regenerate something when the sentences changed

sentence_api/tests/test_api.py

@@ -555,6 +555,10 @@ def test_admin_can_split_sentence_and_reindexes_following_sentences(tmp_path):
         ),
         None,
     )
+    repository.save_sentence_translations(
+        VIDEO_HASH,
+        {0: "之前。", 1: "第一段第二段", 2: "下一句。"},
+    )
 
     class UnavailableTranscriber:
         available = False
@@ -586,6 +590,7 @@ def test_admin_can_split_sentence_and_reindexes_following_sentences(tmp_path):
         "Next sentence.",
     ]
     assert [item.index for item in document.sentences] == [0, 1, 2, 3]
+    assert all(not (item.zh_text or "").strip() for item in document.sentences)
 
 
 def test_admin_can_merge_adjacent_sentences_and_reindexes_following_sentences(tmp_path):
@@ -643,6 +648,10 @@ def test_admin_can_merge_adjacent_sentences_and_reindexes_following_sentences(tm
         ),
         None,
     )
+    repository.save_sentence_translations(
+        VIDEO_HASH,
+        {0: "之前。", 1: "第一段", 2: "第二段", 3: "下一句。"},
+    )
 
     class UnavailableTranscriber:
         available = False
@@ -676,6 +685,7 @@ def test_admin_can_merge_adjacent_sentences_and_reindexes_following_sentences(tm
     assert [item.index for item in document.sentences] == [0, 1, 2]
     assert [item.start_ms for item in document.sentences] == [0, 2000, 9000]
     assert [item.end_ms for item in document.sentences] == [2000, 9000, 12000]
+    assert all(not (item.zh_text or "").strip() for item in document.sentences)
 
     last_response = client.put(
         f"/api/v1/admin/videos/{VIDEO_HASH}/sentences/2/merge",
@@ -684,6 +694,99 @@ def test_admin_can_merge_adjacent_sentences_and_reindexes_following_sentences(tm
     assert last_response.json()["detail"] == "No next sentence to merge with."
 
 
+
+def test_admin_sentence_merge_restarts_subtitle_translation(tmp_path, monkeypatch):
+    client = make_client(tmp_path)
+    _ready_video(client.app.state.video_repository)
+    requested = []
+
+    def translate_sentences(sentences, **kwargs):
+        requested.append([(index, text) for index, text in sentences])
+        return {index: f"中文:{text}" for index, text in sentences}
+
+    class StubAITutor:
+        available = True
+
+    stub = StubAITutor()
+    stub.translate_sentences = translate_sentences
+    monkeypatch.setattr(client.app.state, "ai_tutor", stub)
+    client.app.state.subtitle_translations.ai_tutor = stub
+    client.app.state.subtitle_translations.post_wait_seconds = 0.2
+
+    first = client.post(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
+    if not first.json().get("complete"):
+        first = wait_for_subtitle_translation(client, VIDEO_HASH)
+    assert first.json()["translations"] == [
+        {"index": 0, "zh_text": "中文:Hello."},
+        {"index": 1, "zh_text": "中文:Nice to meet you."},
+    ]
+
+    merge = client.put(f"/api/v1/admin/videos/{VIDEO_HASH}/sentences/0/merge")
+    assert merge.status_code == 200, merge.text
+    assert merge.json()["text"] == "Hello. Nice to meet you."
+
+    finished = wait_for_subtitle_translation(client, VIDEO_HASH)
+    assert finished.json()["complete"] is True
+    assert finished.json()["translations"] == [
+        {"index": 0, "zh_text": "中文:Hello. Nice to meet you."},
+    ]
+    document = client.app.state.video_repository.get_document(VIDEO_HASH)
+    assert [sentence.text for sentence in document.sentences] == ["Hello. Nice to meet you."]
+    assert [sentence.zh_text for sentence in document.sentences] == ["中文:Hello. Nice to meet you."]
+    assert requested[-1] == [(0, "Hello. Nice to meet you.")]
+
+
+def test_admin_sentence_merge_drops_stale_in_flight_translation(tmp_path, monkeypatch):
+    client = make_client(tmp_path)
+    _ready_video(client.app.state.video_repository)
+    started = threading.Event()
+    release = threading.Event()
+    requested = []
+
+    def translate_sentences(sentences, **kwargs):
+        call_index = len(requested)
+        requested.append([(index, text) for index, text in sentences])
+        persist = kwargs.get("persist")
+        if call_index == 0:
+            if persist is not None:
+                persist({0: "旧0", 1: "旧1"})
+            started.set()
+            assert release.wait(timeout=5)
+            if persist is not None:
+                persist({0: "晚0", 1: "晚1"})
+            return {0: "晚0", 1: "晚1"}
+        return {index: f"新:{text}" for index, text in sentences}
+
+    class StubAITutor:
+        available = True
+
+    stub = StubAITutor()
+    stub.translate_sentences = translate_sentences
+    monkeypatch.setattr(client.app.state, "ai_tutor", stub)
+    client.app.state.subtitle_translations.ai_tutor = stub
+    client.app.state.subtitle_translations.post_wait_seconds = 0.2
+
+    response = client.post(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
+    assert started.wait(timeout=2)
+    assert response.status_code == 200, response.text
+
+    merge = client.put(f"/api/v1/admin/videos/{VIDEO_HASH}/sentences/0/merge")
+    assert merge.status_code == 200, merge.text
+    release.set()
+
+    finished = wait_for_subtitle_translation(client, VIDEO_HASH)
+    assert finished.json()["complete"] is True
+    assert finished.json()["translations"] == [
+        {"index": 0, "zh_text": "新:Hello. Nice to meet you."},
+    ]
+    document = client.app.state.video_repository.get_document(VIDEO_HASH)
+    assert [sentence.zh_text for sentence in document.sentences] == [
+        "新:Hello. Nice to meet you."
+    ]
+    assert requested[0] == [(0, "Hello."), (1, "Nice to meet you.")]
+    assert requested[-1] == [(0, "Hello. Nice to meet you.")]
+
+
 def test_admin_boundary_adjustment_updates_linked_next_start(tmp_path):
     settings = replace(Settings.from_env(), data_dir=tmp_path / "data")
     settings.ensure_directories()