tensorflow: Tensorflow v2 Variable name uniquification for Keras Layers in eager is inconsistent

Tensorflow v2.0a

When creating e.g. keras models I would assume, that when I run make_generator_model twice in eager mode that the trainable_variable names are identical.

Why would I assume this? Because the tf.train.Checkpoint and Checkpointable api makes you believe that variables are coupled with their corresponding object/class and uniquification of variables would be no longer necessary. And indeed, this is the case when creating a variable with the same name twice (as can be seen at the end of the code)

What do I get instead? In the below example the variables of the second make_generator_model() call will be uniquified.

# First call
['dense/kernel:0', 'batch_normalization_v2/gamma:0', 'batch_normalization_v2/beta:0', 'conv2d_transpose/kernel:0', 'batch_normalization_v2_1/gamma:0', 'batch_normalization_v2_1/beta:0', 'conv2d_transpose_1/kernel:0', 'batch_normalization_v2_2/gamma:0', 'batch_normalization_v2_2/beta:0', 'conv2d_transpose_2/kernel:0']

# Second
['dense_1/kernel:0', 'batch_normalization_v2_3/gamma:0', 'batch_normalization_v2_3/beta:0', 'conv2d_transpose_3/kernel:0', 'batch_normalization_v2_4/gamma:0', 'batch_normalization_v2_4/beta:0', 'conv2d_transpose_4/kernel:0', 'batch_normalization_v2_5/gamma:0', 'batch_normalization_v2_5/beta:0', 'conv2d_transpose_5/kernel:0']

# Third
['dense/kernel:0', 'batch_normalization_v2/gamma:0', 'batch_normalization_v2/beta:0', 'conv2d_transpose/kernel:0', 'batch_normalization_v2_1/gamma:0', 'batch_normalization_v2_1/beta:0', 'conv2d_transpose_1/kernel:0', 'batch_normalization_v2_2/gamma:0', 'batch_normalization_v2_2/beta:0', 'conv2d_transpose_2/kernel:0']

# Fourth
['dense/kernel:0', 'batch_normalization_v2/gamma:0', 'batch_normalization_v2/beta:0', 'conv2d_transpose/kernel:0', 'batch_normalization_v2_1/gamma:0', 'batch_normalization_v2_1/beta:0', 'conv2d_transpose_1/kernel:0', 'batch_normalization_v2_2/gamma:0', 'batch_normalization_v2_2/beta:0', 'conv2d_transpose_2/kernel:0']

# Manual Creation
<tf.Variable 'test:0' shape=() dtype=int32, numpy=1>
<tf.Variable 'test:0' shape=() dtype=int32, numpy=1>

import tensorflow as tf
from tensorflow.keras import layers

def make_generator_model():
    model = tf.keras.Sequential()
    model.add(layers.Dense(7*7*256, use_bias=False, input_shape=(100,)))
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())

    model.add(layers.Reshape((7, 7, 256)))
    assert model.output_shape == (None, 7, 7, 256) # Note: None is the batch size

    model.add(layers.Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))
    assert model.output_shape == (None, 7, 7, 128)
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())

    model.add(layers.Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))
    assert model.output_shape == (None, 14, 14, 64)
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())

    model.add(layers.Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))
    assert model.output_shape == (None, 28, 28, 1)

    return model


m1 = make_generator_model()
noise = tf.random.normal([1, 100])
generated_image = m1(noise, training=False)
print([v.name for v in m1.trainable_variables])

m2 = make_generator_model()
noise = tf.random.normal([1, 100])
generated_image = m2(noise, training=False)
print([v.name for v in m2.trainable_variables])

with tf.Graph().as_default():
    m1 = make_generator_model()
    noise = tf.random.normal([1, 100])
    generated_image = m1(noise, training=False)
    print([v.name for v in m1.trainable_variables])

with tf.Graph().as_default():
    m2 = make_generator_model()
    noise = tf.random.normal([1, 100])
    generated_image = m2(noise, training=False)
    print([v.name for v in m2.trainable_variables])

a = tf.Variable(1, name='test')
b = tf.Variable(1, name='test')
print(a)
print(b)