Fix only pass RunOptions to keras will trigger core

[zhuzilin]

This is a PR from JIZHI, the AI platform in Tencent.

Right now, keras will trigger segmentation fault when only pass RunOptions but no RunMetadata. This bug can be replicated in tf-nightly in this gist

If you cannot open the gist, the code to replicate the bug is

import tensorflow as tf
mnist = tf.keras.datasets.mnist

tf.compat.v1.disable_eager_execution()

(x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10, activation='softmax')
])

run_options = tf.compat.v1.RunOptions(trace_level=tf.compat.v1.RunOptions.FULL_TRACE)

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'],
              options=run_options)

model.fit(x_train,
          y_train,
          epochs=2)

The reason for this problem is that keras use the TF_SessionRunCallable api instead of TF_Run. While the former was using an unique_ptr to represent the run_metadata and will only new one when RunMetadata was passed. However, the RunOption will need an RunMetadata for tracing. This PR modified the implementation of TF_SessionRunCallable to that of TF_Run and fix this bug.

Thank you for your time on reviewing this PR.

[zhuzilin]

@omalleyt12 Could you have a look at this PR? Thank you!

[zhuzilin]

@tanzhenyu
Could you have a look at this pr?

[gbaned]

@tanzhenyu Can you please review this PR ? Thanks!

[gbaned]

@tanzhenyu, Can you please review this PR ? Thanks!