Repository navigation
对弈竞猜更新 - #1848
对弈竞猜更新#1848
Conversation
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
嘿——我发现了 3 个问题
供 AI 代理使用的提示
请处理本次代码审查中的评论：
## 单独评论
### 评论 1
<location path="tasks/FrogBoss/frog_oas.py" line_range="135-171" />
<code_context>
+ for uid in DASHEN_UIDS:
</code_context>
<issue_to_address>
**issue (performance)：** `fetch_predictions` 在下注前会为 26 个 UID 中的每一个执行两次同步 HTTP 请求，但没有总体截止时间或并发机制；在最坏情况下，单个请求的超时会使下注流程阻塞数分钟，导致阵容发生变化或下注结束，而 `do_bet` 还没来得及点击。
**触发条件：** 下注倒计时期间，预测服务响应缓慢或部分不可用时。
**建议修复：** 设置有上限的总体截止时间和/或使用并发请求，并使用截止时间前已收集到的预测结果继续执行，而不是等待所有 UID 完成。
</issue_to_address>
### 评论 2
<location path="tasks/FrogBoss/frog_oas.py" line_range="77-84" />
<code_context>
+ # Unverified newcomers start neutral; verified sources use raw win rate.
++ return correct / total if total else 0.5
+
+ def settle(self, signature, winner):
+ if winner not in ('LEFT', 'RIGHT'):
+ return None
+ # Include already settled records when checking ambiguity: never transfer
+ # an old result to a newer decision with an identical lineup.
+ matches = [e for e in self.events if e.get('kind') == 'decision'
+ and same_lineup(e['signature'], signature)]
+ if len(matches) != 1:
+ self.append('unverified_result', signature=signature, winner=winner,
+ reason='missing_or_ambiguous_decision')
</code_context>
<issue_to_address>
**issue (bug_risk)：** `settle` 会在整个历史记录中搜索阵容匹配的决策，并要求匹配结果恰好只有一个；当同一阵容指纹被用于两场比赛时，后续结果会被记录为 `unverified_result`，且两个决策都不会获得结果，因此对于重复阵容，来源可靠性将无法继续学习。
**触发条件：** 历史中的两场比赛存在符合 20 位 `same_lineup` 阈值的阵容时。
**建议修复：** 持久化并传递与当前下注关联的具体决策 ID 或槽位，然后直接结算该决策，而不是仅根据阵容指纹解析。
</issue_to_address>
### 评论 3
<location path="tasks/FrogBoss/script_task.py" line_range="197-198" />
<code_context>
logger.info('Right win')
else:
- result = True
+ result = None
return result
def get_bilibili(self) -> RuleImage:
</code_context>
<issue_to_address>
**issue (bug_risk)：** `detect` 被标注为返回 `bool`，其文档字符串也描述了布尔结果，但新的兜底逻辑在无法同时看到结果指示器时返回 `None`；依赖文档所述布尔契约的调用方或重写实现会收到第三种值。
**触发条件：** 既未检测到完整的左方获胜图片对，也未检测到完整的右方获胜图片对时。
**建议修复：** 将类型标注和文档字符串改为 `Optional[bool]`，并说明结果未确定的情况；或者保留布尔契约，并单独处理无法确定的检测结果。
</issue_to_address>Sourcery 评估
需要人工审查。 有 3 个发现需要优先处理；当选择新的 Oas 策略时，外部获取的预测结果和持久化的历史记录会决定自动下注的方向。过期或错误的数据源、阵容匹配结果或评分决策可能导致下注错误的一方，并永久损失已下注的游戏货币；回滚代码也无法挽回这些损失。
阻塞性发现：tasks/FrogBoss/frog_oas.py:171、tasks/FrogBoss/frog_oas.py:84、tasks/FrogBoss/script_task.py:198
Original comment in English
Hey - I've found 3 issues
Prompt for AI Agents
Please address the comments from this code review:
## Individual Comments
### Comment 1
<location path="tasks/FrogBoss/frog_oas.py" line_range="135-171" />
<code_context>
+ for uid in DASHEN_UIDS:
</code_context>
<issue_to_address>
**issue (performance):** `fetch_predictions` performs two synchronous HTTP requests for every one of the 26 UIDs before placing the bet, with no overall deadline or concurrency; in the worst case the per-request timeouts block the betting flow for several minutes, so the lineup changes or betting closes before `do_bet` can click.
**Triggers:** When the prediction service is slow or partially unavailable during the betting countdown.
**Suggested fix:** Use a bounded overall deadline and/or concurrent requests, and proceed with the predictions collected before the deadline instead of waiting for every UID.
