How would you approach the check? Would you do the check on first invocation of the function saving the result for the runtime of the program using something like defonce or at the top of the namespace more or less globally? I don't want to introduce overhead in some bad place, ideally, it should also be visible, which of the implementations is in use - the crypto one or the pseudo-random one.

Btw. the widely support .getRandomValues isn't all that fast - probably the array manipulation has considerable overhead. The .randomUUID version is a bit faster than the current random-uuid, but still considerably slower than the pseudo-random version that I propose.

I think, the array version could be much faster, if we got larger random array at once and just took chunks out of it if multiple random uuids are desired. This would basically require a function with internal state - some kind of "/dev/urandom" as a lazy sequence that would on demand fetch another array if the previous one would be "depleted".

I had an idea how to do that. I have to check if this is thread-safe etc. but this could be it. It seems to be rather efficient too.