David Widmann

OK :slightly_smiling_face: Now the main remaining question is: is it possible to add a test for the fix? I assume it could be a bit challenging since the change only...

> I can think of the Cholesky factorization of the covariance matrix having to be computed only once in the latter case, for example. The factorization is only computed once...

No, ideally we would not mix both paths, also eg for better compatibility with static arrays. Even though probably currently many methods don't work (in an optimized way) with static...

Another reason is that generally it is quite challenging and brittle when one starts to come up with heuristics for the output type.

I meant that generally `logpdf` should be non-mutating, and in particular it should not make any assumptions about the type of the arrays it is called with and eg whether...

Can you make it a weak dependency? Regardless of whether DensityInterface will make it a weak dependency (I assume so), probably we will make DensityInterface a weak dependency in which...

There's already a PR for StatsFuns, just requires someone's approval.

I'm still not completely convinced by this PR, it's a bit unclear to me when these definitions are actually useful and hence justify the increased code complexity (in particular due...

I just ran into the same issue: ```julia using Distributions using Random using Test Random.seed!(1234) v = rand(Beta(0.01, 1.01), 10_000) @test_broken !any(iszero, v) ``` The standard implementation in R does...

I think the main issue with the current implementation is that the Gamma sampler returns incorrect values: ```julia using Distributions using Random using Test Random.seed!(1234) v = rand(Gamma(0.01, 1.0), 10_000)...