Is edit(::Callable) broken on v0.4?

[Sean1708]

I noticed the following behaviour on the master branch from 7 days ago, but it's also present on the current master.

julia> versioninfo()
Julia Version 0.4.0-dev+2833
Commit 5c5bf9d (2015-01-20 23:10 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin14.0.0)
  CPU: Intel(R) Core(TM) i7-2635QM CPU @ 2.00GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

julia> edit(length)  # doesn't work
ERROR: no method found for the specified argument types
 in which at reflection.jl:178
 in edit at interactiveutil.jl:55

julia> edit(edit)  # :(
ERROR: no method found for the specified argument types
 in which at reflection.jl:178
 in edit at interactiveutil.jl:55

julia> edit(length, (AbstractArray,))  # works if you supply the types

julia> edit(length, (AbstractVector,))  # and the type doesn't have to be exact

On v0.3, however

julia> versioninfo()
Julia Version 0.3.5
Commit a05f87b* (2015-01-08 22:33 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin14.0.0)
  CPU: Intel(R) Core(TM) i7-2635QM CPU @ 2.00GHz
  WORD_SIZE: 64
  BLAS: libopenblas (DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

julia> edit(length)  # works fine

julia> edit(edit)  # also works fine

Running edit(which, (Function, Any...)) takes me to this line suggesting that the issue might be related to the deprecation of which(f, args...).

Possibly also related to #9471 or #9536?

[mbauman]

I noticed this, too. I have the following in my .juliarc.jl, but haven't submitted it since it's not really the right solution here:

Base.functionloc(f::Base.Callable) = (m = methods(f); isempty(m) ? error("no methods for $f found") : functionloc(first(m)))

As I remember it, the issue is that you often want the inverse of jl_gf_invoke_lookup. That is, given a set of types that don't match any methods, you want the first method that is a subtype of the request. jl_gf_invoke_lookup does the opposite - it finds the most-specific signature such that the request is a subtype of the signature.

This is most important for things like call overloads. You want to find all signatures that are a subtype of (T, Any...).

[Sean1708]

Awesome thanks, for most of my use-cases the methods are in the same file so this is good enough for me.

[JeffBezanson]

Yes, we used to use the most specific matching method here. However that is really quite arbitrary. In the version of julia I happen to be looking at now, edit(length) would take you to length(mt::MethodTable) at reflection.jl:133. I don't think that's useful.

However you can get that behavior with edit(methods(length, (Any...))[1]).

We should remove the default value for the type argument to functionloc, and instead make functionloc(f) only work if f has one definition.