fix(product-suite): keep retries in current results

2026-07-16 18:15:24 +08:00
parent 25a08decdc
commit becfed81cd
3 changed files with 421 additions and 39 deletions
@@ -697,6 +697,9 @@ class SuiteTaskState:
    settings: dict = field(default_factory=product_suite.default_suite_settings)
    current_job_ids: list = field(default_factory=list)
    show_history: bool = False
+    generation_job_ids: list = field(default_factory=list)
+    generation_mode: str = "batch"
+    generation_retry_job_id: int = None
    worker: object = None
    thread: object = None
    generation_run_token: str = ""
@@ -1582,6 +1585,9 @@ class ProductSuiteTab(QWidget):
        state.project_binding_state = ""
        state.last_saved_prompt = ""
        state.current_job_ids = []
+        state.generation_job_ids = []
+        state.generation_mode = "batch"
+        state.generation_retry_job_id = None
        state.done = state.failed = state.total = 0
        state.started_at = None
        if state is self._displayed_state:
@@ -2517,12 +2523,22 @@ class ProductSuiteTab(QWidget):
        if state is None:
            return
        if state.generation_running():
+            retrying = state.generation_mode == "retry"
            if state.generation_stop_requested:
-                self._status("正在停止当前套图任务", "warning")
+                self._status(
+                    "正在停止当前图片重试"
+                    if retrying
+                    else "正在停止当前套图任务",
+                    "warning",
+                )
                return
            if self._confirm(
-                "停止生成套图",
-                "确认取消当前任务吗？已提交任务会在安全边界停止。",
+                "停止图片重试" if retrying else "停止生成套图",
+                (
+                    "确认停止当前图片重试吗？已提交任务会在安全边界停止。"
+                    if retrying
+                    else "确认取消当前任务吗？已提交任务会在安全边界停止。"
+                ),
                destructive=True,
            ):
                state.generation_stop_requested = True
@@ -2533,14 +2549,21 @@ class ProductSuiteTab(QWidget):
                    "stop_requested",
                )
                self._apply_running_state(state)
-                self._status("已请求停止当前套图任务", "warning")
+                self._status(
+                    "已请求停止当前图片重试"
+                    if retrying
+                    else "已请求停止当前套图任务",
+                    "warning",
+                )
            return
        self.start_generation(state)

-    def start_generation(self, state, specs=None):
+    def start_generation(self, state, specs=None, *, retry_job_id=None):
        if state.generation_running():
            self._status("当前套图任务仍在生成", "warning")
            return False
+        retry_job_id = int(retry_job_id) if retry_job_id is not None else None
+        retrying = retry_job_id is not None
        if state is self._displayed_state:
            self._save_controls_to_state(state)
        template_text = None
@@ -2595,12 +2618,16 @@ class ProductSuiteTab(QWidget):
        state.generation_run_token = run_token
        state.generation_stop_requested = False
        state.generation_terminal_streak = 0
+        state.generation_job_ids = []
+        state.generation_mode = "retry" if retrying else "batch"
+        state.generation_retry_job_id = retry_job_id
        state.done = 0
        state.failed = 0
        state.total = len(specs)
        state.started_at = time.monotonic()
-        state.current_job_ids = []
-        state.show_history = False
+        if not retrying:
+            state.current_job_ids = []
+            state.show_history = False
        self._generation_run_states[run_token] = state.key
        worker.progress.connect(self._on_generation_progress_signal)
        worker.finished.connect(self._on_generation_finished_signal)
@@ -2611,17 +2638,27 @@ class ProductSuiteTab(QWidget):
            state,
            run_token,
            "started",
-            {"total": len(specs), "job_ids": 0},
+            {
+                "total": len(specs),
+                "job_ids": 0,
+                "mode": state.generation_mode,
+            },
        )
-        if state is self._displayed_state:
+        if state is self._displayed_state and not retrying:
            self._loading = True
            try:
                self.history_button.setChecked(False)
            finally:
                self._loading = False
+        if state is self._displayed_state:
            self._apply_running_state(state)
            self._refresh_results(state)
-        self._status("商品套图生成已开始，共%d张；可切换到其他任务" % len(specs), "info")
+        self._status(
+            "图片重试已开始；可切换到其他任务"
+            if retrying
+            else "商品套图生成已开始，共%d张；可切换到其他任务" % len(specs),
+            "info",
+        )
        return True

