Inconsistency with integer valued floats on binomial and multinomial distributions

[mauricelanghinrichs]

Not a major issue, but I observed inconsistent behaviour on how the binomial and multinomial distributions behave on integer valued floats for the trial number $n$. For example, $n=10.0$ works for

julia> rand(Binomial(10.0, 0.6))
7

but will fail in the equivalent multinomial expression:

julia> rand(Multinomial(10.0, [0.6, 0.4]))
ERROR: MethodError: no method matching Multinomial(::Float64, ::Vector{Float64})
The type `Multinomial` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  Multinomial(::Integer, ::AbstractVector{T}; check_args) where T<:Real
   @ Distributions ~/.julia/packages/Distributions/xMnxM/src/multivariate/multinomial.jl:28
  Multinomial(::Integer, ::Integer; check_args)
   @ Distributions ~/.julia/packages/Distributions/xMnxM/src/multivariate/multinomial.jl:37

Stacktrace:
 [1] top-level scope
   @ REPL[37]:1

I feel they should be consistent. And, in my view, preferably, the multinomial implementation should handle these cases as the binomial version.

Maybe there are specific reasons for the current implementation however that I'm not aware of. There might be similar issues for other distributions; I don't know and did not check systematically.

EDIT:
I found the code for the binomial version on real n to be in scr/deprecates.jl:

function Binomial(n::Real, p::Real)
    Base.depwarn("Binomial(n::Real, p) is deprecated. Please use Binomial(n::Integer, p) instead.", :Binomial)
    Binomial(Int(n), p)
end

However, I didn't obtain a deprecation warning on my system (VS code) so that the implementation looked intentional to me..

[devmotion]

I didn't obtain a deprecation warning on my system (VS code)

Unfortunately (IMO), julia by default is started with the commandline flag --depwarn=no, which implies that deprecation warnings are not displayed. A user manually has to set it to --depwarn=yes (default if you run tests, deprecation warnings are displayed) or --depwarn=error (deprecation warnings throw a hard error).

Since I think that's the wrong behaviour and users should always see deprecation warnings (how would they otherwise know they should change their code?), we should consider only using Base.depwarn(...; force=true) in Distributions which forces the display of the deprecation warnings regardless of the commandline flag.

[mauricelanghinrichs]

I have no clue / personal opinion about whether the deprecation warning should be on or off. But in any case, this explains why I didn't see a warning, and turning the warning on by default will at least solve/explain the concrete issue I reported here, as currently intended but deprecated behavior. Thanks!