? behaves strangely as an identifier

[joshkarges]

I have a composite type which has a single Dict property that I thought would be interesting to access with an overloaded ? operator

type J9Type
    propDict::Dict
end
function ?(obj::J9Type,field::Union(Symbol,String))
    return get(obj.propDict,field,Nothing())
end
A = J9Type([:p1=>1,:p2=>2,:p3=>7])

I found that I could only use my overloaded ? operator if I mention it before hand, even with a comment. Otherwise it says that ? is not a generic function.

#?
?(A,:p1)

[JeffBezanson]

It's not supposed to be possible to define methods for ?.

[JeffBezanson]

Also note that ?x is special syntax in the REPL for getting help about x.

[JeffBezanson]

I'm not able to reproduce any comment-related behavior in either master or 0.2. Could you provide some more details about what commit hash you're using (just paste the banner), and exactly how you're running the code --- what is in a file, what is typed at the REPL, etc. ?

[StefanKarpinski]

? isn't an operator in Julia. There is a a ? b : c control flow construct and REPL inputs starting with ? are specially handled as help queries.

[joshkarges]

               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "help()" to list help topics
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.3.0-prerelease+2154 (2014-03-24 13:35 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit 96fb7e1* (3 days old master)
|__/                   |  x86_64-apple-darwin13.1.0

I ran it in both iJulia, and in the terminal julia. It fails in either case if there is no mention of ? before I call it. But it succeeds (though it's not supposed to?) at defining the function and calling it after i mention it with

julia> typeof(?); ?(A,:p1)
juila> methods(?); ?(A,:p1)
etc...

and it succeeds in iJulia if there's even a comment #? in the same cell as the call.

[StefanKarpinski]

Those give me ? not defined which is probably not the right error message here.

[joshkarges]

This is in iJulia

[JeffBezanson]

Well, uses of ? as an identifier really should be errors. Unfortunately I can't fix this due to the @unix? etc. macros, which abuse this feature. #whywecanthavenicethings

Maybe time to fix #4233?
cc @vtjnash

[StefanKarpinski]

There was my suggested non-breaking fix of changing the parsing so that @unix? A : B parses as @unix ? A : B and simultaneously adding a zero-argument @unix macro (and others) that is true or false appropriately at compile time. Is there any issue with that approach?

[JeffBezanson]

I think that will work.

[vtjnash]

No. We use it sometimes in places (like ccall) which don't allow conditionals

[StefanKarpinski]

No. We use it sometimes in places (like ccall) which don't allow conditionals

That can be changed. The current @unix? A : B macros are terrible.

[StefanKarpinski]

There are only two places in Base where that kind of thing is used in a macro:

base/stream.jl
881-
882:dup(x::RawFD) = RawFD(ccall((@windows? :_dup : :dup),Int32,(Int32,),x.fd))
883-dup(src::RawFD,target::RawFD) = systemerror("dup",-1==
884:    ccall((@windows? :_dup2 : :dup2),Int32,
885-    (Int32,Int32),src.fd,target.fd))

Not exactly earth-shattering to work around.

[jiahao]

Ref: #1910

[quinnj]

Closing as dupe of #22025 (even though this came first, the other issue seems to have more recent/productive discussion)