Add at-isdefined

[ararslan]

Note that this is NOT functionally equivalent to @isdefined on 0.7; that isn't possible on earlier versions. Instead this defines @isdefined to wrap isdefined, which can only check for global variables.

[ararslan]

CI failures on macOS are unrelated. Is this something we want or would it be too misleading to have it only support a very limited subset of its 0.7 functionality on 0.6?

[yuyichao]

The subset where this behaves the same as @isdefined can be written as isdefined(@__MODULE__, ...) so I don't think we should have this.

[ararslan]

Okay.

[nalimilan]

I think we should definitely have this because when updating code from 0.6 using isdefined, the most direct replacement is @isdefined, not isdefined(@__MODULE__, ...). Indeed that's what the deprecation suggests. Otherwise, the deprecation message should be changed.

[yuyichao]

No that (edit: the depwarn on 0.7) is unrelated. @isdefined is the closest replacement on 0.7. It is not on 0.6 and adding a definition on 0.6 that does not do the same thing at all does not help.

[nalimilan]

If isdefined(@__MODULE__, ...) is a good replacement, why couldn't we define @isdefined to use it on 0.6?

[yuyichao]

couldn't we define @isdefined to use it on 0.6?

Because it's a good replacement of isdefined. It is not a good implementation (replacement) of @isdefined so such a definition will do completely different things on 0.6 and 0.7.

[yuyichao]

In another word, if you really don't want to type @__MODULE__ defining Compat.@isdefined_in_module to do isdefined(@__MODULE__, ...) is certainly fine. The only thing that's wrong is to call it @isdefined.

Also, defining @isdefined on 0.6 is kind of possible. I don't think it can be done with any performance though (you need to try catch). Such a definition is mostly useless due to the performance issue but if the function of it is really useful for someone (i.e. checking isdefined for local var and not global var) it can be added to Compat and called @isdefined.

[yuyichao]

And yet another way you can call it @isdefined is to make sure it either return the correct result or raise an error on old versions, it's not as good as something that is always correct but at least it won't lead to bugs/much confusion.

In this case, this can be done (conservatively) if you make sure the macro is only used in global scope. I'm not sure the best way to do this but here's one of the very old tricks I've used before.

julia> macro assert_global()
           mod = isdefined(Base, Symbol("@__MODULE__")) ? __module__ : current_module()
           @gensym dummy
           quote
               $(esc(dummy)) = nothing
               if !isdefined($mod, $(QuoteNode(dummy)))
                   error("Not in global scope")
               end
           end
       end
@assert_global (macro with 1 method)

julia> @assert_global

julia> let
       @assert_global
       end
ERROR: Not in global scope
Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] macro expansion at REPL[0]:7 [inlined]
 [3] top-level scope at REPL[2]:2

It's pretty ugly though...........................