non-paged pool memory leak in win_iocp_io_service

[Uri Avni <uri@…>]

This program takes 2GB of non-paged pool memory after several seconds:

boost::asio::io_service io;
while (1) 
{
 io.run();
 io.reset();
}

When outstanding_work_==0, calls to either run, run_one, poll or poll_one leak a non-paged pool memory, caused by the PostQueuedCompletionStatus(iocp_.handle, 0, 0, 0) call in win_iocp_io_service::stop()

[chris_kohlhoff]

(In [76432]) Fix non-paged pool "leak" on Windows when io_service is repeatedly run without anything to do. Refs #6321.

[chris_kohlhoff]

(In [76516]) Merge from trunk:

• Chrono support.

• Added object_handle support.

• Need to enable the basic_handle class when object_handle is supported.

• Update copyright notices.

• Fix MSVC "performance warning".

• Fix for NetBSD. Fixes #6098.

• Fix regression in buffered_write_stream. Fixes #6310.

• Fix deadlock on Mac OS X. Fixes #6275.

• On linux, connect can return EAGAIN in certain circumstances. Remap to another error so that it doesn't look like a non-blocking operation. Fixes #6048.

• Fix non-paged pool "leak" on Windows when io_service is repeatedly run without anything to do. Fixes #6321.

• Disable object_handle on Windows CE.

• Add extra include required for OVERLAPPED struct.

• Fix doxygen comments.

• Update documentation.

• Add missing class.

• Update copyright year.