@bli22ard 实现在这里，就是很简单的 Arc 和 clone
```rust
/// Owned read half of a [`TcpStream`], created by [`into_split`].
///
/// Reading from an `OwnedReadHalf` is usually done using the convenience methods found
/// on the [`AsyncReadExt`] trait.
///
/// [`TcpStream`]: TcpStream
/// [`into_split`]: TcpStream::into_split()
/// [`AsyncReadExt`]: trait@crate::io::AsyncReadExt
#[derive(Debug)]
pub struct OwnedReadHalf {
inner: Arc<TcpStream>,
}

/// Owned write half of a [`TcpStream`], created by [`into_split`].
///
/// Note that in the [`AsyncWrite`] implementation of this type, [`poll_shutdown`] will
/// shut down the TCP stream in the write direction. Dropping the write half
/// will also shut down the write half of the TCP stream.
///
/// Writing to an `OwnedWriteHalf` is usually done using the convenience methods found
/// on the [`AsyncWriteExt`] trait.
///
/// [`TcpStream`]: TcpStream
/// [`into_split`]: TcpStream::into_split()
/// [`AsyncWrite`]: trait@crate::io::AsyncWrite
/// [`poll_shutdown`]: fn@crate::io::AsyncWrite::poll_shutdown
/// [`AsyncWriteExt`]: trait@crate::io::AsyncWriteExt
#[derive(Debug)]
pub struct OwnedWriteHalf {
inner: Arc<TcpStream>,
shutdown_on_drop: bool,
}

pub(crate) fn split_owned(stream: TcpStream) -> (OwnedReadHalf, OwnedWriteHalf) {
let arc = Arc::new(stream);
let read = OwnedReadHalf {
inner: Arc::clone(&arc),
};
let write = OwnedWriteHalf {
inner: arc,
shutdown_on_drop: true,
};
(read, write)
}
```

使用 into_split 可以分离读写
https://docs.rs/tokio/latest/tokio/net/struct.TcpStream.html#method.into_split
可以参考这里
https://github.com/FutureOrientedGB/gbt_sip_server/blob/09628a99b89038b65257ad3ea3fc29c45337a01a/src/sip/server.rs#L123

你点击控件成功之后，给 A 窗口发送最小化消息 SendMessage(HWND, SC_MINIMIZE)就可以了吧

@capric 6, 8, 10, A4 的相纸都支持，调整进纸仓的卡扣就行

我家用惠普 4160L 喷墨自动双面，打 7 寸彩色照片或者 1 寸证件都很不能，咸鱼上收的，400-500 块，连供墨仓，墨水买兼容的，很便宜

@clevertension 你这个 rtmp 是实时流吗，如果是实时流，一般只有 25/30/60 几种，不会有 7600 那么高，你统计出来 7600 fps ，应该绝大多数 frame 都是 None 。另外可以用 multiprocessing 和 queue 分发。

@capric 还需要重新编译 opencv ，https://blog.csdn.net/jiexijihe945/article/details/125084488

cv2.VideoCapture 的 read 是调用 ffmpeg 库实现的，涉及到几个耗时操作，解析 rtmp 封装，H.264/H.264 视频解码(最占用 CPU)，yuv 颜色空间转换成 rgb24 ，打包成 numpy 数组。如果你是在 pc 在操作，需要自己重新编译 ffmpeg ，启用 nvdec/qsv 硬件加速，这个解码负载就可以移动到 GPU 上来做，CPU 占用就很低了。

@zhixiz1007 会降，下午问了银行，每年 1 月 1 日调整

Windows 10 企业版 LTSC 21H2 (19044.3448)直接安装 github release 2.0.0 之后更新.wslconfig 后重启，再打开现有 linux
wsl: 不支持镜像网络模式，正在回退到 NAT 网络
wsl: Hyper-V 防火墙不受支持
wsl: DNS 隧道不受支持

京阿尼的紫罗兰永恒花园，吹响吧上低音号， 轻音少女，凉宫春日系列
风灵玉秀，某科学的电磁炮， 装甲重拳，灵笼，雾山五行
一人之下，少年歌行，金光布袋戏

win32 api 采集摄像头数据, mf 编码成 mp4(h264/h265)本地存储
如果需要远程查看, 文件分小片, 一个 http 服务启 hls, 或者实时上传到去服务器

windows 用 pipe，linux 用 sock

改输出名字也是可以的
if(CMAKE_SIZEOF_VOID_P EQUAL 8)
set(suffix "64")
else()
set(suffix "32")
endif()

# set_target_properties(${name} PROPERTIES OUTPUT_NAME "dense_mapping${suffix}")

set_target_properties(
dense_mapping
PROPERTIES
RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/../../rundir"
)

${CMAKE_BINARY_DIR} 是默认输出位置

src = 'ad=0 cd=0 rcode=ok qdcount=1 ancount=0 nscount=0 arcount=1 src=232.190.252.222 dst=45.80.170.1'
dst = dict(parts.split('=') for parts in src.split())
print(dst.get('ad'))