Stinger Type Interface

[rohitvarkey]

• [x] Julia type mapping to directly decode the Stinger C object.
• [x] Unexported function to provide the user with the StingerGraph instance with C values at that point of time.
• [x] Enumerations of the field types. This gives us the StingerFields type to use for functions to dispatch for these enums. It also allows us to figure out the index and calculate the offset easily.
• [x] getindex to read values of the C object in a lazy manner.
• [x] setindex! to write values to the C object.
• [x] Benchmarks for getindex and setindex
• [x] Insert remove edges benchmark
• [x] outdegree and successors implementations
• [x] BFS Implementation

I couldn't use getfield or setfield! and override the a.b syntax due to https://github.com/JuliaLang/julia/issues/1974.

Using getindex and setindex! along with the defined enumerations allows for usage like

s = Stinger()
s[max_nv] = 1
s[max_nv] #Outputs 1

cc: @jpfairbanks

[rohitvarkey]

The benchmarks for the type stable getfield vs the type instable getfield are shown here:

julia> stableget = read(stable, "gettrial")
BenchmarkTools.Trial:
  samples:          10000000
  evals/sample:     1
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  0.00 bytes
  allocs estimate:  0
  minimum time:     38.00 ns (0.00% GC)
  median time:      41.00 ns (0.00% GC)
  mean time:        43.95 ns (0.00% GC)
  maximum time:     219.11 μs (0.00% GC)

julia> unstableget = read(unstable, "gettrial")
BenchmarkTools.Trial:
  samples:          10000000
  evals/sample:     1
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  32.00 bytes
  allocs estimate:  2
  minimum time:     366.00 ns (0.00% GC)
  median time:      398.00 ns (0.00% GC)
  mean time:        433.29 ns (1.10% GC)
  maximum time:     14.44 ms (99.79% GC)

I think this clearly shows that we should stick with the type stable version with the special functions for batch_time and update_time.

[rohitvarkey]

The insert remove benchmark here gives results of

Insert benchmark (attempting to insert 100000), terminates in 10 seconds.

BenchmarkTools.Trial:
  samples:          84062
  evals/sample:     182
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  16.00 bytes
  allocs estimate:  1
  minimum time:     582.00 ns (0.00% GC)
  median time:      586.00 ns (0.00% GC)
  mean time:        650.31 ns (0.12% GC)
  maximum time:     21.48 μs (0.00% GC)

Remove benchmark (25000)

BenchmarkTools.Trial:
  samples:          25000
  evals/sample:     178
  time tolerance:   5.00%
  memory tolerance: 1.00%
  memory estimate:  16.00 bytes
  allocs estimate:  1
  minimum time:     599.00 ns (0.00% GC)
  median time:      604.00 ns (0.00% GC)
  mean time:        658.24 ns (0.14% GC)
  maximum time:     10.11 μs (91.55% GC)

[rohitvarkey]

The outdegree and successors implementation depends on https://github.com/stingergraph/stinger/pull/204 being merged.