File size: 17,792 Bytes
bc50ea0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 | #!/usr/bin/env python3
"""Real PNG/WAV writes with tiny CPU VAEs: partial chapters and safe append."""
import copy
import json
import os
from pathlib import Path
import tempfile
import unittest
from unittest.mock import patch
import wave
import torch
from safetensors.torch import save_file
from _png_export_unit_test import chain, folder_paths
class VideoVAE:
def __init__(self):
self.calls = 0
def decode(self, video):
self.calls += 1
return torch.full((13, 4, 4, 3), float(video.flatten()[0]) / 255)
class AudioVAE:
audio_sample_rate = 8000
def __init__(self):
self.calls = 0
def decode(self, _audio):
self.calls += 1
count = chain.sample_boundary_from_frames(13, 8000, chain.FPS)
return torch.linspace(-0.5, 0.5, count).reshape(1, 1, -1)
class IncrementalExportTests(unittest.TestCase):
def setUp(self):
self.temp = tempfile.TemporaryDirectory()
self.addCleanup(self.temp.cleanup)
folder_paths.output_directory = self.temp.name
self.root = Path(self.temp.name)
self.video = VideoVAE()
self.audio = AudioVAE()
self.node = chain.MiniMaxH3ChainExportPNG()
self.segments = []
for index in range(1, 11):
checkpoint = self.root / "h3_chains" / "chapter_append" / "checkpoints" / (
"clip_%04d.safetensors" % index)
checkpoint.parent.mkdir(parents=True, exist_ok=True)
save_file({"video": torch.full((1, 24, 4, 1, 1), float(index)),
"audio": torch.zeros((1, 32, 2, 22))}, checkpoint)
segment = checkpoint.with_suffix(".mp4")
segment.write_bytes(b"placeholder video")
self.segments.append({
"index": index, "id": "scene_%02d" % index,
"revision": "%032x" % index,
"checkpoint": chain._relative_output_path(str(checkpoint)),
"checkpoint_sha256": chain._file_sha256(str(checkpoint)),
"segment": chain._relative_output_path(str(segment)),
"segment_sha256": chain._file_sha256(str(segment)),
"raw_frames": 13, "delivered_frames": 13,
"blend_frames": 0, "prompt": "test scene %d" % index,
})
self.editorial = {
"format": "h3_chain_editorial_v1", "run_name": "chapter_append",
"scene_order": [{"scene": i, "scene_id": "scene_%02d" % i}
for i in range(1, 11)],
"chapters": [{"id": "one", "title": "One", "start_scene": 1,
"start_scene_id": "scene_01"},
{"id": "two", "title": "Two", "start_scene": 5,
"start_scene_id": "scene_05"}],
"placements": [], "trims": [],
}
self.partial = self.chapter(8)
self.complete = self.chapter(10)
def manifest(self, end):
return {
"format": "h3_chain_partial_manifest_v3",
"run_name": "chapter_append", "planned_clip_count": 10,
"clip_count": end, "total_delivered_frames": 13 * end,
"compatibility": {"continuation_mode": "guide"},
"segments": copy.deepcopy(self.segments[:end]),
"editorial": copy.deepcopy(self.editorial), "archives": {},
}
def chapter(self, end):
return chain._chapter_manifest_from_manifest(self.manifest(end), 2)[0]
def export(self, manifest=None, **kwargs):
options = {"video_vae": self.video, "audio_vae": self.audio,
"embed_workflow": False, "save_workers": 2}
options.update(kwargs)
result = self.node.export(manifest or self.partial, **options)["result"]
directory = Path(result[0])
return directory, json.loads((directory / "export.json").read_text()), result
def png_state(self, directory):
return {path.name: (path.stat().st_mtime_ns, path.read_bytes())
for path in directory.glob("frame_*.png")}
def test_repeat_is_no_decode_and_preserves_png_and_wav(self):
output, record, result = self.export()
before = self.png_state(output)
wav_before = (output / "audio.wav").read_bytes()
wav_mtime = (output / "audio.wav").stat().st_mtime_ns
self.assertEqual(result[1], 52)
self.assertFalse(record["chapter_complete"])
again, record, result = self.export(checkpoint_verification="strict")
self.assertEqual(again, output)
self.assertEqual(self.video.calls, 4)
self.assertEqual(self.audio.calls, 4)
self.assertEqual(self.png_state(output), before)
self.assertEqual((output / "audio.wav").read_bytes(), wav_before)
self.assertEqual((output / "audio.wav").stat().st_mtime_ns, wav_mtime)
self.assertEqual(record["new_frame_count"], 0)
self.assertEqual(record["reused_frame_count"], 52)
self.assertIn("WAV reused", result[2])
def test_16bit_choice_isolated_from_legacy_8bit_and_reusable(self):
import av
import numpy as np
eight, _, _ = self.export(audio_vae=None)
before = self.png_state(eight)
