search

arrayfire / arrayfire-haskell

Haskell bindings to ArrayFire
http://hackage.haskell.org/package/arrayfire
BSD 3-Clause "New" or "Revised" License
59 stars 5 forks source link

Add .travis.yml and test target #12

Open dmjio opened 4 years ago

chessai commented 4 years ago

Why not use haskell-ci (which is really good), using travis for ubuntu? Unless you can have multiple travis setups for nix and non-nix builds. We could potentially test-drive typhon here for nix

dmjio commented 4 years ago

@chessai I'll check out haskell-ci tonight, bigger issue is why cabal on ubuntu doesn't respect extra-lib-dirs and extra-include-dirs in the cabal file after cabal configure

noughtmare commented 4 years ago

@chessai try passing the -v3 flag to cabal.

chessai commented 4 years ago

why is a travis build not triggering on the most recent commits?

noughtmare commented 4 years ago

@chessai It is working again, but you added -v3 in the wrong place.

noughtmare commented 4 years ago

The log shows:

/usr/bin/gcc returned ExitFailure 1 with error message:
gcc: error: /opt/arrayfire/lib: No such file or directory

I think this is because of the ld-options field in the cabal file. I think it is best if the entire field is removed completely.

chessai commented 4 years ago

okay, configure seems to succeed now. however, hspec-discover fails.

noughtmare commented 4 years ago

Yes, I fixed that error on my machine by running cabal install hspec-discover.

chessai commented 4 years ago

most recent failure seems spurious (just adding hvr-ppa)

chessai commented 4 years ago

changed cabal to ${CABAL}. typo

chessai commented 4 years ago

use $WITHCOMPILER

chessai commented 4 years ago

sorry, i keep forgetting that git with do weird things with '$'-prefixed strings.

noughtmare commented 4 years ago

Now we're seeing real errors :tada:

chessai commented 4 years ago

okay, the library is building, but running anything is failing with:

/tmp/dist-test.N8AS/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.1.0.0/build/libHSarrayfire-0.1.0.0-inplace.a(LAPACK.o)(.text+0x71f): error: undefined reference to 'af_pinverse'
/tmp/dist-test.N8AS/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.1.0.0/build/libHSarrayfire-0.1.0.0-inplace.a(Image.o)(.text+0x48a5): error: undefined reference to 'af_iterative_deconv'
/tmp/dist-test.N8AS/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.1.0.0/build/libHSarrayfire-0.1.0.0-inplace.a(Image.o)(.text+0x4be4): error: undefined reference to 'af_inverse_deconv'
/tmp/dist-test.N8AS/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.1.0.0/build/libHSarrayfire-0.1.0.0-inplace.a(LAPACK.o)(.text+0x1000): error: undefined reference to 'af_pinverse'
/tmp/dist-test.N8AS/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.1.0.0/build/libHSarrayfire-0.1.0.0-inplace.a(Signal.o)(.text+0x4aa): error: undefined reference to 'af_approx1_uniform'
/tmp/dist-test.N8AS/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.1.0.0/build/libHSarrayfire-0.1.0.0-inplace.a(Signal.o)(.text+0xa6f): error: undefined reference to 'af_approx2_uniform'
noughtmare commented 4 years ago

That was fixed in #17

chessai commented 4 years ago

ok, rebasing

chessai commented 4 years ago

once a build with 8.8.1 passes, i will re-add the older ghcs

noughtmare commented 4 years ago

It seems that #23 forgot to also remove the tests for the Ord instance.

chessai commented 4 years ago 
/tmp/dist-test.Iv7D/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.3.0.0/t/test/build/test/test: error while loading shared libraries: libaf.so.3: cannot open shared object file: No such file or directory
Test suite test: FAIL 
Test suite logged to: /tmp/dist-test.Iv7D/dist-newstyle/build/x86_64-linux/ghc-8.8.1/arrayfire-0.3.0.0/t/test/test/arrayfire-0.3.0.0-test.log 
0 of 1 test suites (0 of 1 test cases) passed. 
-----BEGIN CABAL OUTPUT-----
cabal: Tests failed for test:test from arrayfire-0.3.0.0.
-----END CABAL OUTPUT-----

looks like this isnt getting copied in properly

noughtmare commented 4 years ago

It is strange. There are two builds, pr and push, and they give different outputs:

chessai commented 4 years ago

hmm. that is strange.

