specialized Float32 == Float64 comparison?

[StefanKarpinski]

julia> float32(1/7)
0.14285715

julia> 0.14285715 == float32(1/7)
false

julia> float32(0.14285715) == float32(1/7)
true

I would expect 0.14285715 != 1/7 since the latter has lots of extra digits and the former ought to be presumed to have zeros in all those places.

[StefanKarpinski]

Having poked around a bit, I think we have to live with this:

julia> float32(1/7)
0.14285715

julia> bits(0.14285715)
"0011111111000010010010010010010010100001100111111111100001011000"

julia> bits(float32(1/7))
"00111110000100100100100100100101"

julia> bits(float64(float32(1/7)))
"0011111111000010010010010010010010100000000000000000000000000000"

The only viable alternative is to print Float32 values by upcasting to Float64 first, but then printing them gives long, nasty strings with extra digits:

julia> float64(float32(1/7))
0.1428571492433548

The only other thing that would make this work out is to compare Float32 and Float64 by truncating to Float32, but that's clearly unacceptable.