    def _generation_signal_token(self, payload=None):
@@ -2651,7 +2688,7 @@ class ProductSuiteTab(QWidget):
        state.failed = int(payload.get("failed", state.failed) or 0)
        job_ids = payload.get("job_ids")
        if job_ids is not None:
-            state.current_job_ids = [int(job_id) for job_id in job_ids]
+            self._set_generation_job_ids(state, job_ids)
        state.generation_terminal_streak = 0
        if state is self._displayed_state:
            self._refresh_results(state)
@@ -2670,7 +2707,15 @@ class ProductSuiteTab(QWidget):
            {"has_error": True},
            level="ERROR",
        )
-        self._status("商品套图生成失败：%s" % _user_error(error), "danger")
+        self._status(
+            (
+                "图片重试失败：%s"
+                if state.generation_mode == "retry"
+                else "商品套图生成失败：%s"
+            )
+            % _user_error(error),
+            "danger",
+        )

    @Slot(dict)
    def _on_generation_finished_signal(self, result):
@@ -2678,6 +2723,8 @@ class ProductSuiteTab(QWidget):
        state = self._generation_state(token)
        if state is None:
            return
+        if result.get("job_ids") is not None:
+            self._set_generation_job_ids(state, result.get("job_ids"))
        self._log_generation_lifecycle(
            state,
            token,
@@ -2779,7 +2826,7 @@ class ProductSuiteTab(QWidget):
            )

    def _generation_job_ids(self, state):
-        job_ids = list(state.current_job_ids)
+        job_ids = list(state.generation_job_ids)
        if not job_ids and state.worker is not None:
            worker_job_ids = getattr(state.worker, "job_ids", [])
            if not isinstance(worker_job_ids, (list, tuple, set)):
@@ -2789,9 +2836,30 @@ class ProductSuiteTab(QWidget):
                for job_id in list(worker_job_ids or [])
            ]
            if job_ids:
-                state.current_job_ids = job_ids
+                self._set_generation_job_ids(state, job_ids)
        return job_ids

