truncated Normal with zero mean and std hangs

[markmbaum]

I know this is an unusual case, but I noticed that:

using Distributions
using Random

rng = Xoshiro(1)

rand(rng, truncated(Normal(0.0, 0.0), 0, Inf))

hangs and doesn't return a number.

I'm using Distributions v0.25.115 if that's relevant. Possibly related to #1910 but I don't know.

[vandenman]

I've been bitten by this before as well.

Looking at randnt we have tp equals 0.0 so it looks like this will go into an infinite while loop.

Perhaps the truncated normal should have some fallback like tp < eps(tp) && return mean(d)? Alternatively, Normal should require sd > 0.0 (most other distributions don't support degenerate distributions anyway), but that's breaking I guess.

[markmbaum]

If other distributions don't support this kind of degenerate case I would encourage the same for the Normal. I don't think the eps solution is good because it introduces unexpected behavior. Maybe can slate it for the next larger release?

[quildtide]

Duplicate with #1264, #1712, #1867

Would be fixed by #1721