Uwe Bonnes wrote:
I hope people don't think of me as a PITA, but here again my RFT (Request for Tests):
Rob, can you please add a testcase in dlls/kernel/tests/pipe.c showing the usage and the functionality of your patch?
I presume you are requesting this for one of two reasons: making sure that overlapped I/O on named pipes works now and making sure it works in the future. Just to put your mind at ease about the first, I put it through a very heavy work-out by converting rpcrt4 to use overlapped I/O on named pipes instead of synchronous I/O. This would have provided a nice test case for the functionality, but I'm not going to submit the work because it noticably slows down RPCs. However, the reason I did it was to achieve closer semantics to Windows' named pipes to find out the reason for a short read on a pipe. This may be of interest to you because it appears to be exactly the same problem you were reporting in another message. If you could force the program to use asynchronous I/O then it should solve that problem.
So, I'll write tests for overlapped I/O on named pipes when I next have some free time, which will probably be at the weekend.