Clojure

'numerator and 'denominator fail to handle integral values (i.e. N/1)

Details

  • Type: Defect Defect
  • Status: Open Open
  • Priority: Major Major
  • Resolution: Unresolved
  • Affects Version/s: Release 1.6
  • Fix Version/s: None
  • Component/s: None
  • Labels:
    None

Description

Because ratio values reduce to lowest terms and, for integral values where the lowest term is N/1, are auto-converted to BigInts (and formerly Longs), the current behavior of clojure.core/numerator and clojure.core/denominator yield unexpected results.

user=> (numerator 1/3)
1
user=> (numerator (+ 1/3 2/3))

ClassCastException clojure.lang.BigInt cannot be cast to clojure.lang.Ratio  clojure.core/numerator (core.clj:3306)
user=> (denominator 1/3)
3
user=> (denominator (+ 1/3 2/3))

ClassCastException clojure.lang.BigInt cannot be cast to clojure.lang.Ratio  clojure.core/denominator (core.clj:3314)
user=>

The auto-conversion to Longs is not really the problem in my mind. I'd like to see numerator return the original value when presented with a BigInt and denominator always return 1 when presented with a BigInt. It seems reasonable to request the same for Longs.

If desired, I'd be happy to produce a patch.

Activity

Hide
Andy Fingerhut added a comment -

I don't know the official stance on this ticket, but will add some notes.

Aaron, numerator and denominator are pretty clearly documented to work on Ratio types only.

It is pretty easy to write my-numerator and my-denominator that work exactly as you wish, checking for the type of arg and using numerator, denominator for Ratio types, and doing whatever you think is correct for other numeric types.

Show
Andy Fingerhut added a comment - I don't know the official stance on this ticket, but will add some notes. Aaron, numerator and denominator are pretty clearly documented to work on Ratio types only. It is pretty easy to write my-numerator and my-denominator that work exactly as you wish, checking for the type of arg and using numerator, denominator for Ratio types, and doing whatever you think is correct for other numeric types.
Hide
Aaron Brooks added a comment - - edited

I'm aware that they are documented as such. Part of my point is that you can be working entirely with Ratio types and, via arithmetic operations between them, sometimes wind up with a non-Ratio number unexpectedly.

Also consider:

user=> (numerator 2/1)
ClassCastException java.lang.Long cannot be cast to clojure.lang.Ratio  clojure.core/numerator (core.clj:3238)

You're then left either implementing a try/catch correction or always checking the type before using numerator or denominator which is a loss in performance.

The patch I have in mind is creating a protocol, extended to Ratio, BigInt and Long which calls the appropriate method (Ratios) or returns either the given number or 1 (numerator/denominator) for the integral types. I expect this to maintain the current level of performance in the cases where it works and behave properly in the cases currently not handled.

Show
Aaron Brooks added a comment - - edited I'm aware that they are documented as such. Part of my point is that you can be working entirely with Ratio types and, via arithmetic operations between them, sometimes wind up with a non-Ratio number unexpectedly. Also consider:
user=> (numerator 2/1)
ClassCastException java.lang.Long cannot be cast to clojure.lang.Ratio  clojure.core/numerator (core.clj:3238)
You're then left either implementing a try/catch correction or always checking the type before using numerator or denominator which is a loss in performance. The patch I have in mind is creating a protocol, extended to Ratio, BigInt and Long which calls the appropriate method (Ratios) or returns either the given number or 1 (numerator/denominator) for the integral types. I expect this to maintain the current level of performance in the cases where it works and behave properly in the cases currently not handled.

People

Vote (1)
Watch (4)

Dates

  • Created:
    Updated: