Build error: 'val.val' may be used uninitialized

akirademoss commented 4 years ago

System information

OS Platform and Distribution (e.g., Linux Ubuntu 18.04):
TensorFlow Serving installed from source:
TensorFlow Serving version 2.3.0:

Problem

Getting several error messages when trying to build. 1.) error: 'val.val' may be used uninitialized in this function [-Werror=maybe-uninitialized] return val;

2.) AttributeError: module 'keras_preprocessing.image' has no attribute 'DataFrameIterator'

Exact Steps to Reproduce

Download the 2.3.0 tars, create a local virtual environment e.g:

mkvirtualenv tf_serving
workon tf_serving

pip3 install -U six numpy wheel mock
pip3 install -U keras_applications==1.0.5 --no-deps
pip3 install -U keras_preprocessing==1.0.3 --no-deps
pip3 install numpy==1.19.4

bazel build --config=nativeopt tensorflow_serving/...

Potential steps to solve

uninstalled keras_applications and keras_preprocessing. Instead ran:

pip3 install -U keras==2.3.1

note that this should take care of the second error.

akirademoss commented 4 years ago

doing the above and using the below build command built properly

bazel build --config=release --copt=-O2 tensorflow_serving/...

akirademoss commented 4 years ago

Note for updating installation instructions later: to build bazel-bin/tensorflow_serving/tools/pip_package/build_pip_package /tmp/pip

then copy to serving directory, workon virtual environment, and installstall same way installed tensorflow with pip3.

akirademoss / serving