Add setutils.complement, setutils.fullSet

[beef331]

Adds complement for builtin sets which inverts the set. Adds fullset by extension as it was required. Breaks the JS testing for an unevident reason.

(EDIT)

notes

proc fullSet*(T: typedesc): auto {.inline.} =
=>
proc fullSet*(T: typedesc): set[T] {.inline.} = would give a BUG

[timotheecour]

IMO the only thing left here is decide whether to use not or something else (eg complement)

IMO complement is preferred for clarity, there's not much value in conflating logical not with set not; the only (or main) times where it makes sense to reuse a name is where it helps generic code, but I can't think of a single non-contrived case where this would hold in this case.

Please vote with a reaction to this comment:

• -1 for keeping not
• +1 for using something different (complement or something else)

note

this is related to nim-lang/RFCs#102; IMO or, not etc for integers should've been bitor, bitnot etc (more searchable); while this may (or not) be too late for those operators, we shouldn't make this worse by extending this to set. refs also #11372 (comment)

[konsumlamm]

IMO complement is preferred for clarity, there's not much value in conflating logical not with set not; the only (or main) times where it makes sense to reuse a name is where it helps generic code, but I can't think of a single non-contrived case where this would hold in this case.

Note that not is also used for bitwise complement for integer types (https://nim-lang.org/docs/system.html#not%2Cint).

P.S.: There is a typo in the title, it should be "setutils.fullSet".

[timotheecour]

Note that not is also used for bitwise complement for integer types (nim-lang.org/docs/system.html#not%2Cint).

yes, but that's an IMO unfortunate historical artifact that we shouldn't make worse, see nim-lang/RFCs#102

[beef331]

I'd say the best name to be is invert or inverted just so it's a more likely to be understood name. complement is rather obscure to the uninitiated.

[timotheecour]

I'd say the best name to be is invert or inverted just so it's a more likely to be understood name. complement is rather obscure to the uninitiated.

IMO complement the most commonly accepted terminology, by far, eg https://en.wikipedia.org/wiki/Complement_(set_theory) and lots of other sources.

set inverse returns much fewer hits on google than set complement, and also the top hits don't refer to this notion.

[Araq]

Yes, it should be complement.

[beef331]

Well since the consensus seems to be to (ironically) not use not, renamed it.

[timotheecour]

nice

[timotheecour]

@beef331 CI fails, see #17066 (comment) for final recommendation before this can be merged

[timotheecour]

@beef331 CI still fails, after investigating in #17095, it's because defined(bsd) will be false in js mode, so there's no simple way via testament to selectively disable a block (timotheecour#598 would make it easy though, yet another use case)

so I suggest for now to do the more coarse disabling instead:

discard """
  targets: "c js"
  disabled: "bsd" # pending #17093
"""

[timotheecour]

@beef331 good that it's green for the 1st time :) I finally figured out the mysterious CI failure for freebsd/openbsd, see #17097; TLDR: builds.sr.ht CI doesn't work like the other CI causing issues for github projects.

The fix here (so we could un-disable bsd) would be to rebase this PR against devel, which would pick up the fix for #17076 (shouldn't be needed for other CI because they merge against latest devel on each PR push, but it's needed for builds.sr.ht CI)

I'll go ahead and merge this to avoid too many ping pong and will followup with another PR to do this.

EDIT => #17098