+    def _set_generation_job_ids(self, state, job_ids):
+        normalized = []
+        for job_id in job_ids or []:
+            value = int(job_id)
+            if value not in normalized:
+                normalized.append(value)
+        state.generation_job_ids = normalized
+        if state.generation_mode != "retry":
+            state.current_job_ids = list(normalized)
+            return
+        retry_job_id = state.generation_retry_job_id
+        current = [int(job_id) for job_id in state.current_job_ids]
+        if retry_job_id in current:
+            index = current.index(retry_job_id)
+            current[index:index + 1] = normalized
+        elif not state.show_history:
+            for job_id in normalized:
+                if job_id not in current:
+                    current.append(job_id)
+        state.current_job_ids = current
+
    def _generation_job_snapshot(self, state):
        job_ids = self._generation_job_ids(state)
        counts = {
@@ -2865,6 +2933,7 @@ class ProductSuiteTab(QWidget):
    def _finalize_generation(self, state, run_token, result, *, source):
        if self._generation_state(run_token) is not state:
            return False
+        retrying = state.generation_mode == "retry"
        snapshot = self._generation_job_snapshot(state)
        if snapshot["job_ids"] and (
            snapshot["all_terminal"]
@@ -2890,6 +2959,8 @@ class ProductSuiteTab(QWidget):
        state.generation_run_token = ""
        state.generation_stop_requested = False
        state.generation_terminal_streak = 0
+        state.generation_job_ids = []
+        state.generation_retry_job_id = None
        state.worker = None
        state.thread = None
        state.done = success + failed + cancelled
@@ -2912,6 +2983,7 @@ class ProductSuiteTab(QWidget):
                "cancelled": cancelled,
                "active": active,
                "elapsed_seconds": elapsed,
+                "mode": "retry" if retrying else "batch",
            },
            level="WARNING" if active or result.get("ok") is False else "INFO",
        )
@@ -2923,39 +2995,76 @@ class ProductSuiteTab(QWidget):
                or "生成线程已结束，部分任务可稍后继续查询"
            )
            if state is self._displayed_state:
-                self._message("商品套图生成未完整结束", _user_error(message))
+                self._message(
+                    "图片重试未完整结束"
+                    if retrying
+                    else "商品套图生成未完整结束",
+                    _user_error(message),
+                )
            else:
                self._status(
-                    "套图任务%d生成未完整结束" % state.serial,
+                    (
+                        "套图任务%d图片重试未完整结束"
+                        if retrying
+                        else "套图任务%d生成未完整结束"
+                    )
+                    % state.serial,
                    "danger",
                )
            return True
        if stop_requested or cancelled:
            if state is self._displayed_state:
-                self._message(
-                    "商品套图生成已停止",
-                    "本轮共%d张：成功%d张，失败%d张，停止%d张；"
-                    "已提交任务可稍后继续查询；总用时%d秒。"
-                    % (total, success, failed, cancelled, elapsed),
-                    icon=QMessageBox.Information,
-                )
+                if retrying:
+                    self._message(
+                        "图片重试已停止",
+                        "本次重试：成功%d张，失败%d张，停止%d张；"
+                        "已提交任务可稍后继续查询；总用时%d秒。"
+                        % (success, failed, cancelled, elapsed),
+                        icon=QMessageBox.Information,
+                    )
+                else:
+                    self._message(
+                        "商品套图生成已停止",
+                        "本轮共%d张：成功%d张，失败%d张，停止%d张；"
+                        "已提交任务可稍后继续查询；总用时%d秒。"
+                        % (total, success, failed, cancelled, elapsed),
+                        icon=QMessageBox.Information,
+                    )
            self._status(
-                "商品套图生成已停止：成功%d张，失败%d张，停止%d张"
+                (
+                    "图片重试已停止：成功%d张，失败%d张，停止%d张"
+                    if retrying
+                    else "商品套图生成已停止：成功%d张，失败%d张，停止%d张"
+                )
                % (success, failed, cancelled),
                "warning",
            )
            return True
        if state is self._displayed_state:
-            self._message(
-                "商品套图生成完成",
-                "本轮共%d张：成功%d张，失败%d张，停止%d张；总用时%d秒。"
-                % (total, success, failed, cancelled, elapsed),
-                icon=QMessageBox.Information,
+            if retrying:
+                self._message(
+                    "图片重试成功" if success and not failed else "图片重试失败",
+                    "本次重试：成功%d张，失败%d张，停止%d张；总用时%d秒。"
+                    % (success, failed, cancelled, elapsed),
+                    icon=QMessageBox.Information,
+                )
+            else:
+                self._message(
+                    "商品套图生成完成",
+                    "本轮共%d张：成功%d张，失败%d张，停止%d张；总用时%d秒。"
+                    % (total, success, failed, cancelled, elapsed),
+                    icon=QMessageBox.Information,
+                )
+        if retrying:
+            self._status(
+                "图片重试成功" if success and not failed else "图片重试失败，请查看失败卡片",
+                "success" if success and not failed else "danger",
+            )
+        else:
+            self._status(
+                "商品套图生成完成：成功%d张，失败%d张" % (success, failed),
+                "success",
            )
-        self._status(
-            "商品套图生成完成：成功%d张，失败%d张" % (success, failed),
-            "success",
-        )
        return True

    def _log_generation_lifecycle(
@@ -3012,6 +3121,8 @@ class ProductSuiteTab(QWidget):
            self.generate_button.setText(
                "正在停止..."
                if state.generation_stop_requested
+                else "停止重试"
+                if state.generation_mode == "retry"
                else "停止生成"
            )
            self.generate_button.setStyleSheet(
@@ -3035,7 +3146,8 @@ class ProductSuiteTab(QWidget):
            return
        elapsed = int(max(0, time.monotonic() - state.started_at)) if state.started_at else 0
        self.elapsed_label.setText(
-            "套图 %d/%d（%d秒） · 失败 %d" % (
+            "%s %d/%d（%d秒） · 失败 %d" % (
+                "重试" if state.generation_mode == "retry" else "套图",
                state.done,
                state.total,
                elapsed,
@@ -3067,8 +3179,12 @@ class ProductSuiteTab(QWidget):
            return []
        if state.show_history:
            return jobs
-        current = set(state.current_job_ids)
-        return [job for job in jobs if int(job.id) in current]
+        by_id = {int(job.id): job for job in jobs}
+        return [
+            by_id[int(job_id)]
+            for job_id in state.current_job_ids
+            if int(job_id) in by_id
+        ]

