1.起因
用Quick-Cocos2d-Community 和服务器联调 mac 上一切正常
windows下在服务器关闭的时候发现一直没有收到socket close 的消息
当时我手上主要是在折腾服务端，其实可以放下不管，因为在要出包的iOS和Android平台都没有问题，不影响线上
但是不方便客户端哥们进行调试，毕竟大部分哥们还是在win下工作
2.初步处理
将quick的framework中的SocketTcp中以下输出日志注释打开
function SocketTCP:_tick(dt)
local __body, __status, __partial = self.tcp:receive("*a") -- read the package body
//打开日志
print("body:", __body, "__status:", __status, "__partial:", __partial)
if __status == STATUS_CLOSED or __status == STATUS_NOT_CONNECTED then
self:close()
if self.isConnected then
self:_onDisconnect()
else
self:_connectFailure()
end
return
end
if (__body and string.len(__body) == 0) or
(__partial and string.len(__partial) == 0)
then return end
if __body and __partial then __body = __body .. __partial end
self:dispatchEvent({name=SocketTCP.EVENT_DATA, data=(__partial or __body), partial=__partial, body=__body})
end
运行 连上服务器 关闭服务器
日志中一直输出__status:timeout
猜测跟设置的超时未0有关 在以下代码中：
function SocketTCP:connect(__host, __port, __retryConnectWhenFailure)
if __host then self.host = __host end
if __port then self.port = __port end
if __retryConnectWhenFailure ~= nil then self.isRetryConnect = __retryConnectWhenFailure end
assert(self.host or self.port, "Host and port are necessary!")
--printInfo("%s.connect(%s, %d)", self.name, self.host, self.port)
if isIpv6(self.host) then
self.tcp = socket.tcp6()
else
self.tcp = socket.tcp()
end
self.tcp:settimeout(0)//此处
if not self:_checkConnect() then
-- remove the old tcp scheduler
if self.connectTimeTickScheduler then scheduler.unscheduleGlobal(self.connectTimeTickScheduler) end
self.connectTimeTickScheduler = scheduler.scheduleUpdateGlobal(handler(self, self._connectTimeTick))
end
end
`
尝试将self.tcp:settimeout(0) 改为self.tcp:settimeout(0.001)
测试windows平台可以收到服务器断开的消息
改成一下代码
function SocketTCP:connect(__host, __port, __retryConnectWhenFailure)
if __host then self.host = __host end
if __port then self.port = __port end
if __retryConnectWhenFailure ~= nil then self.isRetryConnect = __retryConnectWhenFailure end
assert(self.host or self.port, "Host and port are necessary!")
--printInfo("%s.connect(%s, %d)", self.name, self.host, self.port)
if isIpv6(self.host) then
self.tcp = socket.tcp6()
else
self.tcp = socket.tcp()
end
//change by darklost.me 2017-01-06 17:47:21
if device.platform == "windows" then
self.tcp:settimeout(0.0001)
else
self.tcp:settimeout(0)
end
if not self:_checkConnect() then
-- remove the old tcp scheduler
if self.connectTimeTickScheduler then scheduler.unscheduleGlobal(self.connectTimeTickScheduler) end
self.connectTimeTickScheduler = scheduler.scheduleUpdateGlobal(handler(self, self._connectTimeTick))
end
end
3.经一步研究
按道理说通过以上的方法已经绕过这个问题 就可以不管了 毕竟不需要出win32的货 orz
但是生命不朽折腾不止
设置self.tcp:settimeout(0.0001) 以后即使这个时间再小也是会阻塞的 不再是（non-blocking）无阻塞的了
虽然是客户端但还是会有影响和其他平台有差异 影响开发对网络通信问题的判断
通过搜索引擎检索和GitHub上查issue 还真发现遇到相同问题的哥们
按照这哥们的说法 注释掉这行代码就可以了
local data, status, partial = self.tcp:receive("_a")
if __status == STATUS_CLOSED or _status == STATUS_NOT_CONNECTED then
--remote was closed.
end
I use code above to check remote socket is closed，but the status is allways timeout because the code at wsocket.c（socket_waitfd）：
if (timeout_iszero(tm)) return IO_TIMEOUT; / optimize timeout == 0 case */
because my socket is non-blocking，so it allways return timeout here。after delete this line，it's return closed。
讨论的最后也没有给出解决的方案
倒是给我指明了方向
我review了wsocket.c的代码
突然看到下面这段代码
/*-------------------------------------------------------------------------*\
* Receive with timeout
\*-------------------------------------------------------------------------*/
int socket_recv(p_socket ps, char *data, size_t count, size_t *got,
p_timeout tm)
{
int err, prev = IO_DONE;
*got = 0;
if (*ps == SOCKET_INVALID) return IO_CLOSED;
for ( ;; ) {
int taken = recv(*ps, data, (int) count, 0);
if (taken > 0) {
*got = taken;
return IO_DONE;
}
if (taken == 0) return IO_CLOSED;
err = WSAGetLastError();
/* On UDP, a connreset simply means the previous send failed.