</issue_to_address>
### Comment 2
<location path="tasks/FrogBoss/frog_oas.py" line_range="77-84" />
<code_context>
+ # Unverified newcomers start neutral; verified sources use raw win rate.
+ return correct / total if total else 0.5
+
+ def settle(self, signature, winner):
+ if winner not in ('LEFT', 'RIGHT'):
+ return None
+ # Include already settled records when checking ambiguity: never transfer
+ # an old result to a newer decision with an identical lineup.
+ matches = [e for e in self.events if e.get('kind') == 'decision'
+ and same_lineup(e['signature'], signature)]
+ if len(matches) != 1:
+ self.append('unverified_result', signature=signature, winner=winner,
+ reason='missing_or_ambiguous_decision')
</code_context>
<issue_to_address>
**issue (bug_risk):** `settle` searches the entire history for decisions with a matching lineup and requires exactly one match; when the same lineup fingerprint is used in two competitions, the later result is recorded as `unverified_result` and neither decision receives an outcome, so source reliability stops learning for repeated lineups.
**Triggers:** When two competitions in the history have lineups within the 20-bit `same_lineup` threshold.
**Suggested fix:** Persist and carry the specific decision ID or slot associated with the active bet, then settle that decision directly rather than resolving it only from the lineup fingerprint.
</issue_to_address>
### Comment 3
<location path="tasks/FrogBoss/script_task.py" line_range="197-198" />
<code_context>
logger.info('Right win')
else:
- result = True
+ result = None
return result
def get_bilibili(self) -> RuleImage:
</code_context>
<issue_to_address>
**issue (bug_risk):** `detect` is annotated as returning `bool` and its docstring describes a boolean result, but the new fallback returns `None` when the result indicators are not both visible; callers or overrides relying on the documented boolean contract receive a third value.
**Triggers:** When neither complete left-win nor complete right-win image pair is detected.
**Suggested fix:** Change the annotation and docstring to `Optional[bool]` and document the unresolved-result case, or preserve the boolean contract and handle unresolved detection separately.
</issue_to_address>Sourcery assessment
Needs a human reviewer. 3 findings to address first, and when the new Oas strategy is selected, externally fetched predictions and persisted history determine which side is automatically wagered on. A stale or incorrect feed, lineup match, or scoring decision can place bets on the wrong side and permanently lose the wagered in-game currency; reverting the code would not recover those losses.
Blocking findings: tasks/FrogBoss/frog_oas.py:171, tasks/FrogBoss/frog_oas.py:84, tasks/FrogBoss/script_task.py:198
| for uid in DASHEN_UIDS: | ||
| feed_id = None | ||
| try: | ||
| response = session.get('https://inf.ds.163.com/v1/web/feed/basic/getSomeOneFeeds', | ||
| params={'feedTypes': '1,2,3,4,6,7,10,11', 'someOneUid': uid}, timeout=(3, 5)) | ||
| response.raise_for_status() | ||
| feeds = response.json().get('result', {}).get('feeds', []) | ||
| if not feeds: | ||
| history.append('fetch', uid=uid, accepted=False, reason='no_feed') | ||
| continue | ||
| feed_id = feeds[0]['id'] | ||
| response = session.get('https://inf.ds.163.com/v1/web/feed/basic/facade', | ||
| params={'feedId': feed_id}, timeout=(3, 5)) | ||
| response.raise_for_status() | ||
| feed = response.json()['result']['feed'] | ||
| content = feed['content'] | ||
| if isinstance(content, str): | ||
| content = json.loads(content) | ||
| body = content['body']['text'] | ||
| raw_time = feed.get('createTime') | ||
| published = None | ||
| try: | ||
| published = datetime.fromtimestamp(float(raw_time) / 1000) | ||
| except (TypeError, ValueError, OverflowError, OSError): | ||
| pass | ||
| side = parse_side(body) | ||
| fresh = published is not None and published.date() == now.date() and published.hour // 2 == now.hour // 2 and published <= now | ||
| accepted = fresh and side is not None and (uid, feed_id) not in used | ||
| history.append('fetch', uid=uid, feed_id=feed_id, create_time=raw_time, | ||
| body=body, side=side, accepted=accepted, | ||
| reason='accepted' if accepted else 'stale_unknown_ambiguous_or_duplicate') | ||
| if accepted: | ||
| predictions.append(dict(uid=uid, side=side, feed_id=feed_id)) | ||
| except (requests.RequestException, ValueError, KeyError, IndexError, TypeError) as exc: | ||
| history.append('fetch', uid=uid, feed_id=feed_id, accepted=False, | ||
| reason='request_or_parse_error', error=str(exc)) | ||
| return predictions |
There was a problem hiding this comment.