    def _refresh_results(self, state):
        while self.result_grid.count():
@@ -3122,12 +3238,19 @@ class ProductSuiteTab(QWidget):
        if state.generation_running():
            self._message("当前任务正在生成", "请等待当前生成结束或停止后再重试单张图片。")
            return
+        if str(getattr(job, "status", "") or "") not in {
+            "failed",
+            "expired",
+            "cancelled",
+        }:
+            self._status("当前图片无需重试", "warning")
+            return
        spec = {
            "source_asset_id": job.source_asset_id,
            "job_type": job.job_type,
            "prompt": job.prompt,
        }
-        self.start_generation(state, specs=[spec])
+        self.start_generation(state, specs=[spec], retry_job_id=job.id)

    def _show_job_menu(self, job, global_position):
        menu = QMenu(self)
@@ -3,7 +3,7 @@ id: T-640
 title: 商品套图单张失败重试保持当前轮结果
 phase: 7
 deps: [T-639]
-status: TODO
+status: DONE
 created: 2026-07-16
 ---

@@ -107,4 +107,7 @@ git diff --check

 ## 执行记录

- 待执行。
+- 2026-07-16：`SuiteTaskState` 新增内存态 `generation_job_ids`、`generation_mode` 与 `generation_retry_job_id`。`generation_job_ids` 专门供当前活动请求的进度、停止、线程结束和终态看门狗使用；`current_job_ids` 只表示当前结果区的有效槽位，解决单张重试时旧成功 job 被错误纳入本次汇总的问题。未修改 SQLite schema。
+- 2026-07-16：失败、过期或停止卡片的「重试」改为显式单张重试模式。worker 仍新建 job 保留历史和计费记录；新 job ID 返回后会原位替换当前结果区中的旧失败 job，其他成功图片保持不变。当前结果改按 `current_job_ids` 顺序展示；历史视图继续展示原失败与每次重试记录，从历史发起重试时不会强制退出历史视图。
+- 2026-07-16：重试运行状态、停止确认、进度文案和完成弹窗改为「图片重试」语义；重试成功、失败、停止和未完整结束分别给中文反馈。正常整轮生成、停止、T-639 `run_token` 隔离与终态兜底保持原行为。
+- 2026-07-16：补齐当前轮原位替换、历史记录保留、活动快照只统计新 job、重试再次失败仍可继续重试、历史视图保持和真实 QThread 完整退出测试。相关 42 项测试通过；在不包含用户未提交默认提示词改名的临时隔离 worktree 中，全量 535 项 unittest 通过，Ruff、`compileall` 与 `git diff --check` 均通过。
@@ -856,6 +856,7 @@ class ProductSuiteGuiTests(TempDirMixin, unittest.TestCase):
            state.thread = mock.Mock()
            state.generation_run_token = "watchdog-run"
            state.current_job_ids = [job.id for job in jobs]
+            state.generation_job_ids = [job.id for job in jobs]
            state.total = len(jobs)
            state.started_at = time.monotonic()
            tab._generation_run_states["watchdog-run"] = state.key
@@ -972,6 +973,7 @@ class ProductSuiteGuiTests(TempDirMixin, unittest.TestCase):
            state.thread = mock.Mock()
            state.generation_run_token = "thread-fallback"
            state.current_job_ids = [job.id]
+            state.generation_job_ids = [job.id]
            state.total = 1
            state.started_at = time.monotonic()
            tab._generation_run_states["thread-fallback"] = state.key
@@ -1127,6 +1129,260 @@ class ProductSuiteGuiTests(TempDirMixin, unittest.TestCase):

        self.assert_removed(temp_dir)

