This is an investigation of whether implementations return #t or #f to (exact? (sqrt 25/4)). RnRS recommends, but does not require, that the value is #t, but implementations are free to convert 25/4 to an inexact value and return the inexact square root of that. On implementations where exact rationals are not supported, (exact? (sqrt 4)) was substituted.
Supports exact sqrt: Racket, Gauche, MIT, Gambit, Chicken with the numbers egg, Guile, Chez, Vicare, Larceny, Ypsilon, Mosh, IronScheme, STklos, Unlikely, Rep, Sizzle, LMU, Sagittarius, Owl Lisp
Supports only inexact sqrt: plain Chicken, Bigloo, Scheme48/scsh, Kawa, SISC, SCM, NexJ, JScheme, KSi, BDC, Schemik, UMB, Elk, SXM, Dfsch, Inlab, Foment, Picrin, Chibi
Does not support sqrt: SigScheme, Shoe, Llava, FemtoLisp, Oaklisp
Indeterminate results (exact? does not exist and neither eq? nor the printer can be relied on to distinguish): SIOD, XLisp