Julia: Error in hash of Rational{Int}

[Jutho]

There is a subtle bug in the hashing code of Rational{<:BigInteger64}, in the file hashing2.jl, line 158:

            return hash(ldexp(Float64(num),pow))

This does not pass the incoming hash argument h, and thus, while it produces the correct result when just comparing hash(3//2) with hash(1.5), it does not produce the same result for hash(3//2,h) and hash(1.5,h) for any non-zero h.

I would fix myself but I am traveling and hence not in my normal workflow.

Pinging @StefanKarpinski as I think he wrote the hashing code.

[StefanKarpinski]

Good catch! Yes, this seems like a simple enough fix – just pass the h argument on this line:

https://github.com/JuliaLang/julia/blob/ac389b016702666853bdb80d521a02977705bba1/base/hashing2.jl#L158

I.e. change it to

return hash(ldexp(Float64(num), pow), h)

Then add a test to verify that this is done correctly.

[StefanKarpinski]

Note that we're not going to backport this to 0.4 or 0.5 at this point, but I've tagged this anyway since it could be backported to those releases.

[KristofferC]

What prevents it from being backported to 0.3 though?

[StefanKarpinski]

Only the lack of a label. I drew the line at creating a label for backports that we will never do.