1 2When building libraries (such as custom op libraries) agains the TensorFlow pip 3package, care must be taken to ensure those libraries build against that 4package's headers and with the same compiler and linker flags as that package 5was compiled with. These utilities help ensure that's the case. 6 7First, add the following to your `WORKSPACE` file to configure a repository that 8provides the C++ headers and libraries provided by the TensorFlow pip package. 9 10``` 11load("//fcp/tensorflow/system_provided_tf:system_provided_tf.bzl", "system_provided_tf") 12system_provided_tf(name = "system_provided_tf") 13``` 14 15Then simply load `tf_custom_op_library` from 16`@system_provided_tf//:system_provided_tf.bzl` instead of 17`@org_tensorflow//tensorflow:tensorflow.bzl`. 18 19NOTE: The `gpu_srcs` and `gpu_deps` parameters supported by TensorFlow's version 20of `tf_custom_op_library` are not supported by this version. 21