Issues

• A string constant a summary declares, such as tf.keras.losses.Reduction.NONE, holds nothing, so a loss built with it loses its shape

  wala/ML#1169 · Filter by author khatchad opened on Oct 27, 2026

• The MatMul type feed composes its operands without reading the transpose and adjoint flags, so a flagged call that reaches it gets a wrong shape

  wala/ML#1164 · Filter by author khatchad opened on Oct 26, 2026

• A function that Dataset.map applies to a dataset of pairs gets a type for its first parameter only

  wala/ML#1162 · Filter by author khatchad opened on Oct 26, 2026

• Python's // reaches the IR as the same operator as /, and **, in and not in as the same operator as +

  wala/ML#1156 · Filter by author khatchad opened on Oct 26, 2026

• TensorFlow Datasets' tfds.load has no summary, so the elements of a dataset it loads reach a training step untyped

  wala/ML#1155 · Filter by author khatchad opened on Oct 25, 2026

• The compile summary stores a string optimizer such as "sgd" as the string, so self.optimizer.apply_gradients reaches nothing

  wala/ML#1154 · Filter by author khatchad opened on Oct 24, 2026

• A Layer method read off a built-in layer instance, such as dense.add_weight(...), reaches no callee

  wala/ML#1150 · Filter by author khatchad opened on Oct 24, 2026

• A tuple that a map function builds from its component parameters reads with another component's shape in one part and no type in another

  wala/ML#1142 · Filter by author khatchad opened on Oct 24, 2026

• A value read of a Keras step parameter in call sees the whole elements of the datasets fit was given, not the component the step selected

  wala/ML#1141 · Filter by author khatchad opened on Oct 24, 2026

• The statement import a.b binds b rather than a, so a later a.b.f(x) has no root

  wala/ML#1140 · Filter by author khatchad opened on Oct 24, 2026

• A binary operation whose only tensor operand is typed by a sidecar annotation gives no tensor

  wala/ML#1139 · Filter by author khatchad opened on Oct 24, 2026

• A structured component of a Keras step's data is typed as one tensor carrying its members' shapes and dtypes, not as a tuple

  wala/ML#1138 · Filter by author khatchad opened on Oct 23, 2026

• The @ operator is translated as an elementwise multiply, so a matrix product of tensors loses its shape

  wala/ML#1137 · Filter by author khatchad opened on Oct 23, 2026

• A tensor read back out of a list held inside another container is not seeded, so for g, v in zip(grads, variables) leaves g untyped

  wala/ML#1136 · Filter by author khatchad opened on Oct 23, 2026

• Iterating a dict binds the loop variable to the dict's values rather than its keys

  wala/ML#1135 · Filter by author khatchad opened on Oct 23, 2026

• A slice store lowers to the same slice builtin call as a slice read, and an augmented one loses its operator

  wala/ML#1130 · Filter by author khatchad opened on Oct 20, 2026

• The builtins tuple, min, max, any and all have no summary, so a call to one has no call-graph target

  wala/ML#1129 · Filter by author khatchad opened on Oct 20, 2026

• The Cut release workflow takes no commit to release, so a merge during the run rejects its push

  buginfrastructure
  wala/ML#1126 · Filter by author khatchad opened on Oct 19, 2026

• A Dense layer without a bias is read as having one, so a model's weights gain a bias shape it does not have

  wala/ML#1125 · Filter by author khatchad opened on Oct 17, 2026

• A layer call's deferred update is modeled without its conditions, so a normalization layer in inference writes one and a stateful recurrent layer does not

  wala/ML#1124 · Filter by author khatchad opened on Oct 17, 2026

• A functional model's convolution weights are read as a dense layer's, so a convolution kernel has a rank-2 shape

  wala/ML#1123 · Filter by author khatchad opened on Oct 17, 2026

• A branch on tf.executing_eagerly() in a function that can only run eagerly is read as taking either path

  wala/ML#1122 · Filter by author khatchad opened on Oct 17, 2026

• A Sequential model's call passes its layers no training value, and a summary cannot pass it by keyword

  wala/ML#1120 · Filter by author khatchad opened on Oct 17, 2026

• Applying a BatchNormalization layer records no deferred state update, and its summaries take no training

  wala/ML#1119 · Filter by author khatchad opened on Oct 17, 2026
  1

• A numpy.random draw writes no generator state, so a function that draws shows no side effect

  wala/ML#1118 · Filter by author khatchad opened on Oct 17, 2026