diff options
Diffstat (limited to 'tests/test_chunking.py')
| -rw-r--r-- | tests/test_chunking.py | 48 |
1 files changed, 23 insertions, 25 deletions
diff --git a/tests/test_chunking.py b/tests/test_chunking.py index 7f37f80..3cd926b 100644 --- a/tests/test_chunking.py +++ b/tests/test_chunking.py @@ -4,42 +4,41 @@ import unittest from unittest.mock import patch from converter import config -from converter.chunking import MAX_REQUEST_WORDS, split_into_chunks +from converter.chunking import split_into_chunks class ChunkSizeDefaultTests(unittest.TestCase): - """Guard the request-size settings: each API call is one model - generation, and the servers silently truncate audio past their caps - (~2.5 min faster backend, ~11 min Gradio demo), so both the default - chunk size and the hard ceiling must stay well inside that budget.""" + """Guard the request-size setting: each API call is one model + generation, and the servers silently truncate audio when a single + generation runs too long (~2.5 min faster backend, ~11 min Gradio + demo), so the default chunk size must stay well inside that budget. + There is no hard ceiling beyond CHUNK_SIZE; users raising it accept + the truncation risk themselves.""" - def test_default_chunk_size_within_request_ceiling(self): - self.assertLessEqual(config.CHUNK_SIZE, MAX_REQUEST_WORDS) + def test_default_chunk_size_within_single_generation_budget(self): + self.assertLessEqual(config.CHUNK_SIZE, 300) - def test_request_ceiling_within_single_generation_budget(self): - self.assertLessEqual(MAX_REQUEST_WORDS, 300) - - def test_sizes_are_positive(self): + def test_default_chunk_size_is_positive(self): self.assertGreaterEqual(config.CHUNK_SIZE, 1) - self.assertGreaterEqual(MAX_REQUEST_WORDS, 1) -class RequestCeilingClampTests(unittest.TestCase): - def test_oversized_chunk_size_is_clamped_with_warning(self): +class RequestSizeTests(unittest.TestCase): + def test_oversized_chunk_size_is_honored(self): + # No clamping: whatever size is configured (or requested) is used. text = " ".join(f"word{i}" for i in range(30)) + "." - with patch("converter.chunking.MAX_REQUEST_WORDS", 10), \ - self.assertLogs("converter.chunking", level="WARNING") as logs: - chunks = split_into_chunks(text, max_words=5000) - self.assertTrue(all(len(chunk.split()) <= 10 for chunk in chunks)) - self.assertIn("clamped", " ".join(logs.output)) + chunks = split_into_chunks(text, max_words=5000) + self.assertEqual(len(chunks), 1) + self.assertEqual(len(chunks[0].split()), 30) - def test_default_ceiling_clamps_realistic_configuration(self): + def test_default_uses_runtime_config_chunk_size(self): + # The default resolves config.CHUNK_SIZE at call time, so + # patching the config changes the default split size. sentences = " ".join( f"S{i} " + " ".join(["word"] * 8) + "." for i in range(60)) - chunks = split_into_chunks(sentences, max_words=5000) + with patch.object(config, "CHUNK_SIZE", 120): + chunks = split_into_chunks(sentences) self.assertGreater(len(chunks), 1) - self.assertTrue(all(len(chunk.split()) <= MAX_REQUEST_WORDS - for chunk in chunks)) + self.assertTrue(all(len(chunk.split()) <= 120 for chunk in chunks)) class SplitIntoChunksTests(unittest.TestCase): @@ -94,8 +93,7 @@ class SplitIntoChunksTests(unittest.TestCase): def test_single_oversized_sentence_is_word_split(self): # A punctuation-free sentence longer than the limit is split at word - # boundaries: the request-size ceiling is a hard limit because the - # TTS servers silently truncate oversized generations. + # boundaries so no single request exceeds the configured size. sentence = " ".join(["word"] * 30) + "." chunks = split_into_chunks(sentence, max_words=10) self.assertGreater(len(chunks), 1) |