David Widmann

It always returns "the" probability density, the main question is just with respect to which base measure. Typically for discrete discrete distributions the base measure is the counting measure, which...

Completely agree. I have been thinking about it for a few months but never found the time to put together a PR.

AFAICT this issue (and the suggested fix) are already discussed in https://github.com/JuliaArrays/StaticArrays.jl/issues/1218.

IMO we should add a docstring for generic `fit`, if this can be done in a clear and informative way, before removing the docstring for `Beta`. The possible problem with...

Why was this merged? Docs seem to be broken (missing docstrings): https://juliastats.org/Distributions.jl/dev/fit/

I think the problem are actually `quantile` and `invlogcdf` here. For `Geometric` they should always return an `Int`. For other distributions (and in particular discrete distributions) typically we ensure that...

The `rand` method follows the current design, currently almost all distributions return samples of type `Float64` and `Int` and intentionally the random variate type is independent of the parameter types...

Regarding `len`: IMO we should not add such a field since it is not an actual parameter, could easily cause inconsistencies and I think your example should be fixed differently....

With the current design, the "correct" fix would be to always return Float64 as the variates are supposed to be of type Float64. But as said, I'd rather fix this...

`Normal` is the inconsistent one that does not follow the design. `eltype` is supposed to specify the element type of the random variates, `partype` is the type of the parameters....