set_random(2025) returns false.

[dcnorris]

?- use_module(library(random)).
   true.
?- Seed = 2025, set_random(Seed).
   false. % expected `true`

[bakaq]

?????????????????????????????

[triska]

Expected is a domain error.

?- set_random(seed(2025)).
   true. % expected

Any solutions where the argument is an integer?

?- length(_, L), set_random(L).
   loops, unexpected.

[dcnorris]

Thanks for this clarification, @triska! Random seeds are (in my experience) usually understood to be integers, so I wasn't expecting to need the seed/1 wrapper here. (If the function had reported a domain error, however, that would have given me the needed hint.)

---

This raises the question, however, what useful function seed/1 serves here.

[triska]

The wrapper seed/1 is used to distinguish the argument from other possible arguments which may be supported in the future. Notable examples for this are:

• an attained state of the generator that was previously obtained with a different predicate
• a seed that needs more than one argument
• a way to select a specific kind of generator (?)
• ...

For comparison, please see the SICStus version and possibly other proposals too:

https://sicstus.sics.se/sicstus/docs/latest/html/sicstus.html/lib_002drandom.html

[dcnorris]

I see there's an entire Rust Rand Book, and wonder if the best approach for Scryer might be to leverage this by providing a straightforward interface to those facilities. This could at least serve to flesh out an interface to multiple generators, etc. Presumably, the intense interest in Rust from the scientific community would ensure that this crate is heavily used, tested, etc. Also, someone from that community coming to Scryer would immediately be familiar with the menagerie of RNG options, and able to port work into Scryer as desired.

In any case, I think this is now a non-Issue, and should be closed.

[adri326]

Note that there have been talks in #2670 about doing exactly that :)