* So we try again.
* On TCP, it means our socket is now useless, so the error passes.
* (We will loop again, exiting because the same error will happen) */
if (err != WSAEWOULDBLOCK) {
if (err != WSAECONNRESET || prev == WSAECONNRESET) return err;
prev = err;
}
if ((err = socket_waitfd(ps, WAITFD_R, tm)) != IO_DONE) return err;
}
}
其中的这段注释
/* On UDP, a connreset simply means the previous send failed.
- So we try again.
- On TCP, it means our socket is now useless, so the error passes.
- (We will loop again, exiting because the same error will happen) */
大概意思是tcp 中我们会在循环一次知道遇到相同的错误才退出
但是参考代码逻辑
遇到WSAECONNRESET这个错误一直不会return也不会再循环一次任何一次if (err != WSAECONNRESET || prev == WSAECONNRESET) return err;的判断是prev会一直是IO_DONE
也就是说
if (err != WSAEWOULDBLOCK) {
if (err != WSAECONNRESET || prev == WSAECONNRESET) return err;
prev = err;
//此处少了continue
}
以下方法中也存在相同的问题
int socket_recvfrom(p_socket ps, char *data, size_t count, size_t *got,
SA *addr, socklen_t *len, p_timeout tm)
修改成以下
/*-------------------------------------------------------------------------*\
* Receive with timeout
\*-------------------------------------------------------------------------*/
int socket_recv(p_socket ps, char *data, size_t count, size_t *got,
p_timeout tm)
{
int err, prev = IO_DONE;
*got = 0;
if (*ps == SOCKET_INVALID) return IO_CLOSED;
for ( ;; ) {
int taken = recv(*ps, data, (int) count, 0);
if (taken > 0) {
*got = taken;
return IO_DONE;
}
if (taken == 0) return IO_CLOSED;
err = WSAGetLastError();
/* On UDP, a connreset simply means the previous send failed.
* So we try again.
* On TCP, it means our socket is now useless, so the error passes.
* (We will loop again, exiting because the same error will happen) */
if (err != WSAEWOULDBLOCK) {
if (err != WSAECONNRESET || prev == WSAECONNRESET) return err;
prev = err;
//change by 2017-1-7 01:59:38 darklost.me
//settimeout(0) is always return IO_TIMEOUT when remote server is closed!
//maybe missing continue here?
continue;
}
if ((err = socket_waitfd(ps, WAITFD_R, tm)) != IO_DONE) return err;
}
}
/*-------------------------------------------------------------------------*\
* Recvfrom with timeout
\*-------------------------------------------------------------------------*/
int socket_recvfrom(p_socket ps, char *data, size_t count, size_t *got,
SA *addr, socklen_t *len, p_timeout tm)
{
int err, prev = IO_DONE;
*got = 0;
if (*ps == SOCKET_INVALID) return IO_CLOSED;
for ( ;; ) {
int taken = recvfrom(*ps, data, (int) count, 0, addr, len);
if (taken > 0) {
*got = taken;
return IO_DONE;
}
if (taken == 0) return IO_CLOSED;
err = WSAGetLastError();
/* On UDP, a connreset simply means the previous send failed.
* So we try again.
* On TCP, it means our socket is now useless, so the error passes.
* (We will loop again, exiting because the same error will happen) */
if (err != WSAEWOULDBLOCK) {
if (err != WSAECONNRESET || prev == WSAECONNRESET) return err;
prev = err;
//change by 2017-1-7 01:59:38 darklost.me
//settimeout(0) is always return IO_TIMEOUT when remote server is closed!
//maybe missing continue here?
continue;
}
if ((err = socket_waitfd(ps, WAITFD_R, tm)) != IO_DONE) return err;
}
}
重新编译Quick模拟器 测试可行 没有详细测试，不是是否会存在其他问题orz
在上面那个issue中添加一个 comment
同时新建了一个issue
4.暂告一段落
○|￣|_！！！~