fix: 修复退出串流后客户端显示主机离线的问题

[qiin2333]

根因分析：

1. 僵尸广播(zombie broadcast)：延迟到达的 ABR HTTP 请求在 session 结束后 触发 change_dynamic_param_for_client()，该函数调用 broadcast_shared.ref() 创建了一个临时的广播上下文，随后 end_broadcast() 中的 control_thread.join() 阻塞了 HTTPS handler 线程，导致 HTTPS 服务器 无法接受新连接，客户端显示主机离线。

2. VDD 模式下初始拓扑错误：apply_config 在 VDD 模式下调用 get_current_topology_metadata() 获取的是 VDD 启用后的当前拓扑 （只有 VDD 设备），忽略了 pre_saved_initial_topology（物理显示器拓扑）。 恢复时 remove_vdd_from_topology() 将初始拓扑清空，物理显示器无法 被重新启用。

修复：

• stream.cpp: 在 change_dynamic_param_for_client 中添加 has_ref() 前置 检查，无活跃 session 时不触发僵尸广播
• settings.cpp: VDD 模式下使用 pre_saved_initial_topology 替换 get_current_topology_metadata 返回的 VDD-only 初始拓扑

[coderabbitai]

Summary by CodeRabbit

Bug Fixes

• 虚拟显示设备拓扑恢复优化 - 改进了VDD模式下显示配置的拓扑数据处理，确保物理显示的初始配置在还原操作时得到正确保留。

• 流媒体参数更新稳定性提升 - 优化了动态参数变更逻辑，避免在没有活跃会话时执行不必要的广播控制操作。

概述

两个模块添加了防御性检查以优化特定场景。VDD模式中的拓扑处理现在有条件地恢复预存的初始拓扑元数据，防止物理显示器配置丢失。流广播代码新增早期退出检查，避免在无活跃会话时进行不必要处理。

变更

模块 / 文件	变更摘要
显示设备拓扑管理 src/platform/windows/display_device/settings.cpp	VDD模式下的apply_config函数在获取当前拓扑元数据后，现在会条件性地使用预保存的初始拓扑覆盖topology_result->pair.initial。防止还原过程中remove_vdd_from_topology清除物理显示器的初始拓扑配置。
流广播参数管理 src/stream.cpp	stream::change_dynamic_param_for_client函数开头新增broadcast_shared.has_ref()检查，当无活跃广播引用时立即返回false。避免在无现有会话时尝试获取广播上下文和启动控制流程。

代码审查工作量评估

🎯 2 (简单) | ⏱️ ~10 分钟

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	拉取请求的标题准确反映了主要变更内容——修复客户端在退出串流后显示主机离线的问题，这正是本次代码变更的核心目标。
Description check	✅ Passed	拉取请求的描述详细说明了两个根本原因（僵尸广播和VDD模式下的拓扑问题）以及相应的修复措施，与代码变更内容完全相关。

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/zombie-broadcast-and-vdd-topology

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

🧹 Nitpick comments (1)

src/stream.cpp (1)

3055-3060: TOCTOU 竞态条件仍存在，虽改善常见场景但未完全消除问题

此修复有效改善了主要问题：在明确没有活跃引用时避免触发僵尸广播。然而，has_ref() 和 ref() 各自独立加锁，存在竞态窗口：线程 A 调用 has_ref() 看到 _count > 0 后，线程 B 完全释放引用（_count = 0，调用 end_broadcast），随后线程 A 调用 ref() 时 _count == 0，会再次调用 _construct()（即 start_broadcast），重新触发启停周期。

虽然此竞态窗口很小且实际触发概率低（特别是客户端发送动态参数变更时通常有活跃会话），但从线程安全角度讲仍不完美。建议在 shared_t 类中添加原子的 try_ref() 方法，仅当已有活跃引用时才增加计数，这样可完全消除竞态条件。

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/stream.cpp` around lines 3055 - 3060, The current check using
broadcast_shared.has_ref() followed by broadcast_shared.ref() has a TOCTOU race
where has_ref() and ref() lock separately allowing _count to drop to zero
between calls and re-trigger start_broadcast/end_broadcast; add an atomic
try_ref() method to shared_t that increments the reference count only if it is
already >0 and returns success/failure, then replace the has_ref()+ref() pattern
in the stream logic (where broadcast_shared is used) with a single call to
try_ref() to ensure no new reference is created when the count reached zero and
eliminate the race that recreates the broadcast cycle.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@src/stream.cpp`:
- Around line 3055-3060: The current check using broadcast_shared.has_ref()
followed by broadcast_shared.ref() has a TOCTOU race where has_ref() and ref()
lock separately allowing _count to drop to zero between calls and re-trigger
start_broadcast/end_broadcast; add an atomic try_ref() method to shared_t that
increments the reference count only if it is already >0 and returns
success/failure, then replace the has_ref()+ref() pattern in the stream logic
(where broadcast_shared is used) with a single call to try_ref() to ensure no
new reference is created when the count reached zero and eliminate the race that
recreates the broadcast cycle.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: f2d78b1b-3999-4b19-9d0b-f52e7bf91065

📥 Commits

Reviewing files that changed from the base of the PR and between f520396 and aafa965.

📒 Files selected for processing (2)

• src/platform/windows/display_device/settings.cpp
• src/stream.cpp

📜 Review details

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

• GitHub Check: Windows

🧰 Additional context used

📓 Path-based instructions (2)

src/**/*.{cpp,c,h}

⚙️ CodeRabbit configuration file

src/**/*.{cpp,c,h}: Sunshine 核心 C++ 源码，自托管游戏串流服务器。审查要点：内存安全、 线程安全、RAII 资源管理、安全漏洞。注意预处理宏控制的平台相关代码。

Files:

• src/platform/windows/display_device/settings.cpp
• src/stream.cpp

src/platform/**

⚙️ CodeRabbit configuration file

src/platform/**: 平台抽象层代码（Windows/Linux/macOS）。确保各平台实现一致， 注意 Windows API 调用的错误处理和资源释放。

Files:

• src/platform/windows/display_device/settings.cpp

🔇 Additional comments (1)

src/platform/windows/display_device/settings.cpp (1)

797-805: VDD 初始拓扑覆盖逻辑实现正确。

这里在 pre_saved_initial_topology 有效且非空时覆盖 pair.initial，可避免将 VDD-only 拓扑误持久化为初始拓扑，修复恢复阶段物理显示器无法重新启用的问题。