Re: [3/3] kernel32/tests: add named pipe test using i/o completion ports

[Austin English]

On Thu, Feb 26, 2009 at 1:43 PM, Vincent Povirk  wrote:
> This has passed for me on all the NT systems I tested (NT4, 2000, XP, and Vista). 9x systems skip all named pipe tests.
>
> It does not pass on Wine. The exact nature of the failure is theoretically too unpredictable to properly mark it as todo. It can't be committed until Wine is fixed. (The test always failed the same way for me, but I believe on Wine there are races and accesses to uninitialized memory.)

It seems AJ put in some other fixes before this, then committed your
patch...however, I'm now getting failures on my (previously nearly
pristine) box. Looks like Michael Stefaniuc is affected as well.

Would you mind taking a look?

http://test.winehq.org/data/566cb8c7a32f6128ae7d05f7fab30db08ba64ea4/#group_...

[3/3] kernel32/tests: add named pipe test using i/o completion ports

[Vincent Povirk]

I don't understand Wine's named pipe or completion implementations,
but I can at least explain what the test is observing.

http://test.winehq.org/data/566cb8c7a32f6128ae7d05f7fab30db08ba64ea4/wine_ae...

ConnectNamedPipe is setting ERROR_ALREADY_CONNECTED, but it's also
notifying the completion. In my testing on Windows, the completion was
never notified in this case (we don't have any tests on
test.winehq.org showing this yet, or I'd link one).

http://test.winehq.org/data/566cb8c7a32f6128ae7d05f7fab30db08ba64ea4/wine_ms...

Here ConnectNamedPipe is setting ERROR_IO_PENDING, the client is
connecting, but the completion is never notified.

Vincent Povirk