Bazel build failure on Ubuntu

[johnnogent]

I have a failure when running
bazel build --config=opt :install
on Ubuntu

[cgarciae]

Same here on Ubuntu 16.10.

[johnnogent]

same on Ubuntu 14.04 LTS

[leesunfreshing]

same here. Ubuntu 16.04 LTS

[normanheckscher]

Builds on local ubuntu 16.04 LTS after applying #6

[leesunfreshing]

@normanheckscher problem solved, thx

[malcolmreynolds]

@normanheckscher thanks for the fix - we'll get it in ASAP, and I'll close this issue once the changes are pushed back out to GitHub.

[malcolmreynolds]

@normanheckscher @leesunfreshing @johnnogent @cgarciae We have pushed a fix, please can you try and build again?

[normanheckscher]

@malcolmreynolds It builds, I just can't import it. I'll try and look a little further into it sometime over the next few days when I get more time.

Python 2.7.12 (default, Nov 19 2016, 06:48:10)
[GCC 5.4.0 20160609] on linux2
Type "help", "copyright", "credits" or "license" for more information.

import sonnet as snt
Traceback (most recent call last):
File "", line 1, in
File "/home/norman/.virtualenvs/tf/local/lib/python2.7/site-packages/sonnet/init.py", line 102, in
from sonnet.python.ops.resampler import resampler
File "/home/norman/.virtualenvs/tf/local/lib/python2.7/site-packages/sonnet/python/ops/resampler.py", line 33, in
tf.resource_loader.get_path_to_datafile("_resampler.so"))
File "/home/norman/.virtualenvs/tf/local/lib/python2.7/site-packages/tensorflow/python/framework/load_library.py", line 64, in load_op_library
None, None, error_msg, error_code)
tensorflow.python.framework.errors_impl.NotFoundError: /home/norman/.virtualenvs/tf/local/lib/python2.7/site-packages/sonnet/python/ops/_resampler.so: undefined symbol: ZN10tensorflow15shape_inference16InferenceContext15WithRankAtLeastENS0_11ShapeHandleEiPS2

[normanheckscher]

Managed to get it to work.
My separately installed tensorflow version (outside the sonnet directory) was originally form the head and a version of 1.1.0rc2. I changed this tensorflow version with git, recompiled and installed the resultant tensorflow-1.1.0rc1-cp27-cp27mu-linux_x86_64.whl. After this, sonnet build, installs and runs as expected.

cd ~/workspace/tensorflow
git checkout r1.1
./configure
bazel build -c opt --copt=-mavx --copt=-mfpmath=both --copt=-msse4.2 --copt=-msse4.1 --config=cuda -k //tensorflow/tools/pip_package:build_pip_package --verbose_failures
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp//tensorflow_pkg/
pip install --upgrade /tmp/tensorflow_pkg/tensorflow-1.1.0rc1-cp27-cp27mu-linux_x86_64.whl
cd ~/workspace
git clone --recursive https://github.com/deepmind/sonnet
etc etc etc from the sonnet instructions

[tmke8]

I have the same error as @normanheckscher ("undefined symbol: ZN10tensorflow...") but my tensorflow is simply 1.0.1 from pip and I'm using Python3. So whatever @normanheckscher thinks was the problem, can't be the problem.

[normanheckscher]

@thomkeh I managed to get it working by being very particular with my tensorflow install. I'm guessing that the issue lies somewhere around interoperability with Tensorflow as I've played with differing versions of TensorFlow in both the sonnet directory and in my workstation environment (using virtualenv).

After building and installing an older version of Tensorflow I have been able to get Sonnet running. I'd suggest using git to build tensorflow @ 36a47f2, as this is the version which Sonnet is linked against. I'm currently using it to bang out multi-cpu models with mnist... variable success (recursive pun intended)... so yeah, it does work.

Is python 3 supported?

[tmke8]

@normanheckscher Ok, I see. Thanks for the response! I makes sense that you should use the tensorflow version that this repository links against.

Python3 is not officially supported but #10 made it sound like compatibility is just some small code changes away.

[adeze]

	sudo apt-get dist-upgrade

fixed it for me on ubuntu 14.04 LTS

i found the bazel error fix elsewhere and it was stated it was an earlier version kernel issue
bazelbuild/bazel#1972

[adria-p]

@thomkeh python 3 is supported now :)

Regarding the compilation issues, marking this as duplicate of #25 to follow progress there.