Issues

A bare base name binds to whichever same-named class was parsed last, so a colliding class in another module becomes the superclass

#946

· khatchad opened

on Sep 17, 2026

An inherited method on a class defined inside a function is never dispatched

#945

· khatchad opened

on Sep 17, 2026

A user base class imported from another module resolves only if its module is translated first

#944

· khatchad opened

on Sep 17, 2026

The `tf.shape` summary is a fresh `Tensor` allocation with no generator, so the factory refuses it at seeding and its result reads "not a tensor" (3760 refusals in the test module); a subscript of it inherits that

#943

· khatchad opened

on Sep 17, 2026

A callback passed to `Dataset.reduce`, `Dataset.filter`, `tf.map_fn`, `tf.py_function` or `tf.while_loop` gets no call-graph node, because those summaries never invoke it where `Dataset.map`'s does

#942

· khatchad opened

on Sep 17, 2026

A wildcard source is matched against the parsed scripts by path suffix and the first match wins, so two scripts with one relative path make a wildcard-bound base depend on hash iteration order

#941

· khatchad opened

on Sep 15, 2026

A `tf.reshape` of a dataflow-typed input keeps its unknown-dtype seed member beside the delivered dtype: a same-shape twin from a single reshape

#940

· khatchad opened

on Sep 15, 2026

The unfed-seed restore runs once after the first solve, so a feed whose operand arrives in the second solve leaves a permanent phantom unknown beside its repair

#939

· khatchad opened

on Sep 15, 2026

A layer's `add_weight`-created weight has an empty points-to set where `call` reads it, in every context of the vendored `Conv1d`, so neither the matmul's dtype rule nor its feed can use it

#938

· khatchad opened

on Sep 15, 2026

The same `Conv1d` class yields a float32 output from its first instance and an unknown-dtype output from its second, in two independent layers of the vendored decoder block

#937

· khatchad opened

on Sep 15, 2026

The dtype value reader's fallback chain has less reach than the shape reader's, so one value resolves a shape and floors a dtype

#936

· khatchad opened

on Sep 15, 2026

684 producer delegations in the test module match no manual generator and contribute nothing: the manual registry's coverage, measured

#935

· khatchad opened

on Sep 14, 2026

Navigation Menu

All issues

A bare base name binds to whichever same-named class was parsed last, so a colliding class in another module becomes the superclass

An inherited method on a class defined inside a function is never dispatched

A user base class imported from another module resolves only if its module is translated first

The `tf.shape` summary is a fresh `Tensor` allocation with no generator, so the factory refuses it at seeding and its result reads "not a tensor" (3760 refusals in the test module); a subscript of it inherits that

A callback passed to `Dataset.reduce`, `Dataset.filter`, `tf.map_fn`, `tf.py_function` or `tf.while_loop` gets no call-graph node, because those summaries never invoke it where `Dataset.map`'s does

A wildcard source is matched against the parsed scripts by path suffix and the first match wins, so two scripts with one relative path make a wildcard-bound base depend on hash iteration order

A `tf.reshape` of a dataflow-typed input keeps its unknown-dtype seed member beside the delivered dtype: a same-shape twin from a single reshape

The unfed-seed restore runs once after the first solve, so a feed whose operand arrives in the second solve leaves a permanent phantom unknown beside its repair

A layer's `add_weight`-created weight has an empty points-to set where `call` reads it, in every context of the vendored `Conv1d`, so neither the matmul's dtype rule nor its feed can use it

The same `Conv1d` class yields a float32 output from its first instance and an unknown-dtype output from its second, in two independent layers of the vendored decoder block

The dtype value reader's fallback chain has less reach than the shape reader's, so one value resolves a shape and floors a dtype

684 producer delegations in the test module match no manual generator and contribute nothing: the manual registry's coverage, measured

Search results