Issues

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

The allocation-site extractor declines for a constant key and throws for every other key with no allocation behind it: a definition disagreeing with itself

#933

· khatchad opened

on Sep 14, 2026

The flow-sensitive integer chase declines silently at seventeen points, so a null cannot tell "the program does not say" from "the chase did not reach it"

#932

· khatchad opened

on Sep 13, 2026

A loop-carried producer read directly, with no analysis run, returns a shape set that excludes the program's value

#931

· khatchad opened

on Sep 13, 2026

Two producer self-recursion guards return ⊥ on the null-engine path where the resolver's policy is ⊤

#928

· khatchad opened

on Sep 13, 2026

Any allocation of the array type made outside the array summaries silently loses every method call: the array's methods are per-allocation fields, not class methods

#927

· khatchad opened

on Sep 13, 2026

A list slice used as a shape vector reads the whole list: the receiver's extent survives into a container's slice result

#926

· khatchad opened

on Sep 13, 2026

A generator that throws is floored to unknown shape by the resolver and reads as a gap; a first census finds the allocation-site extractor throwing inside the `np.array` generator

#925

· khatchad opened

on Sep 13, 2026

An elementwise operation over a tensor and an opaque operand floors to "not a tensor" where unknown shape is the sound answer

#924

· khatchad opened

on Sep 13, 2026

An unmodelled producer vanishes from a union, so the modelled branch decides the type alone

#921

· khatchad opened

on Sep 13, 2026

Navigation Menu

Uh oh!

All issues

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

The allocation-site extractor declines for a constant key and throws for every other key with no allocation behind it: a definition disagreeing with itself

The flow-sensitive integer chase declines silently at seventeen points, so a null cannot tell "the program does not say" from "the chase did not reach it"

A loop-carried producer read directly, with no analysis run, returns a shape set that excludes the program's value

Two producer self-recursion guards return ⊥ on the null-engine path where the resolver's policy is ⊤

Any allocation of the array type made outside the array summaries silently loses every method call: the array's methods are per-allocation fields, not class methods

A list slice used as a shape vector reads the whole list: the receiver's extent survives into a container's slice result

A generator that throws is floored to unknown shape by the resolver and reads as a gap; a first census finds the allocation-site extractor throwing inside the `np.array` generator

An elementwise operation over a tensor and an opaque operand floors to "not a tensor" where unknown shape is the sound answer

An unmodelled producer vanishes from a union, so the modelled branch decides the type alone

Uh oh!

Search results