Setting seed < 0 doesn't produce different chains on re-runs

[dmcdougall]

It does produce different chains in parallel, so each processor produces a different chain. However, subsequent runs will produce the same output even in parallel. Here's the relevant part of the documentation that describes this behaviour. While this isn't a bug and the documentation is correct, it might be an idea to add an option to allow the user to seed the the random number generator from reading /dev/urandom.

[pbauman]

I was thinking why not just use the abs of the input option to be the seed? When it's positive that's how the seed is chosen, so why not do the same for the negative case (and keep the current algorithm for the seed on the other processors)? Then we can at least have reproducibility if we need it. Perhaps make the 0 value of this option seed from urandom?