Deal with normalization collisions in reimport

[palday]

• give more informative error messages
• allow customization of normalization
• fix a few small things while hitting that code

closes #418

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 80.15%. Comparing base (62ae259) to head (52d80f8).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #532      +/-   ##
==========================================
+ Coverage   77.49%   80.15%   +2.65%     
==========================================
  Files          26       26              
  Lines        1684     1693       +9     
==========================================
+ Hits         1305     1357      +52     
+ Misses        379      336      -43     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ararslan]

countmap is such a simple function that I almost feel like it could be more convenient to basically inline it rather than load StatsBase (assuming it isn't already a transitive dependency, which it probably is). Or perhaps separate this chunk out into something like

function normalized_names(members, replacements...)
    occurrences = Dict{Symbol,Int}()
    for member in members
        x = Symbol(replace(member, replacements...))
        occurrences[x] = get(occurrences, x, 0) + 1
    end
    duplicated = [k for (k, v) in occurrences if v > 1]
    if !isempty(duplicated)
        throw(ArgumentError("Names are not unique after normalization: " * join(duplicated, ", "))
    end
    return collect(keys(occurrences))
end

But anyway, doesn't matter to me, do what feels right.

[palday]

It is already a transitive dependency (StatsModels), which is why I decided to just grab it that way. But I like your minor refactor, so I'll update to go that way. 😄

[palday]

HUH, so it turns out that the exports are order sensitive ?! That ordering is lost when we do do collect(keys(...)), so everything breaks.