Decide whether `@...` is intended to work

[Keno]

@. is a special parser form that lowers to @__dot__:

julia> :(@.)
:(#= REPL[4]:1 =# @__dot__)

julia> macro __dot__() 1 end
@__dot__ (macro with 1 method)

julia> @.
1

The parser also supports @.. because .. is an operator, which is allowed as a macro name:

julia> :(@+)
:(#= REPL[7]:1 =# @+)

julia> macro ..() 2 end
@.. (macro with 2 methods)

julia> @..
2

But what about ...? It parses in macro call context:

julia> :(@...)
:(#= REPL[13]:1 =# @var"...")

but defining it is an error:

julia> macro ...() 3 end
ERROR: ParseError:
# Error @ REPL[12]:1:7
macro ...() 3 end
#     └─┘ ── invalid identifier
Stacktrace:
 [1] top-level scope
   @ REPL:1

unless you use var:

julia> macro var"..."(args...) 3 end
@... (macro with 1 method)

julia> @...
3

Four dots and up parses as a macro call to @..., with normal dot parsing for the remaining arguments, but since it's mostly illegal, it's a bit confusing:

julia> (@....)
ERROR: ParseError:
# Error @ REPL[33]:1:7
(@....)
#     ╙ ── unexpected `)`
Stacktrace:
 [1] top-level scope
   @ REPL:1

julia> (@.....)
3

julia> (@......)
ERROR: ParseError:
# Error @ REPL[35]:1:6
(@......)
#    └─┘ ── invalid identifier
Stacktrace:
 [1] top-level scope
   @ REPL:1

julia> (@.......)
ERROR: ParseError:
# Error @ REPL[36]:1:6
(@.......)
#    └─┘ ── invalid identifier
Stacktrace:
 [1] top-level scope
   @ REPL:1

My preference would be to make @... an illegal identifier parse (and thus by implication also the four-dots-and up variants). I don't think this was ever intended to work and is just an artifact of how the parser treats ... currently.

[Keno]

Relatedly, I think we should consider deprecating the import A... syntax for importing the .. operator from A, requiring import A.:(..) like everywhere else.

[Keno]

Also, we should decide whether @$x is really supposed to parse as @$ x. @A.$ b is disallowed of course. To me @$x really reads as trying to interpolate a macro name - I think we should make this an error.

[LilithHafner]

... is not a valid identifier in any case (except for macro call case) and Triage approves making ... also not a valid identifier in the macro call case. It's okay to deviate from the flisp parser in this case.

@$x can go when we have syntax evolution and remove $ as an identifier altogether.