issue (performance)： fetch_predictions 在下注前会为 26 个 UID 中的每一个执行两次同步 HTTP 请求，但没有总体截止时间或并发机制；在最坏情况下，单个请求的超时会使下注流程阻塞数分钟，导致阵容发生变化或下注结束，而 do_bet 还没来得及点击。
触发条件： 下注倒计时期间，预测服务响应缓慢或部分不可用时。
建议修复： 设置有上限的总体截止时间和/或使用并发请求，并使用截止时间前已收集到的预测结果继续执行，而不是等待所有 UID 完成。
Original comment in English
issue (performance): fetch_predictions performs two synchronous HTTP requests for every one of the 26 UIDs before placing the bet, with no overall deadline or concurrency; in the worst case the per-request timeouts block the betting flow for several minutes, so the lineup changes or betting closes before do_bet can click.
Triggers: When the prediction service is slow or partially unavailable during the betting countdown.
Suggested fix: Use a bounded overall deadline and/or concurrent requests, and proceed with the predictions collected before the deadline instead of waiting for every UID.
| def settle(self, signature, winner): | ||
| if winner not in ('LEFT', 'RIGHT'): | ||
| return None | ||
| # Include already settled records when checking ambiguity: never transfer | ||
| # an old result to a newer decision with an identical lineup. | ||
| matches = [e for e in self.events if e.get('kind') == 'decision' | ||
| and same_lineup(e['signature'], signature)] | ||
| if len(matches) != 1: |
There was a problem hiding this comment.
issue (bug_risk)： settle 会在整个历史记录中搜索阵容匹配的决策，并要求匹配结果恰好只有一个；当同一阵容指纹被用于两场比赛时，后续结果会被记录为 unverified_result，且两个决策都不会获得结果，因此对于重复阵容，来源可靠性将无法继续学习。
触发条件： 历史中的两场比赛存在符合 20 位 same_lineup 阈值的阵容时。
建议修复： 持久化并传递与当前下注关联的具体决策 ID 或槽位，然后直接结算该决策，而不是仅根据阵容指纹解析。
Original comment in English
issue (bug_risk): settle searches the entire history for decisions with a matching lineup and requires exactly one match; when the same lineup fingerprint is used in two competitions, the later result is recorded as unverified_result and neither decision receives an outcome, so source reliability stops learning for repeated lineups.
Triggers: When two competitions in the history have lineups within the 20-bit same_lineup threshold.
Suggested fix: Persist and carry the specific decision ID or slot associated with the active bet, then settle that decision directly rather than resolving it only from the lineup fingerprint.
| result = None | ||
| return result |
There was a problem hiding this comment.
issue (bug_risk)： detect 被标注为返回 bool，其文档字符串也描述了布尔结果，但新的兜底逻辑在无法同时看到结果指示器时返回 None；依赖文档所述布尔契约的调用方或重写实现会收到第三种值。
触发条件： 既未检测到完整的左方获胜图片对，也未检测到完整的右方获胜图片对时。
建议修复： 将类型标注和文档字符串改为 Optional[bool]，并说明结果未确定的情况；或者保留布尔契约，并单独处理无法确定的检测结果。
Original comment in English
issue (bug_risk): detect is annotated as returning bool and its docstring describes a boolean result, but the new fallback returns None when the result indicators are not both visible; callers or overrides relying on the documented boolean contract receive a third value.
Triggers: When neither complete left-win nor complete right-win image pair is detected.
Suggested fix: Change the annotation and docstring to Optional[bool] and document the unresolved-result case, or preserve the boolean contract and handle unresolved detection separately.
Change Summary
Reconstructed Intent点击此处展开
Observed Constraints点击此处展开
Intent Alignment
Release Risk
Validation Gaps点击此处展开
Warning Firewall blocked 1 domainThe following domain was blocked by the firewall during workflow execution:
To allow these domains, add them to the network:
allowed:
- defaults
- "ab.chatgpt.com"See Network Configuration for more information.
|
Findings
Warning Firewall blocked 1 domainThe following domain was blocked by the firewall during workflow execution:
To allow these domains, add them to the network:
allowed:
- defaults
- "ab.chatgpt.com"See Network Configuration for more information.
|
迁移 runhey/OnmyojiAutoScript PR runhey#1848 的提交 86b2a7e。
2 participants
更新十周年对弈竞猜资源
Sourcery 摘要
更新 FrogBoss 对周年庆资源的下注支持，并引入持久化的 Oas 预测策略。
新功能：
错误修复：
增强功能：
测试：
杂项：
Original summary in English
Summary by Sourcery
Update FrogBoss betting support for the anniversary resources and introduce the persistent Oas prediction strategy.
New Features:
Bug Fixes:
Enhancements:
Tests:
Chores: