Add a Deque based on a circular buffer

[timholy]

The main attraction of CircularDeque is its performance:

julia> using DataStructures, BenchmarkTools

julia> D = Deque{Int}()
Deque [Int64[]]

julia> C = CircularDeque{Int}(5)
CircularDeque{Int64}([])

julia> function foo!(D)
           for i = 1:5
               push!(D, i)
           end
           v = back(D)
           for i = 1:5
               shift!(D)
           end
           v
       end
foo! (generic function with 1 method)

julia> foo!(D)
5

julia> foo!(C)
5

julia> @benchmark foo!($D)
BenchmarkTools.Trial: 
  samples:          10000
  evals/sample:     870
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  0.00 bytes
  allocs estimate:  0
  minimum time:     153.00 ns (0.00% GC)
  median time:      155.00 ns (0.00% GC)
  mean time:        156.19 ns (0.00% GC)
  maximum time:     265.00 ns (0.00% GC)

julia> @benchmark foo!($C)
BenchmarkTools.Trial: 
  samples:          10000
  evals/sample:     995
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  0.00 bytes
  allocs estimate:  0
  minimum time:     30.00 ns (0.00% GC)
  median time:      33.00 ns (0.00% GC)
  mean time:        33.22 ns (0.00% GC)
  maximum time:     56.00 ns (0.00% GC)

[codecov-io]

Current coverage is 92.21% (diff: 100%)

Merging #215 into master will increase coverage by 0.15%

@@             master       #215   diff @@
==========================================
  Files            27         28     +1   
  Lines          2291       2338    +47   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits           2109       2156    +47   
  Misses          182        182          
  Partials          0          0          

Powered by Codecov. Last update a4a2d41...4ba6e0c

[tkelman]

only a test dependency, isn't it?

useful addition though

[timholy]

With a couple more performance optimizations, I got it down to a median of 20ns, which is almost 8 times faster than the existing Deque.

[rawls238]

Isn't it sufficient to just do this in the runtests.jl file?

[timholy]

It seems to have been set up so that you can also run at least some of the test files independently. For example, if I'm hacking on one particular structure, it's nice to be able to quickly run the tests for just that structure.