+    def test_failed_job_retry_replaces_current_slot_and_keeps_history(self):
+        with self.make_temp_dir() as temp_dir:
+            config = self._config(temp_dir)
+            project, sources = self._create_project_with_assets(temp_dir, config, 1)
+            source = sources[0]
+
+            success_path = os.path.join(temp_dir, "success.jpg")
+            retry_path = os.path.join(temp_dir, "retry.jpg")
+            self._write_image(success_path)
+            self._write_image(retry_path)
+            success_asset = image_studio.add_asset(
+                project.id,
+                "generated_main",
+                local_path=success_path,
+                parent_asset_id=source.id,
+                path=config["db_path"],
+            )
+            success_job = image_studio.create_job(
+                project.id,
+                source_asset_id=source.id,
+                job_type="白底图",
+                prompt="成功图",
+                path=config["db_path"],
+            )
+            success_job = image_studio.update_job_status(
+                success_job.id,
+                "succeeded",
+                output_asset_id=success_asset.id,
+                path=config["db_path"],
+            )
+            failed_job = image_studio.create_job(
+                project.id,
+                source_asset_id=source.id,
+                job_type="场景图",
+                prompt="失败图",
+                path=config["db_path"],
+            )
+            failed_job = image_studio.update_job_status(
+                failed_job.id,
+                "failed",
+                error="上游生成失败",
+                path=config["db_path"],
+            )
+
+            tab = ProductSuiteTab(config=config, db_path=config["db_path"])
+            self.addCleanup(tab.close)
+            state = tab._displayed_state
+            state.account_alias = "alias-a"
+            state.item_id = project.item_id
+            state.project_id = project.id
+            state.project_binding_state = project.binding_state
+            state.current_job_ids = [success_job.id, failed_job.id]
+            tab._load_state(state)
+            messages = []
+            tab._message = lambda title, message, **kwargs: messages.append(
+                (title, message)
+            )
+
+            def fake_run_jobs(jobs, **kwargs):
+                job = list(jobs)[0]
+                retry_asset = image_studio.add_asset(
+                    project.id,
+                    "generated_main",
+                    local_path=retry_path,
+                    parent_asset_id=source.id,
+                    path=config["db_path"],
+                )
+                image_studio.update_job_status(
+                    job.id,
+                    "succeeded",
+                    output_asset_id=retry_asset.id,
+                    path=config["db_path"],
+                )
+                return {
+                    "total": 1,
+                    "success": 1,
+                    "failed": 0,
+                    "cancelled": 0,
+                    "jobs": [],
+                }
+
+            with mock.patch(
+                "app.gui.workers.image_studio_generation.run_jobs",
+                side_effect=fake_run_jobs,
+            ):
+                tab.retry_job(failed_job)
+                generation_thread = state.thread
+                deadline = time.monotonic() + 3
+                while state.worker is not None and time.monotonic() < deadline:
+                    QTest.qWait(20)
+                    self.app.processEvents()
+                while (
+                    generation_thread is not None
+                    and generation_thread.isRunning()
+                    and time.monotonic() < deadline
+                ):
+                    QTest.qWait(20)
+                    self.app.processEvents()
+                self.assertFalse(generation_thread.isRunning())
+
+            all_jobs = image_studio.list_jobs(project.id, path=config["db_path"])
+            retry_jobs = [
+                job
+                for job in all_jobs
+                if job.id not in {success_job.id, failed_job.id}
+            ]
+            self.assertEqual(1, len(retry_jobs))
+            retry_job = retry_jobs[0]
+            self.assertEqual(
+                [success_job.id, retry_job.id],
+                state.current_job_ids,
+            )
+            self.assertEqual([], state.generation_job_ids)
+            self.assertEqual(
+                [success_job.id, retry_job.id],
+                [job.id for job in tab._jobs_for_state(state)],
+            )
+            self.assertEqual("图片重试成功", messages[-1][0])
+
+            state.show_history = True
+            history_ids = {job.id for job in tab._jobs_for_state(state)}
+            self.assertEqual(
+                {success_job.id, failed_job.id, retry_job.id},
+                history_ids,
+            )