chessai commented 4 years ago

do i need to remove the Ord tests from the test suite?

noughtmare commented 4 years ago

Yes, I think so.

chessai commented 4 years ago

now we at least see the same error on both.

noughtmare commented 4 years ago

Oh, I think I know how to fix this. The linux documentation of arrayfire mentions that you have to do this:

Given sudo permissions, you can add the ArrayFire libraries via ldconfig like so:

echo /opt/arrayfire/lib64 > /etc/ld.so.conf.d/arrayfire.conf
sudo ldconfig

Otherwise, you will need to set the LD_LIBRARY_PATH environment variable in order to let your shared library loader find the ArrayFire libraries.

noughtmare commented 4 years ago

I think that is also the reason for the ld-options field in the cabal file.

So another alternative we have is putting:

ld-options: -Wl,-rpath /opt/arrayfire/lib64

back in the cabal file.

(not -Wl,-rpath /opt/arrayfire/lib)

chessai commented 4 years ago

we could do that. would it be better instead to do export LD_LIBRARY_PATH="/opt/arrayfire/lib64:$LD_LIBRARY_PATH"?

noughtmare commented 4 years ago

Yes, that would also work.

noughtmare commented 4 years ago

I don't know if it carries over from before-install to script, but we will see.

chessai commented 4 years ago

me either. i should probably just do it right after the arrayfire script finishes.

chessai commented 4 years ago

failure on push is spurious.

chessai commented 4 years ago

actually, those have to be persisting, because a lot of them are actually used during install. so i'm less worried about that. for example, if the extension to PATH to include $CABALHOME/bin didn't persist, the hspec-discover failure would still be occurring.

chessai commented 4 years ago

extending LD_LIBRARY_PATH worked.

Test suite now fails with:

Failures:
  test/ArrayFire/ArithSpec.hs:31:7: 
  1) ArrayFire.Arith, Arith tests, Should take cubed root
       expected: ArrayFire Array
                 [1 1 1 1]
                     3.0000 

        but got: ArrayFire Array
                 [1 1 1 1]
                     3.0000 

  To rerun use: --match "/ArrayFire.Arith/Arith tests/Should take cubed root/"
Randomized with seed 938744424

Which i think is a problem with floating point arithemtic

noughtmare commented 4 years ago 
test/ArrayFire/ArithSpec.hs:31:7: 
1) ArrayFire.Arith, Arith tests, Should take cubed root
     expected: ArrayFire Array
               [1 1 1 1]
                   3.0000 

      but got: ArrayFire Array
               [1 1 1 1]
                   3.0000

:thinking:

chessai commented 4 years ago

perhaps we should add a shouldBeEps :: Array Double -> ArrayFire Double -> Expectation that compares for equality within some epsilon.

chessai commented 4 years ago

push failure is spurious

chessai commented 4 years ago

nice. it built and passed.

chessai commented 4 years ago

added ghc 8.6.5 and 8.4.4.

chessai commented 4 years ago

Configure fails on older ghcs/cabal 2.4. Havent looked into it.

chessai commented 4 years ago

i am going to get arrayfire locally so i can test without travis. i don't know why configure is failing on older ghcs, but not 881 with cabal 3.0

noughtmare commented 4 years ago

lapack spec: failing with "free(): invalid next size (normal)"

I get a similar error message in https://github.com/arrayfire/arrayfire-haskell/issues/13#issuecomment-549938655.

chessai commented 4 years ago

just trying to appease travis right now, because locally everything is fine.

chessai commented 4 years ago

OK, we now have CI for GHC 8.4.4, 8.6.5, and 8.8.1. @noughtmare @dmjio please review when you have time.

chessai commented 4 years ago

have to add back LAPACK/doctest tests