make kwarg-methods whereis as nothing, rather than as ("none",0)

[oxinabox]

Before:

julia> whereis(@which sum([1]; dims=1))
("none", 0)

julia> definition(String, @which sum([1]; dims=1))
ERROR: SystemError: opening file "none": No such file or directory

After, both of these return nothing.

[KristofferC]

This test is redundant with the one below.

[oxinabox]

It is indeed, but it serves to illustrate what was wrong before.
It can be removed, but I thought it might be helpful in future changes.

[oxinabox]

Do you think it should be removed?

Suggested change

@test loc != ("none", 0) # old incorrect behavour.

[KristofferC]

Suggested change

	If the method could not be found for some reason, returns `nothing`.
	If the method could not be found for some reason, then return `nothing`.

[codecov]

Codecov Report

Merging #48 into master will not change coverage.
The diff coverage is 0%.

@@         Coverage Diff          @@
##           master   #48   +/-   ##
====================================
  Coverage       0%    0%           
====================================
  Files           3     3           
  Lines         165   169    +4     
====================================
- Misses        165   169    +4

Impacted Files	Coverage Δ
src/CodeTracking.jl	0% <0%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3df13a7...4642055. Read the comment docs.

[timholy]

The better approach is to fix the underlying problem. In this case the issue is

julia> m
(::getfield(Base, Symbol("#kw##sum")))(::Any, ::typeof(sum), a::AbstractArray) in Base

julia> m.file
:none

which makes it impractical to find it. However, with a suitably-fixed version of bodymethod (in LoweredCodeUtils), we have

julia> mbody = LoweredCodeUtils.bodymethod(m)
#sum#558(dims, ::typeof(sum), a::AbstractArray) in Base at /home/tim/src/julia-1/base/reducedim.jl:652

and voila we have it.

I'll submit the PRs (LCU, CodeTracking, and Revise) needed to make this work.

I'm not against this change, but I am about to invalidate your test 😄.

[oxinabox]

@timholy great to have this fixed properly.
What happens for the case when Revise isn't loaded?
I'ld like to fail gracefully then. For what ever gracefully means.

[timholy]

I'd call this graceful:

julia> whereis(@which sum([1]; dims=1))
("none", 0)

[oxinabox]

I would not call that graceful, because potentially
"none" could be a valid filename.

Further definition(String, @which sum([1]; dims=1) will still error right?
Because it tries to open that path.

[timholy]

But no actual method has a line number of 0. (Internally, Julia checks this kind of thing fairly frequently.) You can also use isfile to see if "none" really is a file. I'm not completely against returning nothing, but whereis is kind of the more powerful/accurate variant of checking m.file and m.line. And you can always do that and get something back.

The definition example is perhaps more problematic. What should it do besides error? Return nothing?

[oxinabox]

but whereis is kind of the more powerful/accurate variant of checking m.file and m.line.

That sounds like a reasonable argument for the current behavour.
So I am sold.

The definition example is perhaps more problematic. What should it do besides error? Return nothing?

I think so.
Sometimes already definition{Expr, method) returns nothing. (#45)
So it makes sense to me that sometimes definition(String, method) might return nothing.