Remove a convert method

[davidanthoff]

Fixes #51.

@JeffBezanson I guess this would be the PR, but it fails tests with the method ambiguity error. Opening a PR so that we can see what happens on julia nightly.

[JeffBezanson]

Ok, it looks like a recent change to type intersection changed this, but I'm not totally confident it won't change again, so let's just keep the method for now.

[codecov-io]

Codecov Report

Merging #52 into master will decrease coverage by 0.04%.
The diff coverage is n/a.

@@            Coverage Diff            @@
##           master     #52      +/-   ##
=========================================
- Coverage   74.95%   74.9%   -0.05%     
=========================================
  Files          13      13              
  Lines         519     518       -1     
=========================================
- Hits          389     388       -1     
  Misses        130     130

Impacted Files	Coverage Δ
src/scalar/core.jl	92.59% <ø> (-0.06%)	⬇️

---

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 7f32c4d...6c55dfb. Read the comment docs.

[davidanthoff]

I think removing the method seems to work on both julia 1.1 and 1.2, so maybe that is the right way to do it for now?

[tkf]

Would it be crazy to dynamically determine if you need disambiguation at precompile-time? Something like this:

try
    convert(Any, NA)
    true
catch err
    err isa MethodError || rethrow()
    false
end || begin
    Base.convert(::Type{Any}, ::DataValue{Union{}}) = NA
end

[davidanthoff]

That seems very involved :) I almost feel having the static version check there is simpler?

[tkf]

I thought it's nice to be able to define it only using stable public API; i.e., it does/will not have to record which past/future Julia versions need this definition.

[davidanthoff]

@JeffBezanson, do you have an opinion about that kind of approach?