ridgeworks

@yoquierotacob 's analysis is correct: `float_parts/4` uses the `frexp` function from C which, I assume, is defined in the ISO-C standard and therefore unlikely to change. It appears this is...

I see you've made progress solving your particular issue, but for the record: I am unaware of any particular motivating use case for `float_parts/4`. However, all the atomic types have...

> I don't think the Prolog engine should expose these. ... And I totally agree. But I do have an issue with: > It is really easy to do so...

Just some performance numbers: ``` Welcome to SWI-Prolog (threaded, 64 bits, version 8.3.25) SWI-Prolog comes with ABSOLUTELY NO WARRANTY. This is free software. Please run ?- license. for legal details....

I suspected changing `nb_current/2` would be problematical but, if I understand your PR, this change only helps if there are multiple attempts to read the same undefined global variable. To...

Nope, see my second post in this issue: 1 global var (`'$$$')`. Accessing it using `nb_current` takes less than 0.1 microseconds including backtracking loop overhead. Same call on a non-existant...

"Then I do not understand. You get one call to exception/3 for the undefined ??? var and thus..." In my case (no patch) there is a potential call to exception...

I don't get to pick a user's platform of choice. So if a given "feature" is unacceptable for an intended purpose on one platform, it's unusable on all IMO. In...

> ...We keep track of the number of glbal vars that need this threadment and if there is at least one we process the entire table. I think I see...

Unreproducible in 8.5.18 ; believed to be fixed with latest round of changes to `mpz_fdiv()`.