feat: word-based chunker mit sentence-boundary look-back

tests/test_chunker.py

@@ -0,0 +1,54 @@
+from app.ingest.chunker import chunk_text, Chunk
+
+
+def test_chunk_short_text_single_chunk():
+    text = "Das ist ein kurzer Text mit wenigen Worten."
+    chunks = chunk_text(text, size_words=500, overlap_words=50, page=1)
+    assert len(chunks) == 1
+    assert chunks[0].text == text
+    assert chunks[0].page == 1
+
+
+def test_chunk_size_and_overlap():
+    words = [f"w{i}" for i in range(1200)]
+    text = " ".join(words)
+    chunks = chunk_text(text, size_words=500, overlap_words=50, page=1)
+
+    # 1200 words, size 500, overlap 50 → step 450 → starts at 0, 450, 900 → 3 chunks
+    assert len(chunks) == 3
+    # First chunk has up to 500 words
+    assert len(chunks[0].text.split()) <= 500
+    # Overlap: last 50 words of chunk 0 are first 50 words of chunk 1
+    last_50_of_first = chunks[0].text.split()[-50:]
+    first_50_of_second = chunks[1].text.split()[:50]
+    assert last_50_of_first == first_50_of_second
+
+
+def test_chunk_respects_sentence_boundary_in_lookback_window():
+    # 600 words, with a sentence ending around word 480 (within last 20% = words 400-500)
+    words = [f"w{i}" for i in range(600)]
+    words[479] = "ende."
+    text = " ".join(words)
+    chunks = chunk_text(text, size_words=500, overlap_words=50, page=1)
+
+    # First chunk should end at the sentence boundary, not at word 500
+    first_chunk_words = chunks[0].text.split()
+    assert first_chunk_words[-1] == "ende."
+    assert len(first_chunk_words) == 480
+
+
+def test_chunk_no_sentence_boundary_in_window_falls_back_to_word_count():
+    words = [f"w{i}" for i in range(600)]
+    text = " ".join(words)
+    chunks = chunk_text(text, size_words=500, overlap_words=50, page=1)
+    # No sentence-end → exactly 500 words in first chunk
+    assert len(chunks[0].text.split()) == 500
+
+
+def test_chunk_empty_text_returns_empty_list():
+    assert chunk_text("", size_words=500, overlap_words=50, page=1) == []
+
+
+def test_chunk_carries_page_number():
+    chunks = chunk_text("hallo welt", size_words=500, overlap_words=50, page=7)
+    assert chunks[0].page == 7