# FP16 white must not overflow when scaled to RGB16's 65535 range.
with patch.object(self.video, "decode", return_value=torch.ones((13, 4, 4, 3), dtype=torch.float16)):
sixteen, record, result = self.export(audio_vae=None, png_bit_depth="16")
self.assertNotEqual(sixteen, eight)
self.assertEqual(record["settings"]["png_bit_depth"], 16)
self.assertIsNone(result[4])
with av.open(str(sixteen / "frame_00000001.png")) as container:
actual = next(container.decode(video=0)).to_ndarray(format="rgb48le")
np.testing.assert_array_equal(actual, np.full((4, 4, 3), 65535, dtype=np.uint16))
with patch.object(self.video, "decode", side_effect=AssertionError("decoded reused scene")):
reused, _, _ = self.export(audio_vae=None, png_bit_depth="16", checkpoint_verification="strict")
self.assertEqual(reused, sixteen)
self.assertEqual(self.png_state(eight), before)
def test_timestamp_only_changes_reuse_png_and_wav_without_decoding(self):
output, _, _ = self.export()
for path in [*output.glob("frame_*.png"), output / "audio.wav"]:
saved = path.stat()
os.utime(path, ns=(saved.st_atime_ns, saved.st_mtime_ns + 49_000_000_000))
png_before = self.png_state(output)
wav_before = ((output / "audio.wav").stat().st_mtime_ns, (output / "audio.wav").read_bytes())
calls = self.video.calls, self.audio.calls
for verification in ("cached", "strict"):
with self.subTest(verification=verification):
reused, record, result = self.export(checkpoint_verification=verification)
self.assertEqual(reused, output)
self.assertEqual((self.video.calls, self.audio.calls), calls)
self.assertEqual(self.png_state(output), png_before)
self.assertEqual(((output / "audio.wav").stat().st_mtime_ns, (output / "audio.wav").read_bytes()), wav_before)
self.assertEqual(record["new_frame_count"], 0)
self.assertIn("WAV reused", result[2])
def test_append_exports_only_new_pngs_and_rebuilds_frame_locked_audio(self):
output, _, _ = self.export()
before = self.png_state(output)
extended, record, result = self.export(self.complete)
self.assertEqual(extended, output)
self.assertEqual(self.video.calls, 6)
self.assertEqual(self.audio.calls, 10) # Four initially, six for new joins.
