For small files, don't recompile anything

[quinnj]

This was a change I had wanted to experiment with and @tk3369 prompted the work by showing how much over-compiling was costing us in #275. On the benchmarks we were running in that issue with this branch, we now see:

julia> perf()
2018-09-19T07:53:00.698 reading 50 x 1 file
  0.110446 seconds (112.06 k allocations: 5.198 MiB)
2018-09-19T07:53:00.928 reading 100 x 1 file
  0.005255 seconds (3.70 k allocations: 166.945 KiB)
2018-09-19T07:53:00.933 reading 150 x 1 file
  0.014252 seconds (5.58 k allocations: 237.680 KiB, 62.17% gc time)
2018-09-19T07:53:00.948 reading 200 x 1 file
  0.002573 seconds (7.96 k allocations: 361.148 KiB)
2018-09-19T07:53:00.95 reading 250 x 1 file
  0.002736 seconds (8.75 k allocations: 399.305 KiB)
2018-09-19T07:53:00.953 reading 300 x 1 file
  0.003196 seconds (10.42 k allocations: 461.695 KiB)
2018-09-19T07:53:00.957 reading 350 x 1 file
  0.003694 seconds (12.13 k allocations: 546.273 KiB)
2018-09-19T07:53:00.961 reading 400 x 1 file
  0.004209 seconds (13.83 k allocations: 612.805 KiB)
2018-09-19T07:53:00.965 reading 450 x 1 file
  0.006593 seconds (15.55 k allocations: 680.398 KiB)
2018-09-19T07:53:00.972 reading 500 x 1 file
  0.005080 seconds (17.29 k allocations: 747.898 KiB)
2018-09-19T07:53:00.978 reading 550 x 1 file
  0.005779 seconds (19.21 k allocations: 812.602 KiB)
2018-09-19T07:53:00.984 reading 600 x 1 file
  0.006613 seconds (23.38 k allocations: 1.069 MiB)
2018-09-19T07:53:00.99 reading 650 x 1 file
  0.006636 seconds (22.89 k allocations: 1.062 MiB)
2018-09-19T07:53:00.997 reading 700 x 1 file
  0.007651 seconds (24.83 k allocations: 1.130 MiB)
2018-09-19T07:53:01.005 reading 750 x 1 file
  0.007381 seconds (26.75 k allocations: 1.197 MiB)
2018-09-19T07:53:01.013 reading 800 x 1 file
  0.008498 seconds (28.70 k allocations: 1.266 MiB)
2018-09-19T07:53:01.022 reading 850 x 1 file
  0.009268 seconds (30.65 k allocations: 1.335 MiB)
2018-09-19T07:53:01.031 reading 900 x 1 file
  0.009300 seconds (32.60 k allocations: 1.404 MiB)
2018-09-19T07:53:01.041 reading 950 x 1 file
  0.012026 seconds (34.55 k allocations: 1.472 MiB)
2018-09-19T07:53:01.053 reading 1000 x 1 file
  0.009923 seconds (36.49 k allocations: 1.541 MiB)

[quinnj]

We probably still need to experiment with the right threshold of when we switch from no-recompiling to specialized kernels for larger files.

[nalimilan]

That sounds really nice, but it's impossible to spot what changes in the middle of moving code. Maybe make a first commit just to move code around?

[codecov]

Codecov Report

Merging #284 into master will increase coverage by 0.15%.
The diff coverage is 97.56%.

@@            Coverage Diff             @@
##           master     #284      +/-   ##
==========================================
+ Coverage   94.56%   94.72%   +0.15%     
==========================================
  Files           7        7              
  Lines         607      625      +18     
==========================================
+ Hits          574      592      +18     
  Misses         33       33

Impacted Files	Coverage Δ
src/typedetection.jl	97.75% <100%> (-0.03%)	⬇️
src/tables.jl	96.82% <100%> (+1.47%)	⬆️
src/CSV.jl	100% <100%> (ø)	⬆️
src/filedetection.jl	86.76% <87.5%> (ø)	⬆️
src/deprecated.jl	95.39% <0%> (-0.03%)	⬇️
src/validate.jl	94.44% <0%> (+0.32%)	⬆️

---

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 eb02e13...5f6f0e9. Read the comment docs.

[quinnj]

Good idea @nalimilan, I pushed a commit to master that just did the code movement, and rebased this PR so the diff is much cleaner.