rand(1:n) gives numbers outside 1:n

[fcoUnda]

for i=1:100000
j = rand(1:7^7)
if(j<0) println(j) end
end

gives a bunch of negative numbers. I think there might be an abs missing in the last while of rand.

Cheers!

[lindahua]

It seems to be working fine on my mac. I tried the following:

x = rand(1:7^7, 10^7)
all(x .> 0)   # outputs true

and

x = [rand(1:7^7) for i = 1 : 10^7]
all(x .> 0)   # outputs true

So, in both cases, no negative numbers were generated.

[fcoUnda]

Both of these output false for me. I'm on a windows machine. Can anyone reproduce this?

[jiahao]

What version are you running? I cannot reproduce this on osx with 2356fb8.

[quinnj]

I can reproduce on the Windows binary, v0.1.2. I think it's related to #2587 which has been fixed in v0.2.

EDIT: I can't reproduce on my latest Windows 64-bit build (v0.2), just to confirm that it's been resolved.

[fcoUnda]

Yes, I got my version from http:https://code.google.com/p/julialang/downloads/list, so its v0.1.2... my bad :P

[jiahao]

Should be tagged for 0.1.3.

[GunnarFarneback]

To reproduce on 64-bit Julia you need to explicitly make the range Int32.

And yes, there's clearly a bug there. The redrawing loop at the end also needs abs around the rand, except abs is fundamentally not the right operation in that function in the first place.

[GunnarFarneback]

There are two reasons why abs is wrong. First it makes zero underrepresented compared to the non-zero values, skewing the distribution. Not necessarily a huge deal but to a large extent invalidating the redrawing loop at the end. Second, if you are unlucky enough to draw typemin of a signed type with at least as many bits as Int, abs won't change it and you end up with a negative number.