v1.6 major performance regression

[ChrisRackauckas]

#139 shows that master is just fine on v1.5, but #137 shows a massive regression on v1.6.

https://github.com/JuliaDiff/SparseDiffTools.jl/runs/2317160332#step:6:147

When I run it locally, it runs forever and then just goes OOM. I might need some external help here.

[ChrisRackauckas]

Specifically it looks like this test is it: https://github.com/JuliaDiff/SparseDiffTools.jl/blob/master/test/test_acyclic.jl

[JeffBezanson]

Bisects to JuliaLang/julia#37914, so I suspect it's due to different random numbers. It looks like even the size of the graph is random, which could certainly do the trick!

[ChrisRackauckas]

Oh interesting, #141 yeah that did the trick. I'll keep it in mind and maybe make that StableRNGs.jl. Thanks! Definitely not what I was expecting.