Add caching

[DrChainsaw]

This is the proposed fix for #172

Unfortunately, only caching of the solution seems to have a pretty disappointing impact on the benchmark:

julia> function bench(optimizer, N = 30_000)
           model = Model(with_optimizer(optimizer))
           @variable(model, x[i = 1:N] >= i)
           @objective(model, Min, sum(x))
           optimize!(model)
           @benchmark value.($x)
        end
bench (generic function with 2 methods)

julia> bench(Cbc.Optimizer)
Optimal - objective value 4.50015e+08
Optimal objective 450015000 - 0 iterations time 0.002
BenchmarkTools.Trial: 2 samples with 1 evaluation.
 Range (min … max):  3.524 s …    4.264 s  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     3.894 s               ┊ GC (median):    0.00%        
 Time  (mean ± σ):   3.894 s ± 523.174 ms  ┊ GC (mean ± σ):  0.00% ± 0.00%

  █                                                        █  
  █▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█ ▁
  3.52 s         Histogram: frequency by time         4.26 s <

Profiling shows that Cbc_isProvenOptimal and Cbc_getNumIntegers seems to dominate the computation time.

After adding caching of those two performance appears to be back on track:

julia> bench(Cbc.Optimizer)
Optimal - objective value 4.50015e+08
Optimal objective 450015000 - 0 iterations time 0.002
BenchmarkTools.Trial: 205 samples with 1 evaluation.
 Range (min … max):  19.858 ms … 34.276 ms  ┊ GC (min … max): 0.00% … 13.42%
 Time  (median):     23.973 ms              ┊ GC (median):    0.00%
 Time  (mean ± σ):   24.479 ms ±  2.629 ms  ┊ GC (mean ± σ):  1.76% ±  4.52%

  ▄   █                               ▄  █                    ▁
  █▁▁▁█▁▁▁▁▁▁▁▁▆▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▆▁▁▁█▁▁█▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ █
  20.4 ms         Histogram: frequency by time        23.4 ms <

 Memory estimate: 6.18 MiB, allocs estimate: 330002.

There are probably alot of other Cbc_X functions which are also too slow, but I lack the overall understanding to pinpoint them.

I have tried to make a minimum of changes for easier reviewing of the "core changes", but please suggest better structure and names if you think this is the way to go. I don't mind nitpicking.

[codecov]

Codecov Report

Merging #173 (b9f4ac8) into master (4a80397) will increase coverage by 0.07%.
The diff coverage is 94.11%.

@@            Coverage Diff             @@
##           master     #173      +/-   ##
==========================================
+ Coverage   53.48%   53.55%   +0.07%     
==========================================
  Files           3        3              
  Lines         531      534       +3     
==========================================
+ Hits          284      286       +2     
- Misses        247      248       +1     

Impacted Files	Coverage Δ
src/MOI_wrapper/MOI_wrapper.jl	80.30% <94.11%> (-0.13%)	⬇️

---

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 4a80397...b9f4ac8. Read the comment docs.

[odow]

The fact that it's all the Cbc_ functions makes me think there is a bigger issue going on...

[DrChainsaw]

The fact that it's all the Cbc_ functions makes me think there is a bigger issue going on...

I agree!

[odow]

What's the benchmark with your latest changes?

[odow]

Just a note to say that I haven't forgotten about this PR. But I've been busy with the MOI 0.10 release so I haven't had a chance to take a deeper look yet.

[DrChainsaw]

No problem at all!

I'm not blocked by this, just trying to do my duty as a user and report issues (and fix what I feel is within reach for me to fix). Let me know if there is anything more I can do (the "deeper issue" feels a bit out of reach for me atm).