Fix #97, decode ordered containers with 'fromDistinctAscList'

[sjakobi]

Performance differences:

Time:

Memory allocations:

50c50
<    1,000,000 Set Int              Encode/Decode: Store      420,580,936     759  OK   
---
>    1,000,000 Set Int              Encode/Decode: Store      424,580,920     766  OK   
55c55
<    2,000,000 Set Int              Encode/Decode: Store      841,194,312   1,518  OK   
---
>    2,000,000 Set Int              Encode/Decode: Store      849,194,296   1,533  OK   
60c60
<   10,000,000 Set Int              Encode/Decode: Store    4,206,073,096   7,592  OK   
---
>   10,000,000 Set Int              Encode/Decode: Store    4,246,073,080   7,666  OK   
65c65
<    1,000,000 IntSet               Encode/Decode: Store      549,981,184   1,006  OK   
---
>    1,000,000 IntSet               Encode/Decode: Store      252,081,248     435  OK   
70c70
<    2,000,000 IntSet               Encode/Decode: Store    1,139,994,240   2,088  OK   
---
>    2,000,000 IntSet               Encode/Decode: Store      504,194,624     871  OK   
75c75
<   10,000,000 IntSet               Encode/Decode: Store    6,123,454,120  11,252  OK   
---
>   10,000,000 IntSet               Encode/Decode: Store    2,521,068,864   4,355  OK   
78c78
<    1,000,000 Map Int Int          Encode: Store             328,000,384     599  OK   
---
>    1,000,000 Map Int Int          Encode: Store             288,000,368     522  OK   
80c80
<    1,000,000 Map Int Int          Encode/Decode: Store    1,864,730,296   3,513  OK   
---
>    1,000,000 Map Int Int          Encode/Decode: Store      536,580,856     968  OK   
83c83
<    2,000,000 Map Int Int          Encode: Store             656,000,384   1,198  OK   
---
>    2,000,000 Map Int Int          Encode: Store             576,000,368   1,044  OK   
85,86c85,86
<    2,000,000 Map Int Int          Encode/Decode: Store    3,873,491,688   7,306  OK   
---
>    2,000,000 Map Int Int          Encode/Decode: Store    1,073,194,232   1,936  OK   
88c88
<   10,000,000 Map Int Int          Encode: Store           3,280,005,024   5,992  OK   
---
>   10,000,000 Map Int Int          Encode: Store           2,880,005,008   5,220  OK   
90c90
<   10,000,000 Map Int Int          Encode/Decode: Store   21,013,631,352  39,721  OK   
---
>   10,000,000 Map Int Int          Encode/Decode: Store    5,366,073,912   9,684  OK   
93c93
<    1,000,000 IntMap Int           Encode: Store             352,000,448     644  OK   
---
>    1,000,000 IntMap Int           Encode: Store             312,000,432     567  OK   
95c95
<    1,000,000 IntMap Int           Encode/Decode: Store      940,404,952   1,742  OK   
---
>    1,000,000 IntMap Int           Encode/Decode: Store      568,581,024   1,029  OK   
98c98
<    2,000,000 IntMap Int           Encode: Store             704,000,448   1,289  OK   
---
>    2,000,000 IntMap Int           Encode: Store             624,000,432   1,134  OK   
100c100
<    2,000,000 IntMap Int           Encode/Decode: Store    1,920,842,928   3,560  OK   
---
>    2,000,000 IntMap Int           Encode/Decode: Store    1,137,194,400   2,058  OK   
103c103
<   10,000,000 IntMap Int           Encode: Store           3,520,004,288   6,447  OK   
---
>   10,000,000 IntMap Int           Encode: Store           3,120,002,384   5,674  OK   
105c105
<   10,000,000 IntMap Int           Encode/Decode: Store   10,254,619,448  19,050  OK   
---
>   10,000,000 IntMap Int           Encode/Decode: Store    5,686,071,536  10,291  OK   

Apparently this version of pokeMap / pokeOrdMap is a bit faster too! It'd be interesting to see if HashMap can benefit too, but I haven't added a benchmark yet.

[sjakobi]

Actually I'm not sure how to explain the speedup seen for Set Int and Set [Char]. I guess these results are noisier than I would have thought.

Would you mind running the benchmarks on your box, @mgsloan?

[sjakobi]

Seems to have worked.

Here's an overview of the various Travis CI environments.

Not sure why the OS X jobs didn't error out too.

[sjakobi]

I've fixed the merge conflicts.

[mgsloan]

LGTM, thanks!