-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/18893/
-----------------------------------------------------------
(Updated March 10, 2014, 4:04 p.m.)
Review request for qpid and Rafael Schloming.
Changes
-------
fixes 1,2, and 4. driver.c does not contain any use of selectors. So I will
have to rework it to do that before it can be OS agnostic. I'm not sure that
will be ready later today for the first RC. I'm also unsure how to test the
change now that messenger doesn't use it.
Bugs: PROTON-529
https://issues.apache.org/jira/browse/PROTON-529
Repository: qpid
Description
-------
This patch is a minimal port of the existing Windows code base to incorporate
the refactored IO and selector mechanism of PROTON-525. It does not
incorporate any changes to how the sockets and "pipes" operate but some thought
was given to accommodating completion ports right after 0.7.
Although there was a clear attempt in PROTON-525 to provide rich abstractions
that would allow Posix and Windows implementations to optimize their low level
IO for each OS, awkwardness remains.
Windows sockets are outside the regular IO system, for a long time a bolt on
third party implementation. For Posix, they are an extension of regular file
descriptors and things like close(socket_fd) makes sense.
This short term Windows fix hides the above. Because the "sockets only"
select() call in the Windows api was the easiest to work with initially, it was
necessary to fake a socket-based pipe for the control port. Since everything
is sockets anyway, we can mix the control fd with other fds. If we used
Windows pipes, we couldn't. But we definitely don't want to use select() long
term and we may not want to use fake socket pipes, or even any pipes at all.
Note that the pipe control mechanism is not needed on Windows if you use
completion ports and is not used in the Qpid C++ Windows implementation.
Even though it is probable that the existing new api could be made to work
without penalizing Windows significantly, I would prefer to stop mingling
sockets with non socket io objects in proton.
Perhaps some interrupt interface could be added to the selector, implemented
with pipes on (some) Posix systems and optimized in other ways on other
platforms.
Diffs (updated)
-----
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/CMakeLists.txt
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/include/proton/cproton.i
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/include/proton/io.h
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/include/proton/selectable.h
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/messenger/messenger.c
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/posix/driver.c
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/posix/io.c
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/selectable.h
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/selectable.c
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/ssl/ssl-internal.h
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/ssl/ssl_stub.c
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/windows/driver.c
1575326
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/windows/io.c
PRE-CREATION
http://svn.apache.org/repos/asf/qpid/proton/trunk/proton-c/src/windows/selector.c
PRE-CREATION
Diff: https://reviews.apache.org/r/18893/diff/
Testing
-------
rhel 6, Win7, all 64 bit for now
Thanks,
Cliff Jansen
