tensorflow: Bug: reshape shape inference for parital defined shape

---

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):Linux Ubuntu 16.04
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): v1.4.0-rc1-11-g130a514 1.4.0
• Python version: Python 3.6.3 :: Anaconda custom (64-bit)
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version:
• GPU model and memory:
• Exact command to reproduce: see example

Describe the problem

When the input shape for ‘tf.reshape’ is partial defined and the new shape contains a -1 for known dimensions, ‘tf.reshape’ does not predict the shape. See the example code.

Source code / logs

Second and third example have working shape inference:

def foo(*shape):
    x = tf.placeholder(tf.float32, shape)
    return tf.reshape(x, tf.concat([tf.shape(x)[:-2], [-1]], 0))

print(foo(2, 3, 4, 5))  # Tensor("Reshape_8:0", shape=(2, 3, 20), dtype=float32)  # correct
print(foo(None, 3, 4, 5))  # Tensor("Reshape_9:0", shape=(?, 3, ?), dtype=float32)  # shape inference possible
print(foo(None, None, 4, 5))  # Tensor("Reshape_10:0", shape=(?, ?, ?), dtype=float32)  # shape inference possible
print(foo(2, 3, 4, None))  # Tensor("Reshape_11:0", shape=(2, 3, ?), dtype=float32)  # correct

Proof that shape inference is possible:

import functools, operator 
def bar(*shape):
    x = tf.placeholder(tf.float32, shape)
    
    tmp = x.shape[-2:]
    if not tmp == tf.TensorShape(None):
        tmp = functools.reduce(operator.mul, tmp, tf.Dimension(1))
    if str(tmp) == '?' or tmp == tf.TensorShape(None):
        shape = [-1]
    else:
        shape = [tmp]
    
    return tf.reshape(x, tf.concat([tf.shape(x)[:-2], shape], 0))


print(bar(2, 3, 4, 5))  # Tensor("Reshape_21:0", shape=(2, 3, 20), dtype=float32)
print(bar(None, 3, 4, 5))  # Tensor("Reshape_22:0", shape=(?, 3, 20), dtype=float32)
print(bar(None, None, 4, 5))  # Tensor("Reshape_23:0", shape=(?, ?, 20), dtype=float32)
print(bar(2, 3, 4, None))  # Tensor("Reshape_24:0", shape=(2, 3, ?), dtype=float32)