self.assertTrue(record["chapter_complete"])
self.assertEqual(record["new_frame_count"], 26)
self.assertEqual(record["reused_frame_count"], 52)
self.assertEqual(result[1], 78)
self.assertEqual([r["index"] for r in record["clips"]], list(range(5, 11)))
after = self.png_state(output)
self.assertEqual({key: after[key] for key in before}, before)
with wave.open(str(output / "audio.wav"), "rb") as audio:
self.assertEqual(audio.getnframes(),
chain.sample_boundary_from_frames(78, 8000, chain.FPS))
restored, _ = chain._load_chapter_manifest(
"chapter_append", 2, self.partial["chapter_manifest_id"])
self.assertEqual(restored["scene_end"], 8)
self.assertTrue(list((output / "export.history").glob("*.json")))
def test_changed_revision_uses_new_folder_and_preserves_old_files(self):
output, _, _ = self.export()
before = self.png_state(output)
changed = copy.deepcopy(self.partial)
changed["segments"][0]["revision"] = "f" * 32
other, _, _ = self.export(changed)
self.assertNotEqual(output, other)
self.assertEqual(self.png_state(output), before)
self.assertEqual(self.video.calls, 8)
def test_changed_trim_or_placement_does_not_append(self):
output, _, _ = self.export(audio_vae=None)
self.editorial["trims"] = [
{"scene": 5, "scene_id": "scene_05", "out_frame": 9}]
trimmed, record, _ = self.export(self.chapter(8), audio_vae=None)
self.assertNotEqual(trimmed, output)
self.assertEqual(record["frame_count"], 48)
changed = copy.deepcopy(self.partial)
changed["editorial"]["placements"][0]["start_frame"] = 1
moved, _, _ = self.export(changed, audio_vae=None)
self.assertNotEqual(moved, output)
def test_settings_and_explicit_fresh_export_are_isolated(self):
output, _, _ = self.export(audio_vae=None)
for options in ({"png_compression": 0}, {"first_frame_number": 10},
{"embed_workflow": True}, {"reuse_existing": False}):
with self.subTest(options=options):
other, _, _ = self.export(audio_vae=None, **options)
self.assertNotEqual(other, output)
def test_missing_png_does_not_repair_or_overwrite_old_folder(self):
output, _, _ = self.export(audio_vae=None)
removed = output / "frame_00000001.png"
removed.unlink()
before = self.png_state(output)
other, _, _ = self.export(audio_vae=None)
self.assertNotEqual(other, output)
self.assertFalse(removed.exists())
self.assertEqual(self.png_state(output), before)
def test_strict_detects_same_size_same_mtime_file_corruption(self):
output, _, _ = self.export(audio_vae=None)
damaged = output / "frame_00000001.png"
before = damaged.stat()
contents = bytearray(damaged.read_bytes())
contents[-1] ^= 1
damaged.write_bytes(contents)
os.utime(damaged, ns=(before.st_atime_ns, before.st_mtime_ns))
other, _, _ = self.export(audio_vae=None, checkpoint_verification="strict")
self.assertNotEqual(other, output)
self.assertEqual(damaged.read_bytes(), contents)
def test_orphan_png_or_legacy_sidecar_is_not_overwritten(self):
output, record, _ = self.export(audio_vae=None)
orphan = output / "frame_99999999.png"
orphan.write_bytes(b"untracked")
other, _, _ = self.export(audio_vae=None)
self.assertNotEqual(other, output)
self.assertEqual(orphan.read_bytes(), b"untracked")
old = json.loads((other / "export.json").read_text())
old.pop("incremental")
chain._atomic_json(str(other / "export.json"), old)
third, _, _ = self.export(audio_vae=None)
self.assertNotIn(third, (output, other))
def test_failed_append_keeps_old_success_and_retries_in_new_folder(self):
output, before_record, _ = self.export(audio_vae=None)
before = self.png_state(output)
with patch.object(self.video, "decode", side_effect=RuntimeError("decode failed")):
with self.assertRaisesRegex(RuntimeError, "decode failed"):
self.export(self.complete, audio_vae=None)
self.assertEqual(self.png_state(output), before)
self.assertEqual(json.loads((output / "export.json").read_text()), before_record)
self.assertTrue((output / "export.partial.json").exists())
retried, _, _ = self.export(self.complete, audio_vae=None)
self.assertNotEqual(retried, output)
def test_audio_only_can_repeat_and_grow(self):
output, record, _ = self.export(video_vae=None)
self.assertEqual(record["frame_count"], 0)
self.assertEqual(record["timeline_frame_count"], 52)
same, _, _ = self.export(video_vae=None)
self.assertEqual(same, output)
self.assertEqual(self.audio.calls, 4)
same, record, _ = self.export(self.complete, video_vae=None)
