Jacek Caban jacek@codeweavers.com writes:
On 08.12.2016 16:47, Alexandre Julliard wrote:
Jacek Caban jacek@codeweavers.com writes:
Signed-off-by: Jacek Caban jacek@codeweavers.com
dlls/kernel32/tests/pipe.c | 74 ++----- dlls/ntdll/file.c | 7 +- dlls/ntdll/tests/file.c | 2 +- server/async.c | 15 ++ server/file.h | 2 + server/named_pipe.c | 499 ++++++++++++++++++++++++++++++++++++++++++--- 6 files changed, 510 insertions(+), 89 deletions(-)
This is causing various random failures during make test. I don't have a good reproducible case, but it seems to randomly break communication with services. For instance I see mountmgr-related errors, or hangs in the msi tests.
I could reproduce it by running services.exe tests in a loop. I found that it's fixed by patches I have in my branch that are meant to improve immediate returns from read/write. On closer look, they not only improve perf, but also fix how we handle errors on ntdll side. Sorry I didn't catch it earlier, I would plan patch series differently. They are only drafts I meant to send as follow ups, so I need to finish them, test a bit more and put earlier in the patch series. I will work on that, but how does it fit release plans?
With rc1 being out today, I'm afraid it's not going to make it into 2.0, it's too big and risky to apply during code freeze. Sorry about that...