hasktorch as a nix package

[wavewave]

On top of #162, I packaged hasktorch 0.0.2 as a nix package, also with libtorch-ffi as a nix package.
For user (not a hasktorch developer), I created user-shell.nix which set up the dependencies for using hasktorch by running

$ nix-shell user-shell.nix

[idontgetoutmuch]

It failed for me :(

Preprocessing test suite 'sysinfo-test' for sysinfo-0.1.1..
Building test suite 'sysinfo-test' for sysinfo-0.1.1..
[1 of 1] Compiling Main             ( test/Spec.hs, dist/build/sysinfo-test/sysinfo-test-tmp/Main.o )
Initialized empty Git repository in /nix/store/jl51p1qmwk94w8wf6jq5pr8s1rq7ipq4-source/third_party/fbgemm/.git/
Linking dist/build/sysinfo-test/sysinfo-test ...
remote: Enumerating objects: 48, done.
remote: Counting objects: 100% (48/48), done.
remote: Compressing objects: 100% (31/31), done.
remote: Total 1401 (delta 19), reused 28 (delta 13), pack-reused 1353
Receiving objects: 100% (1401/1401), 658.64 KiB | 1.44 MiB/s, done.
Resolving deltas: 100% (1008/1008), done.
running tests
From https://github.com/pytorch/fbgemm
 * [new branch]      fb-config          -> origin/fb-config
 * [new branch]      jianyuh-patch-1    -> origin/jianyuh-patch-1
 * [new branch]      jspark1105-patch-1 -> origin/jspark1105-patch-1
 * [new branch]      master             -> origin/master
Running 1 test suites...
Test suite sysinfo-test: RUNNING...
Switched to a new branch 'fetchgit'

sysInfo function
  retuns valid data FAILED [1]
  uptime > 0 FAILED [2]
  freeram > 0 FAILED [3]

Failures:

  test/Spec.hs:24:16:
  1) sysInfo function retuns valid data
       predicate failed on: Left 96

  To rerun use: --match "/sysInfo function/retuns valid data/"

  test/Spec.hs:16:23:
  2) sysInfo function uptime > 0
       uncaught exception: ErrorCall
       Invalid value Left 96
       CallStack (from HasCallStack):
         error, called at test/Spec.hs:16:23 in main:Main

  To rerun use: --match "/sysInfo function/uptime > 0/"

  test/Spec.hs:16:23:
  3) sysInfo function freeram > 0
       uncaught exception: ErrorCall
       Invalid value Left 96
       CallStack (from HasCallStack):
         error, called at test/Spec.hs:16:23 in main:Main

  To rerun use: --match "/sysInfo function/freeram > 0/"

Randomized with seed 426662167

Finished in 0.0010 seconds
3 examples, 3 failures
Test suite sysinfo-test: FAIL
Test suite logged to: dist/test/sysinfo-0.1.1-sysinfo-test.log
0 of 1 test suites (0 of 1 test cases) passed.
builder for '/nix/store/mmhd4mdngkshymnwirvjr0cni7qkmbzl-sysinfo-0.1.1.drv' failed with exit code 1
cannot build derivation '/nix/store/abv36w9dikdzd57fj7wqygnnc1j28f32-ghc-8.6.5-with-packages.drv': 1 dependencies couldn't be built
error: build of '/nix/store/5vw5pmr3n57vchbq6ryq54sf7v57xbfh-mkl-2019.3.199.drv', '/nix/store/abv36w9dikdzd57fj7wqygnnc1j28f32-ghc-8.6.5-with-packages.drv', '/nix/store/vmjmm3588z26m5d4h3h1qb9rgffin94p-python3.7-pytorch-1.2.0.drv' failed

[tscholak]

Did you do that on a Mac? I had the same problem. I had to override sysinfo with an otherwise identical derivation that skips tests, see #165

[idontgetoutmuch]

Yes a Mac. Great! I will take a look.

[junjihashimoto]

@wavewave @tscholak
Thank you for great contribution to make nix package.
There are some PRs for nix-build.
Which should we merge first?

[wavewave]

@junjihashimoto This PR is superseded by @tscholak 's PR and the new PR uses a more modern approach. I am now closing this PR and the nix-shell PR and let us all focus on PR #165

[stites]

@junjihashimoto @wavewave @tscholak.

In the spirit of all-contributors (but since both parties have pushed code), I'd like to bring this PR back, merge it first, then merge #164. This might require a bandaid or two. Both @wavewave and @tscholak have chipped away at the idiosyncrasies of this build.

I'd like to take a stab at this, but need roughly 2-4 days to wrap up home- and work- transitions.

[tscholak]

@stites I couldn't agree more. the work @wavewave did was instrumental in delivering #165, and it should be recognized!