self.assertEqual(same, output)
self.assertEqual(record["timeline_frame_count"], 78)
self.assertFalse(list(output.glob("*.png")))
def test_failed_audio_publish_keeps_previous_wav_and_export_index(self):
output, before_record, _ = self.export()
before = (output / "audio.wav").read_bytes()
with patch.object(chain, "_write_wav", side_effect=RuntimeError("WAV failed")):
with self.assertRaisesRegex(RuntimeError, "WAV failed"):
self.export(self.complete)
self.assertEqual((output / "audio.wav").read_bytes(), before)
self.assertEqual(json.loads((output / "export.json").read_text()), before_record)
self.assertTrue((output / "export.partial.json").exists())
def test_older_snapshot_never_truncates_a_grown_export(self):
output, _, _ = self.export(self.complete, audio_vae=None)
before = self.png_state(output)
older, record, _ = self.export(audio_vae=None)
self.assertNotEqual(older, output)
self.assertEqual(record["frame_count"], 52)
self.assertEqual(self.png_state(output), before)
def test_whole_run_keeps_fresh_folder_behavior(self):
first, _, _ = self.export(self.manifest(8), audio_vae=None)
second, _, _ = self.export(self.manifest(8), audio_vae=None)
self.assertNotEqual(first, second)
def test_finished_chapter_is_reused_when_other_chapters_grow(self):
first = chain._chapter_manifest_from_manifest(self.manifest(8), 1)[0]
later = chain._chapter_manifest_from_manifest(self.manifest(10), 1)[0]
output, _, _ = self.export(first, audio_vae=None)
before = self.png_state(output)
same, record, _ = self.export(later, audio_vae=None)
self.assertEqual(output, same)
self.assertEqual(self.png_state(output), before)
self.assertEqual(record["new_frame_count"], 0)
@unittest.skipIf(chain.av is None, "PyAV is required for real MP4 assembly")
def test_partial_chapter_can_also_assemble_saved_mp4s_without_vae(self):
for segment in self.segments[4:8]:
path = chain._absolute_output_path(segment["segment"])
chain._write_segment_video(torch.zeros((13, 16, 16, 3)), path, 24, 18)
segment["segment_sha256"] = chain._file_sha256(path)
chapter = self.chapter(8)
result = chain.MiniMaxH3ChainAssemble().assemble(
chapter, audio_source="none", filename="partial_chapter", audio_bitrate=192)
output = Path(result["result"][0])
self.assertIn("chapters/02_two/final/", output.as_posix())
with chain.av.open(str(output)) as movie:
self.assertEqual(sum(1 for _ in movie.decode(video=0)), 52)
def test_numbering_can_cross_eight_digit_boundary(self):
output, _, _ = self.export(audio_vae=None, first_frame_number=99999990)
same, record, _ = self.export(audio_vae=None, first_frame_number=99999990)
self.assertEqual(output, same)
self.assertEqual(record["new_frame_count"], 0)
def test_unrecorded_checkpoint_hash_is_still_checked_for_reuse(self):
legacy = copy.deepcopy(self.partial)
legacy["segments"][0].pop("checkpoint_sha256")
output, _, _ = self.export(legacy, audio_vae=None)
checkpoint = Path(chain._absolute_output_path(legacy["segments"][0]["checkpoint"]))
before = checkpoint.stat()
tensors = chain._st_load(str(checkpoint))
tensors["video"] += 1
save_file(tensors, checkpoint)
os.utime(checkpoint, ns=(before.st_atime_ns, before.st_mtime_ns))
self.assertEqual(checkpoint.stat().st_size, before.st_size)
other, _, _ = self.export(legacy, audio_vae=None)
self.assertNotEqual(other, output)
def test_export_lock_blocks_competitor_and_releases_on_failure(self):
with chain._chapter_png_export_lock(self.partial, "png_sequence"):
with self.assertRaisesRegex(ValueError, "Another export"):
self.export(audio_vae=None)
with self.assertRaisesRegex(RuntimeError, "test failure"):
with chain._chapter_png_export_lock(self.partial, "png_sequence"):
raise RuntimeError("test failure")
self.export(audio_vae=None)
def test_symlinked_png_and_sidecar_are_not_reused(self):
output, _, _ = self.export(audio_vae=None)
image = output / "frame_00000001.png"
target = self.root / "preserve.png"
image.rename(target)
image.symlink_to(target)
second, _, _ = self.export(audio_vae=None)
self.assertNotEqual(second, output)
sidecar = second / "export.json"
target_record = self.root / "preserve.json"
sidecar.rename(target_record)
sidecar.symlink_to(target_record)
third, _, _ = self.export(audio_vae=None)
self.assertNotIn(third, (output, second))
if __name__ == "__main__":
unittest.main()
|