Fix #15, error with latest StaticArrays due to dot change.

[tkoolen]

CC: @goretkin.

Microbenchmark results: on Julia 1.1.1:

julia> @btime transpose(a) * b setup = begin
           a = rand(SVector{2})
           b = SVector(rand(SVector{2}), rand(SVector{2}))
           end
  10.521 ns (0 allocations: 0 bytes)

which at least doesn't allocate, but on latest master:

julia> @btime transpose(a) * b setup = begin
           a = rand(SVector{2})
           b = SVector(rand(SVector{2}), rand(SVector{2}))
           end
  1.762 ns (0 allocations: 0 bytes)

I think the difference is that an assertion error is being 'outlined' automatically in newer versions. I think this is the cleanest solution going forward, but if you consider this to be too much of a performance bug on 1.1, I can change it.

[goretkin]

Thanks! Did you bump the StaticArrays version because this change isn't compatible with older versions?

I was going to suggest first restricting StaticArrays at v0.5.1-v0.10, and then applying a change that makes it compatible with v0.11, but it seems that Pkg does not support version ranges yet: JuliaLang/Pkg.jl#843

[tkoolen]

Good point, I guess bumping the StaticArrays lower bound isn't necessary. Pkg does support version ranges, just not spelled that way. The 'correct' thing to do is to retroactively upper bound the StaticArrays version on previous releases, but if we just release a new version with the fix then just doing a pkg> up will be fine.

I'm going to replace this implementation with a generated function to get the same performance on 1.1 later today. I had trouble getting down to the performance on master without using a generated function.

Also, not sure what's going on on Julia 1.0 Travis builds.

[goretkin]

Pkg does support version ranges, just not spelled that way

Do you mean like: JuliaLang/Pkg.jl#1232 (comment) , or something else?

[codecov-io]

Codecov Report

Merging #16 into master will increase coverage by 0.15%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master      #16      +/-   ##
==========================================
+ Coverage   93.26%   93.42%   +0.15%     
==========================================
  Files           8        8              
  Lines         208      213       +5     
==========================================
+ Hits          194      199       +5     
  Misses         14       14

Impacted Files	Coverage Δ
src/gjk.jl	95.45% <100%> (+0.37%)	⬆️
src/reference_distance.jl	95% <100%> (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 f6a6553...19489ae. Read the comment docs.

[tkoolen]

OK, switched to the generated function approach, which is fast for both current and future versions of Julia.

As for the Travis failures on Julia 1, I can reproduce them locally with Julia 1.0.4. only with code coverage turned on! So I think this is another instance of JuliaLang/julia#30872.

[tkoolen]

@rdeits, would you be OK with allowing failures on 1.0?

[tkoolen]

Went ahead with allowing failures on 1.0.

[tkoolen]

@goretkin, I started the process for tagging a new version: JuliaRegistries/General#1483.