+
+        self.assert_removed(temp_dir)
+
+    def test_retry_tracks_only_new_job_and_preserves_history_view(self):
+        with self.make_temp_dir() as temp_dir:
+            config = self._config(temp_dir)
+            project, sources = self._create_project_with_assets(temp_dir, config, 1)
+            source = sources[0]
+            success_job = image_studio.create_job(
+                project.id,
+                source_asset_id=source.id,
+                job_type="白底图",
+                prompt="成功图",
+                path=config["db_path"],
+            )
+            success_job = image_studio.update_job_status(
+                success_job.id,
+                "succeeded",
+                path=config["db_path"],
+            )
+            failed_job = image_studio.create_job(
+                project.id,
+                source_asset_id=source.id,
+                job_type="场景图",
+                prompt="失败图",
+                path=config["db_path"],
+            )
+            failed_job = image_studio.update_job_status(
+                failed_job.id,
+                "failed",
+                path=config["db_path"],
+            )
+            retry_job = image_studio.create_job(
+                project.id,
+                source_asset_id=source.id,
+                job_type="场景图",
+                prompt="重试图",
+                path=config["db_path"],
+            )
+
+            tab = ProductSuiteTab(config=config, db_path=config["db_path"])
+            self.addCleanup(tab.close)
+            state = tab._displayed_state
+            state.account_alias = "alias-a"
+            state.item_id = project.item_id
+            state.project_id = project.id
+            state.project_binding_state = project.binding_state
+            state.current_job_ids = [success_job.id, failed_job.id]
+            state.generation_mode = "retry"
+            state.generation_retry_job_id = failed_job.id
+            state.total = 1
+
+            tab._set_generation_job_ids(state, [retry_job.id])
+
+            self.assertEqual(
+                [success_job.id, retry_job.id],
+                state.current_job_ids,
+            )
+            self.assertEqual([retry_job.id], state.generation_job_ids)
+            snapshot = tab._generation_job_snapshot(state)
+            self.assertEqual(1, snapshot["job_ids"])
+            self.assertEqual(1, snapshot["active"])
+
+            image_studio.update_job_status(
+                retry_job.id,
+                "failed",
+                error="重试仍失败",
+                path=config["db_path"],
+            )
+            state.worker = mock.Mock()
+            state.thread = mock.Mock()
+            state.generation_run_token = "retry-failed"
+            state.started_at = time.monotonic()
+            tab._generation_run_states["retry-failed"] = state.key
+            messages = []
+            tab._message = lambda title, message, **kwargs: messages.append(
+                (title, message)
+            )
+
+            self.assertTrue(
+                tab._finalize_generation(
+                    state,
+                    "retry-failed",
+                    {"total": 1, "success": 0, "failed": 1},
+                    source="worker",
+                )
+            )
+            self.assertEqual("图片重试失败", messages[-1][0])
+            retry_cards = [
+                card
+                for card in tab.findChildren(SuiteResultCard)
+                if card.job.id == retry_job.id
+            ]
+            self.assertEqual(1, len(retry_cards))
+            self.assertTrue(
+                any(
+                    button.text() == "重试"
+                    for button in retry_cards[0].findChildren(QPushButton)
+                )
+            )
+
+            state.show_history = True
+            tab._load_state(state)
+            with mock.patch.object(
+                tab,
+                "_start_thread",
+                return_value=mock.Mock(),
+            ):
+                self.assertTrue(
+                    tab.start_generation(
+                        state,
+                        specs=[
+                            {
+                                "source_asset_id": source.id,
+                                "job_type": failed_job.job_type,
+                                "prompt": failed_job.prompt,
+                            }
+                        ],
+                        retry_job_id=failed_job.id,
+                    )
+                )
+            self.assertTrue(state.show_history)
+            self.assertTrue(tab.history_button.isChecked())
+            state.worker = None
+            state.thread = None
+            state.generation_run_token = ""
+
+        self.assert_removed(temp_dir)
+
    def test_original_list_expands_without_internal_scrollbars(self):
        with self.make_temp_dir() as temp_dir:
            config = self._config(temp_dir)