Edge case for truncated

[nicomignoni]

Hi!
While using Distributions.jl (v. 0.25.118), I stumbled on this possible corner case

using Distributions

n = truncated(Normal(0, 0), lower=0)
rand(n, 10)

which does not seem to terminate. I think the expected behaviour should be a Vector of 0.

It behaves similarly for (it seems any) lower >= 0, while it works for lower < 0. Could it be that truncated does not check if the distributions can produce a sample within the lower and upper bounds?

Unless I'm mistaken, this occurs only when the distribution to be truncated degenerates into a Dirac distribution. I don't know if it is worth adding a fix to this; in my case, it occurred while constructing distributions for photovoltaic generation data: I have a bunch of null generation values for dawn hours in my dataset, which generated the described bug.

Thank you!

Best,
Nico

[devmotion]

Closing as a duplicate of #1867.