@pydef improvements

[cstjean]

I used MacroTools.splitdef to clean up the method definition code. This enables support for long-form function definitions (along with parametric types and return types, which might not get much use). I also now create types with a method dictionary. It will help #470.

(Sorry for the other PR; I had forgotten to branch)

[stevengj]

What are all the 0.6 test failures?

[cstjean]

Fixed.

[stevengj]

indent

[cstjean]

Fixed indentation.

[rdeits]

I think this is causing errors on v0.7 that are being masked by the fact that the 0.7 build was stalling out. Previously, the 0.7 build would hit JuliaLang/julia#26357 but now it errors out earlier: https://travis-ci.org/JuliaPy/PyCall.jl/jobs/357206714#L674

As I understand it, that means that the with_ioraise macro is being called with 0 arguments instead of the 1 it's expecting. It's not clear how that could be happening, but I guess macro parsing inside other macros is different in 0.7?

[cstjean]

Looks like some interaction with JuliaLang/julia#21746 ?

All macros receive an extra argument source::LineNumberNode which describes the parser location in the source file for the @ of the macro call. It can be accessed as a normal argument variable in the body of the macro. This is implemented by inserting an extra leading argument into the Expr(:macrocall, :@name, LineNumberNode(...), args...) surface syntax.

It shouldn't affect us, though. Maybe it's a bug?

This line may have to be changed, but it looks unrelated.

[rdeits]

I don't think it's the new line number argument. The error message is:

ERROR: LoadError: LoadError: LoadError: MethodError: no method matching @with_ioraise(::LineNumberNode, ::Module)
Closest candidates are:
  @with_ioraise(::LineNumberNode, ::Module, ::Any) at /home/travis/.julia/v0.7/PyCall/src/io.jl:22

which shows that the macro was defined to take the LineNumberNode, the Module, and then it's "actual" argument expr::Any, but was called with just the LineNumberNode and Module. If it were an issue of the missing LineNumberNode, then the first argument types wouldn't match.

[Keno]

What's happening is that pydef now generates definitions like this:

    $(Expr(:escape, :(function ##fileno#1531(self; )::Any
      #= /home/keno/.julia/v0.7/MacroTools/src/utils.jl:260 =#
      begin
          @with_ioraise
      end

[Keno]

I'll have a fix shortly.

[Keno]

See FluxML/MacroTools.jl#68.