round for all FloatingPoint types is broken

[rickhg12hs]

julia> versioninfo()                                                                                                                                         
Julia Version 0.4.0-dev+3073
Commit 2f97c78* (2015-02-02 22:28 UTC)
Platform Info:
  System: Linux (x86_64-linux-gnu)
  CPU: AMD Opteron 63xx class CPU
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Piledriver)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

julia> [(n, round(n)) for n::float_type = 0.5:1.0:10.5, float_type = subtypes(FloatingPoint)]                                                                
11x4 Array{(Any,Any),2}:
 (5e-01 with 256 bits of precision,0e+00 with 256 bits of precision)     (float16(0.5),float16(0.0))    (0.5f0,0.0f0)    (0.5,0.0)  
 (1.5e+00 with 256 bits of precision,2e+00 with 256 bits of precision)   (float16(1.5),float16(2.0))    (1.5f0,2.0f0)    (1.5,2.0)  
 (2.5e+00 with 256 bits of precision,2e+00 with 256 bits of precision)   (float16(2.5),float16(2.0))    (2.5f0,2.0f0)    (2.5,2.0)  
 (3.5e+00 with 256 bits of precision,4e+00 with 256 bits of precision)   (float16(3.5),float16(4.0))    (3.5f0,4.0f0)    (3.5,4.0)  
 (4.5e+00 with 256 bits of precision,4e+00 with 256 bits of precision)   (float16(4.5),float16(4.0))    (4.5f0,4.0f0)    (4.5,4.0)  
 (5.5e+00 with 256 bits of precision,6e+00 with 256 bits of precision)   (float16(5.5),float16(6.0))    (5.5f0,6.0f0)    (5.5,6.0)  
 (6.5e+00 with 256 bits of precision,6e+00 with 256 bits of precision)   (float16(6.5),float16(6.0))    (6.5f0,6.0f0)    (6.5,6.0)  
 (7.5e+00 with 256 bits of precision,8e+00 with 256 bits of precision)   (float16(7.5),float16(8.0))    (7.5f0,8.0f0)    (7.5,8.0)  
 (8.5e+00 with 256 bits of precision,8e+00 with 256 bits of precision)   (float16(8.5),float16(8.0))    (8.5f0,8.0f0)    (8.5,8.0)  
 (9.5e+00 with 256 bits of precision,1e+01 with 256 bits of precision)   (float16(9.5),float16(10.0))   (9.5f0,10.0f0)   (9.5,10.0) 
 (1.05e+01 with 256 bits of precision,1e+01 with 256 bits of precision)  (float16(10.5),float16(10.0))  (10.5f0,10.0f0)  (10.5,10.0)

julia>

[jiahao]

This is correct. The default rounding mode changed in 0.4 (see #8750, #9344):

By default ("RoundNearest"), this will round to the nearest integer, with ties (fractional values of 0.5) being rounded to the even integer.

[rickhg12hs]

Wow, I'll have to re-calibrate my understanding (and use) of rounding.

@jiahao : Thanks for the references.