Changes between and of Version 5Version 6Ticket #5562, comment 45


Ignore:
Timestamp:
06/17/2017 07:28:53 PM (2 years ago)
Author:
Bear
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #5562, comment 45

    v5 v6  
    11Thanks BMtthws. It is really a bug. Only adding the write_pending can solve the problem very safely. The second send command will then be sent until our program gets the callback of the first send command from the iocp.
    22
    3 So the self.offset in 17.5 still destroys the order of sending. In fact, the IOCP can ensure the enqueue order. We don't even need write_pending, self.offset, _cbWrite. It is nealy 100 percent that all the bytes will be sent[I listed the explanation below]
     3So the self.offset in 17.5 still destroys the order of sending. In fact, the IOCP can ensure the enqueue order. We don't even need write_pending, self.offset, _cbWrite. It is nealy 100 percent that all the bytes will be sent[I listed the explanation below]. So if we get a error_io_pending, then set the write_pending=True.
    44
    55I use the IOCP reactor to avoid the limited file descriptor so the patch is very good.