Why does Julia 1.6.5 and 1.7.1 generate differnent random values using the same seed?

I found that if you start a fresh session in both 1.6.5. and 1.7.1 you get different results. I found this glitch while running doctests… and I will never be able to run Documenter.jl with this kind of craziness.

using Random
rng = MersenneTwister(1)

Try it and you will see that doctesting is impossible if you use rand()s

Any thoughts? Thanks :slight_smile:

This is documented. The random number generator is not guaranteed to produce the same values between minor versions.


If you would like a stable stream of random values for a particular seed, try StableRNGs.jl.


Thanks for the packages. The doc seems to address the the whole doctest thing. I will try it out.

1 Like

Turns out I was not passing the rng argument to the rand(). The correct way to seed would be rand(rng, 5)

1 Like

Documenter.jl has filter expressions for the docstrings, which allow you to filter out parts of outputs. Only relying on doctests for functionality is quite brittle, especially with a fixed seed - if possible, I’d recommend testing properties you’d expect of the output in your testsuite instead.