  
 2.0.0-p247 :081 > BigDecimal(1).quo(3) 
  => #<BigDecimal:8a92d94,'0.3333333333 33333333E0',18(36)>  
 ``` 

  

 Precision does not seem to be automatic: 

 ``` 
 2.0.0-p247 :141 > BigDecimal::limit(500) 
  => 100  
 2.0.0-p247 :142 > BigDecimal(1).quo(229) 
  => #<BigDecimal:8be67f4,'0.4366812227 074236E-2',18(36)>  
 ``` 

 229 is a full period prime, its reciprocal yields 228 fractional digits before repetition.  

 Whereas #div's precision can be larger than #div's: 

 ``` 
 2.0.0-p247 :109 > BigDecimal(1).div(3,19) 
  => #<BigDecimal:8acb2d4,'0.3333333333 333333333E0',27(54)> 
 ``` 

 And for 229: 

 ``` 
 2.0.0-p247 :144 > BigDecimal(1).div(229,250) 
  => #<BigDecimal:8bc8b28,'0.4366812227 0742358078 6026200873 3624454148 4716157205 2401746724 8908296943 2314410480 3493449781 6593886462 8820960698 6899563318 7772925764 1921397379 9126637554 5851528384 2794759825 3275109170 3056768558 9519650655 0218340611 3537117903 9301310043 6681222707... 
 ``` 

 Expected behavior: 
 - One division method to rule them all :-) (#Division) 
 - Never truncate a result (aka no Fixnum/Bignum as result). If someone uses BigDecimal, they probably wanted large precision instead of truncating the results by default. 
 - #Division should accept Float, Rational, String, Complex, Integer, BigDecimal as divisor, even Float w/o precision. (This is intended as a full list of acceptable classes, #div and #quo can already take different classes as arguments).  
 - #Division should accept a precision argument which would override ::Limit (as this happens in many instance method), this argument is optional. Without precision argument, use ::Limit 

 as for now proper calculation only happens if: 
 - method is #div 
 - the divisor is converted to BigDecimal 
 - a precision argument is given to #div 

Back