th cannot import packages

[caseus-viridis]

Same problem as in Issue #33

th> require 'nn'
/Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: /Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: loop or previous error loading module 'torch'
stack traceback:
    [C]: in function 'error'
    /Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: in function 'require'
    [string "_RESULT={require 'nn'}"]:1: in main chunk
    [C]: in function 'xpcall'
    /Users/___/torch/install/share/lua/5.1/trepl/init.lua:630: in function 'repl'
    .../___/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:187: in main chunk
    [C]: at 0x01020a11b0    

luajit -l xlua gives:

luajit: error loading module 'libsys' from file '/Users/___/torch/install/lib/lua/5.1/libsys.so':
    dlopen(/Users/___/torch/install/lib/lua/5.1/libsys.so, 6): Library not loaded: libmkl_intel_lp64.dylib
  Referenced from: /Users/___/torch/install/lib/lua/5.1/libsys.so
  Reason: image not found
stack traceback:
    [C]: at 0x010a86f960
    [C]: in function 'require'
    /Users/___/torch/install/share/lua/5.1/sys/init.lua:14: in main chunk
    [C]: in function 'require'
    /Users/___/torch/install/share/lua/5.1/xlua/init.lua:37: in main chunk
    [C]: at 0x010a8702c0
    [C]: at 0x010a7f71b0

Running on MAC OSX 10.11

[soumith]

Library not loaded: libmkl_intel_lp64.dylib

That should help you fix the issue? Intel MKL seems to not be in your LD_LIBRARY_PATH / DYLD_LIBRARY_PATH

[caseus-viridis]

Thanks soumith.
Adding Intel MKL path makes loading torch packages in luajit behaving normally.

luajit
LuaJIT 2.1.0-alpha -- Copyright (C) 2005-2015 Mike Pall. http://luajit.org/

 _____              _     
|_   _|            | |    
  | | ___  _ __ ___| |__  
  | |/ _ \| '__/ __| '_ \ 
  | | (_) | | | (__| | | |
  \_/\___/|_|  \___|_| |_|

JIT: ON SSE2 SSE3 SSE4.1 fold cse dce fwd dse narrow loop abc sink fuse
th> require 'nn'
th> 

But the same problem persists when attempting to load packages in trepl.

th

  ______             __   |  Torch7                                         
 /_  __/__  ________/ /   |  Scientific computing for Lua. 
  / / / _ \/ __/ __/ _ \  |  Type ? for help                                
 /_/  \___/_/  \__/_//_/  |  https://github.com/torch         
                          |  http://torch.ch                  

th> 
th> 
th> require 'nn'
/Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: /Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: loop or previous error loading module 'torch'
stack traceback:
    [C]: in function 'error'
    /Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: in function 'require'
    [string "_RESULT={require 'nn'}"]:1: in main chunk
    [C]: in function 'xpcall'
    /Users/___/torch/install/share/lua/5.1/trepl/init.lua:630: in function 'repl'
    .../___/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:187: in main chunk
    [C]: at 0x01065501b0

Ideas?

[soumith]

@caseus-viridis that is very strange. How about in luajit, you do:
require('trepl')()

That will activate trepl in luajit

[caseus-viridis]

Invoking trepl in luajit works fine. Using it as a workaround for now.
Would be nice to solve the prob tho. Thanks @soumith

[caseus-viridis]

One more test: invoking trepl in luajit works fine only after MKL path was manually added to LD_LIBRARY_PATH, DYLD_LIBRARY_PATH and DYLD_FALLBACK_LIBRARY_PATH.
Another test: without adding MKL path manually, and after a fresh install, I had the following:

luajit -l torch
luajit: /Users/___/torch/install/share/lua/5.1/torch/init.lua:14: cannot load '/Users/___/torch/install/lib/lua/5.1/libtorch.so'
stack traceback:
    [C]: in function 'require'
    /Users/___/torch/install/share/lua/5.1/torch/init.lua:14: in main chunk
    [C]: at 0x010ea5e2c0
    [C]: at 0x010e9e52e0

and

luajit
LuaJIT 2.1.0-alpha -- Copyright (C) 2005-2015 Mike Pall. http://luajit.org/

 _____              _     
|_   _|            | |    
  | | ___  _ __ ___| |__  
  | |/ _ \| '__/ __| '_ \ 
  | | (_) | | | (__| | | |
  \_/\___/|_|  \___|_| |_|

JIT: ON SSE2 SSE3 SSE4.1 fold cse dce fwd dse narrow loop abc sink fuse
th> require 'torch'
/Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: loop or previous error loading module 'torch'
stack traceback:
    [C]: in function 'error'
    /Users/___/torch/install/share/lua/5.1/trepl/init.lua:363: in function 'require'
    [string "_RESULT={require 'torch'}"]:1: in main chunk
    [C]: in function 'xpcall'
    /Users/___/torch/install/share/lua/5.1/trepl/init.lua:630: in function </Users/___/torch/install/share/lua/5.1/trepl/init.lua:539>
    stdin:1: in main chunk
    [C]: at 0x010fc362e0

Hope this helps pinpoint the problem.

[gcr]

I'm affected by this too. Digging deeper...

@caseus-viridis , do you use Anaconda? If so, do you have gcc installed into your Conda environment?

On my system, libmkl_intel_lp64.dylib only exists in the Anaconda library folder. I note that install.sh has some special logic for removing Anaconda from the build path... but I also note that whenever luarocks installs a package, it builds with ~/miniconda/bin/cc, which always generates libraries that depend on libmkl_intel_lp64.dylib.

[caseus-viridis]

I don't use conda. I use enthought.

[gcr]

Hm. Does enthought install mkl? Where is the mkl library on your machine?

As a workaround, deleting ./pkg/torch/lib/TH/cmake/FindMKL.cmake and rebuilding torch from scratch with ./clean.sh; ./install.sh completely fixes the issue. (What cmake can't find won't hurt it, right?)

But not sure why the library fails to load. Are there any people who use torch with MKL without any issues?

[caseus-viridis]

Yes @gcr , MKL is with my version of enthought (v2.7.9), located /Library/Frameworks/EPD64.framework/Versions/7.2/lib on my mac.
Linking to it manually after torch install now as a workaround.

[caseus-viridis]

@gcr , I verified your workaround on my mac with enthought MKL.

[admercs]

This solution of @gcr worked for me as well. Anaconda really messes with things on OS X.