Use StableRNGs for adapted_grid()

[ffreyer]

Julia 1.11 changed something about random number generation which causes rand(MersenneTwister(1337)) to return a different result from <1.11. This results in a different point selection in adapted_grid and causes a test failure in Makie. Making rng a kwarg doesn't fix the difference between versions, but allows for other packages to use StableRNG if they care about consistency

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 62.15%. Comparing base (f16e07a) to head (39d2e34).
Report is 6 commits behind head on master.

❗ There is a different number of reports uploaded between BASE (f16e07a) and HEAD (39d2e34). Click for more details.

HEAD has 3 uploads less than BASE

Flag	BASE (f16e07a)	HEAD (39d2e34)
	9	6

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #169      +/-   ##
==========================================
- Coverage   72.05%   62.15%   -9.91%     
==========================================
  Files           7        6       -1     
  Lines         662      650      -12     
==========================================
- Hits          477      404      -73     
- Misses        185      246      +61     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[asinghvi17]

Looks good to me, thanks!

[ffreyer]

I'm not sure about this yet. The master version always reset the RNG so the results are always consistent for same inputs. This may or may not, depending on how it's used. E.g. if you set const RNG = Xoshiro(123) and pass that, you will advance the rng between calls and not get the same output. Re-seeding the rng is dodgy, since it could also be Random.default_rng().

I wonder if this could use a weaker factor instead, or if we should just make this a StableRNG in general.

[BeastyBlacksmith]

we should just make this a StableRNG in general.

I'd say we do this, since we just need some random numbers. Possibly making the rand_factor a callsite argument, since people might be more interested to tweak that instead of the RN stream.

[jkrumbiegel]

we should just make this a StableRNG in general.

Sounds good to me, and if you really want to give people the ability to try different seeds then you could also just make the seed itself a kwarg and then do rng = StableRNG(seed) in the function. But for me it would also be fine without that.

[ffreyer]

This fails:

let f = sinc, int = (0, 40)  # JuliaPlots/Plots.jl/issues/3894
    xs, fs = adapted_grid(f, int)
    @test length(xs) > 400
end

but it doesn't reproduce the problem from the issue. With lines(-40..40, sinc) in Makie:

The 0..40 range also looks good.