from __future__ import annotations from types import SimpleNamespace import pytest import torch from ComfyUI_H3_Continuum_Join import nodes as root_nodes from ComfyUI_H3_Continuum_Join.graph_contract import build_upstream_graph_contract from ComfyUI_H3_Continuum_Join.reference import ReferenceAssets, encode_reference_prompt from ComfyUI_H3_Continuum_Join.reference_audio import ( H3ContinuumReferenceAudios, REFERENCE_AUDIOS_TYPE, ReferenceAudioError, combine_reference_audio_identity, encode_reference_audio_input, prepare_reference_audio_bundle, prepare_reference_audio_source, resolve_reference_audio_input, validate_reference_audio_prompts, ) from ComfyUI_H3_Continuum_Join.v2.h3_builder import encode_prompt_conditioning from ComfyUI_H3_Continuum_Join.v3.driving_nodes import ( H3ContinuumSamplerV37, H3ContinuumSamplerV38, ) def _audio(value: float) -> dict: return { "waveform": torch.full((1, 2, 320), value, dtype=torch.float32), "sample_rate": 32000, } class _AudioVAE: audio_sample_rate = 32000 def __init__(self): self.inputs = [] def encode(self, waveform): self.inputs.append(waveform.clone()) value = float(waveform.mean().item()) return torch.full((1, 32, 2, len(self.inputs)), value, dtype=torch.float32) class _Clip: def __init__(self): self.options = None def tokenize(self, prompt, **kwargs): self.options = kwargs return prompt def encode_from_tokens_scheduled(self, tokens): return [[torch.zeros((1, 1, 2)), {}]] def test_public_helper_and_v38_socket_are_the_only_surface_additions(): helper_id = "H3ContinuumReferenceAudios" helper_schema = H3ContinuumReferenceAudios.INPUT_TYPES() v37 = H3ContinuumSamplerV37.INPUT_TYPES() v38 = H3ContinuumSamplerV38.INPUT_TYPES() assert root_nodes.NODE_CLASS_MAPPINGS[helper_id] is H3ContinuumReferenceAudios assert root_nodes.NODE_DISPLAY_NAME_MAPPINGS[helper_id] == ( "H3 Continuum Reference Audios" ) assert tuple(helper_schema) == ("optional",) assert tuple(helper_schema["optional"]) == ( "reference_audio_1", "reference_audio_2", "reference_audio_3", "reference_audio_vae", ) assert H3ContinuumReferenceAudios.RETURN_TYPES == (REFERENCE_AUDIOS_TYPE,) assert "audio_references" not in v37["optional"] assert list(v38["optional"])[:-1] == list(v37["optional"]) assert v38["optional"]["audio_references"][0] == REFERENCE_AUDIOS_TYPE def test_helper_preserves_order_and_requires_contiguous_core_numbering(): vae = _AudioVAE() bundle = H3ContinuumReferenceAudios().pack( _audio(0.1), _audio(0.2), _audio(0.3), vae )[0] assert bundle.count == 3 assert bundle.reference_audio_vae is vae assert [item.source_sha256 for item in bundle.sources] == [ item["reference_audio"]["source_sha256"] for item in bundle.contract["items"] ] assert [item["audio_index"] for item in bundle.contract["items"]] == [1, 2, 3] with pytest.raises(ReferenceAudioError, match="without gaps"): prepare_reference_audio_bundle(_audio(0.1), None, _audio(0.3), vae) assert prepare_reference_audio_bundle(None, None, None, vae) is None def test_bundle_order_changes_run_identity_and_legacy_identity_stays_v1(): vae = _AudioVAE() first = prepare_reference_audio_bundle(_audio(0.1), _audio(0.2), None, vae) swapped = prepare_reference_audio_bundle(_audio(0.2), _audio(0.1), None, vae) assert first is not None and swapped is not None first_identity = combine_reference_audio_identity("visual", first) swapped_identity = combine_reference_audio_identity("visual", swapped) assert first_identity != swapped_identity assert first.contract != swapped.contract legacy = SimpleNamespace(combined_hash="audio") assert combine_reference_audio_identity("visual", legacy) == ( "59fa60ba2a9a19525222c867aa2c2b8350dbf6af1e24ad34056317bcd40c81cb" ) def test_legacy_and_bundle_inputs_cannot_be_mixed(): vae = _AudioVAE() bundle = prepare_reference_audio_bundle(_audio(0.1), _audio(0.2), None, vae) with pytest.raises(ReferenceAudioError, match="not both"): resolve_reference_audio_input(_audio(0.1), vae, bundle) resolved, resolved_vae = resolve_reference_audio_input(None, object(), bundle) assert resolved is bundle assert resolved_vae is vae def test_each_reference_is_encoded_independently_in_order(): vae = _AudioVAE() bundle = prepare_reference_audio_bundle(_audio(0.1), _audio(0.2), _audio(0.3), vae) assert bundle is not None assets = encode_reference_audio_input(vae, bundle, cache_enabled=False) assert isinstance(assets, tuple) assert len(assets) == 3 assert len(vae.inputs) == 3 assert [round(float(item.audio_latent.mean().item()), 3) for item in assets] == [ 0.1, 0.2, 0.3, ] def test_three_assets_follow_core_tokenizer_and_packed_ref_order(): latents = tuple( SimpleNamespace(audio_latent=torch.full((1, 32, 2, index), float(index))) for index in (1, 2, 3) ) clip = _Clip() conditioning = encode_prompt_conditioning( clip, "