Repeated names

[mcabbott]

Maybe it's time to define & test the behaviour with repeated indices. This fixes:

• sum(A, dims=:i) to use the first dimension named :i, as it already did, and
• A[i=1] is an error if more than one dimension is named :i. Previously it fixed all such dimensions.

The one exception I can think of is that, for a Diagonal matrix, it may make sense to index both dimensions together. But maybe I have overlooked other uses?

The indexing change did not break any tests. But may still count as a breaking change, not sure.

[codecov]

Codecov Report

Merging #148 (1d3f4c7) into master (1ef0b73) will increase coverage by 0.04%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #148      +/-   ##
==========================================
+ Coverage   93.71%   93.76%   +0.04%     
==========================================
  Files          10       10              
  Lines         382      385       +3     
==========================================
+ Hits          358      361       +3     
  Misses         24       24              

Impacted Files	Coverage Δ
src/name_core.jl	95.55% <100.00%> (+0.10%)	⬆️

---

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 1ef0b73...1d3f4c7. Read the comment docs.

[oxinabox]

I know @nickrobinson251 likes for covarience matrixes that indexing with just the keyword gives you on the diagonal.
Not enough to use undefined behavour, but he thinks it is cute