test_generated_cores

Generated cores go through the same seams as checked-in ones.

A tile whose HDL is rendered is the first thing in this registry that has no file beside its entry. That is a place where a parallel mechanism could easily grow – a build-layer side channel that renders somewhere the registry cannot see. The rule against it is the one already learned the hard way: a tile known only to the build layer never reaches core_names, so its HDL goes unstaged, its spec hash collides with its siblings, and its resources go unpriced.

So: the generator is reached through the registry, its parameters are hydra configuration like every other fact, and the rendered file arrives at source_path() – the same accessor dependency closure and synthesis already use.

Attributes

_BASE
EXP_EXACT
EXP_LIGHT

Functions

_registered_point(name) → dau_core.fixed_point.ExpDatapath
_definition(**overrides) → dau_core.cores.CoreDefinition
test_a_generated_core_refuses_its_source_until_rendered(...) → None	Rendering implicitly on read would hide which parameters produced the
test_rendering_writes_the_source_the_entry_names(tmp_path) → None
test_a_checked_in_core_refuses_to_be_rendered() → None	The inverse mistake: rendering over a hand-written tile would replace
test_the_operating_point_comes_from_configuration(tmp_path) → None	The parameters are hydra-injected, not Python literals — the same rule
test_each_operating_point_renders_a_different_source() → None	Two points must not render identical HDL, or the spec hash could not
test_the_rendered_source_carries_the_parameters_that_made_it() → None	A generated file with no record of its inputs is untraceable back to

Module Contents

test_generated_cores._BASE: dict[str, Any]

test_generated_cores._registered_point(name: str) → dau_core.fixed_point.ExpDatapath

test_generated_cores.EXP_EXACT

test_generated_cores.EXP_LIGHT

test_generated_cores._definition(**overrides) → dau_core.cores.CoreDefinition

test_generated_cores.test_a_generated_core_refuses_its_source_until_rendered(tmp_path: pathlib.Path, monkeypatch) → None

Rendering implicitly on read would hide which parameters produced the file that then gets synthesized.

Points the render root at an empty directory: the session fixture renders the real palette, and without this isolation that fixture would satisfy the very refusal under test.

test_generated_cores.test_rendering_writes_the_source_the_entry_names(tmp_path: pathlib.Path) → None

test_generated_cores.test_a_checked_in_core_refuses_to_be_rendered() → None

The inverse mistake: rendering over a hand-written tile would replace reviewed HDL with generated text.

test_generated_cores.test_the_operating_point_comes_from_configuration(tmp_path: pathlib.Path) → None

The parameters are hydra-injected, not Python literals — the same rule the rest of the registry follows. Instantiating from a config tree is the thing that must work, so it is what is tested.

test_generated_cores.test_each_operating_point_renders_a_different_source() → None

Two points must not render identical HDL, or the spec hash could not tell them apart and one measurement would be quoted for both.

test_generated_cores.test_the_rendered_source_carries_the_parameters_that_made_it() → None

A generated file with no record of its inputs is untraceable back to the configuration that produced it.