should definition(Expr, meth) fall back to parsing, when Revise isn't loaded?

[oxinabox]

When Revise isn't loaded
it seems that definition(String, method) works (to my surprise),
but definition(Expr, method) returns nothing.

Seems like we could have a fallback to try parsing the definition(String, method),
if we were going to return nothing.

julia> using CodeTracking

julia> definition(String, @which abs(1))
("abs(x::Signed) = flipsign(x,x)", 136)

julia> definition(Expr, @which abs(1))

julia> Meta.parse(first(definition(String, @which abs(1))))
:(abs(x::Signed) = begin
          #= none:1 =#
          flipsign(x, x)
      end)

[pfitzseb]

Makes sense to me.

[timholy]

This would be a nice addition

[timholy]

A counterpoint is that this is a worse version of definition, as it won't work for @evaled methods. Is it better to have it return the nothing, given that one can then detect this and parse the return from definition(String, ...)?

[timholy]

Though I guess if we do #46, we'd have to return the best we can find.

[oxinabox]

I think i wanted this for Cthulu.
What it currently does is complain that Revise is not loaded.
I guess it would be a small change to first fallback to parsing itself.

https://github.com/JuliaDebug/Cthulhu.jl/blob/f4ffcc4ed919061e7772656695eabcba192cc147/src/codeview.jl#L51-L71

[timholy]

Maybe this would be a good strategy to decide:

• work on Should we rename definition to code_expr and code_string? #46
• document the whole "code_expr is more powerful than code_string if you have Revise loaded, but if not it will return nothing in which case you should parse code_string"
• see if reading our own documentation makes us slap our head and say "that's crazy!"

[timholy]

This exercise seemed to support the current strategy. Close?