David Widmann

As far as I can tell, for `Gumbel` currently `xval` is just not used anywhere. I assume for most distributions it is tested implicitly since it is used in a...

I'd consider a `cdf` that returns values < 0 or > 1 a bug. This is the case in your example: ```julia julia> cdf(D, 3.741058503233821e-17) -3.00686029885372e-50 julia> cdf(D, 1.4354474178676617e-18) -3.00686029885372e-50...

IIRC there are quite a few suboptimal implementations in ParetoSmooth (even a few more in the function alone that is touched by this PR), but I haven't really used the...

Regarding this PR more concretely: Can't we avoid parsing strings completely and just work with `VarName`s (as shown in https://github.com/TuringLang/DynamicPPL.jl/blob/3b8dceab76ef37d57d0abdecb205a5c46550679f/src/loglikelihoods.jl#L201-L206)? And do we actually have to sort, given that (nowadays)...

But is it clear whether the sorting is actually needed? https://github.com/TuringLang/ParetoSmooth.jl/pull/97#issuecomment-2285464006

Other packages such as LinearSolve actually use `Base.USE_GPL_LIBS` to disable functionality that requires GPL code: https://github.com/search?q=repo%3ASciML%2FLinearSolve.jl%20USE_GPL_LIBS&type=code

I think this PR can be closed since the problem was already fixed by #1863 and #1888.

I opened a PR a while ago but the reactions where a bit mixed: https://github.com/JuliaStats/KernelDensity.jl/pull/117 The main issue with the AbstractFFTs approach taken in that PR is that as @andreasnoack...

I'll have to refresh my memory on this issue. There's quite a few places in the example above though that will increase memory usage, so I'd have to check to...

IMO (at first glance at least) the current default is reasonable since `@addlogprob!` just indicates that something is added to the joint log density that is computed by default. That...