Flags enums: prevent All and None keywords

[decioferreira]

Should the compiler prevent the definition of All and None keywords on flag enums?

@[Flags]
enum IOMode
  None # 1
  All  # 2
end

puts IOMode::None.value #=> 1
puts IOMode::All.value  #=> 2

[asterite]

Probably. I don't know if there's a use case where you'd want to overwrite those, and in any case it would yield unexpected results. Thanks for noticing this!

[ervumlens]

All and None values would be pretty nice to have in a flag enum, defined to the appropriate values. And as long as I'm dreaming, it would be good to have the option to specify the names as well. 😄

Borrowing decioferreira's example:

@[Flags(all: :All, none: :Empty)]
enum IOMode
  In # 1
  Out  # 2
end

puts IOMode::Empty.value #=> 0 because the "none" name was set to :Empty
puts IOMode::All.value  #=> 3 because the "all" name was set to :All
puts IOMode::None.value # error: no such constant

For a default setting, though, it makes sense to provide at least a warning if things like All and None are named in the enum itself.

[oprypin]

I think that having All in all enums without opt-out is bad, because it's an API contract. Someone may want to modify their library and add a flag that makes no sense to have under normal circumstances but people are already using All. And being unable to override it (to keep previous behavior) makes it worse.

[RX14]

All Doesn't seem to be used much in the stdlib, I'd be in favour of removing it for the problems it causes when adding new values.

[jhass]

I see the argument OTOH it seems like a handy tool too. Maybe we can come up with some more explicit way of defining it, like All = {@all}, which could even allow All = {@all & ~Foo}?

[bcardiff]

I agree that it would be nice for an opt-out for All/None. And if opted out let the user define/use them (just by plain values as other constants).

All/None have special meaning in Enum#to_s, Enum#each, etc. I wouldn't allow All/None to lose their special semantic.