tensorflow: tf.layers uses wrong variable scope

TF 1.4.

import tensorflow as tf
def f(x):
    return tf.layers.conv2d(x, 30, 3)

x = tf.zeros([3, 20, 20, 1])

with tf.variable_scope('a'):
    print(f(x))
with tf.variable_scope('a', reuse=True):
    print(f(x)) # works

print(f(x))
with tf.variable_scope(tf.get_variable_scope(), reuse=True):
    print(f(x)) # failed with:
"""
ValueError: Variable conv2d_1/kernel does not exist, or was not created with tf.get_variable(). Did you mean to set reuse=None in VarScope?
"""

From what I can see, tf.layers is trying to create the variables under a wrong variable scope name, making variable sharing impossible if used under the root scope.

I found that this works:

with tf.variable_scope(tf.get_variable_scope()):
    print(f(x))
with tf.variable_scope(tf.get_variable_scope(), reuse=True):
    print(f(x)) # works

But the first line seems redundant and counter-intuitive.

//UPDATE: It also works if I don’t use tf.layers:

def f(x):
    W = tf.get_variable('w', shape=[1])
    return x + W