Hi Ismael, have a look at http://test.winehq.org/data/tests/dplayx:dplayx.html Since your commits on the 4th of Aug, the dplayx tests have been hanging. This is getting in the way of my regression testing. Can you have a look? Thanks, Dan
On 8/10/08, Dan Kegel dank@kegel.com wrote:
Hi Ismael, have a look at http://test.winehq.org/data/tests/dplayx:dplayx.html Since your commits on the 4th of Aug, the dplayx tests have been hanging. This is getting in the way of my regression testing. Can you have a look? Thanks, Dan
Are they really hanging or taking too long? Dplay is quite slow and the tests are intensive so they take a lot of time. They take 7 minutes in my XP and some seconds in wine.
Anyway I'll try to see if there's any way of speeding them up, I'm sorry for the hassle. Maybe the best solution is to remove dplay tests until I've finished implementing the library, as right now they are useful only for me, and a problem for the rest.
Cheers Ismael
On Sun, Aug 10, 2008 at 10:19 AM, Ismael Barros razielmine@gmail.com wrote:
http://test.winehq.org/data/tests/dplayx:dplayx.html Since your commits on the 4th of Aug, the dplayx tests have been hanging.
Are they really hanging or taking too long? Dplay is quite slow and the tests are intensive so they take a lot of time. They take 7 minutes in my XP and some seconds in wine.
7 minutes is unreasonable. winetest has a timeout of 120 seconds. You'd better shoot for having them run in under 60 seconds on XP.
In my case, they really were hanging, I waited 20 minutes on Wine.
Anyway I'll try to see if there's any way of speeding them up, I'm sorry for the hassle. Maybe the best solution is to remove dplay tests until I've finished implementing the library, as right now they are useful only for me, and a problem for the rest.
That could well be. - Dan
On 8/10/08, Dan Kegel dank@kege.com wrote:
In my case, they really were hanging, I waited 20 minutes on Wine.
That's really weird, in wine most of the tests are skipped, and they take around 5 seconds on my machine with the latest git head. Could you provide a log with WINEDEBUG=+dplay?
2008/8/10 Ismael Barros razielmine@gmail.com:
On 8/10/08, Dan Kegel dank@kegel.com wrote:
Hi Ismael, have a look at http://test.winehq.org/data/tests/dplayx:dplayx.html Since your commits on the 4th of Aug, the dplayx tests have been hanging. This is getting in the way of my regression testing. Can you have a look? Thanks, Dan
Are they really hanging or taking too long? Dplay is quite slow and the tests are intensive so they take a lot of time. They take 7 minutes in my XP and some seconds in wine.
Anyway I'll try to see if there's any way of speeding them up, I'm sorry for the hassle.
Taking 7 minutes is too long, even if it is only on Windows. Perhaps you could run each test that depends on network latency in a separate thread. This could cause problems with the test output being printed in a weird order as several tests will be running at once, but the number of failures should typically be low and it should the tests to complete faster.
Maybe the best solution is to remove dplay tests until I've finished implementing the library, as right now they are useful only for me, and a problem for the rest.
Please keep them - having the tests running reliably on Windows before implementing the functions they test is very useful as it ensures that no one is going to introduce obviously incorrect code when implementing the functions.
On Sunday 10 August 2008 18:34:02 Dan Kegel wrote:
http://test.winehq.org/data/tests/dplayx:dplayx.html Since your commits on the 4th of Aug, the dplayx tests have been hanging. This is getting in the way of my regression testing.
I've toyed with the tests a little and the following tests took over 30s to finish:
test_Open() 49s test_EnumSessions() 180s test_CreatePlayer() 32s test_GetPlayerAccount() 63s, also 1 test failure on XP
A bunch of other tests took > 20s, and test_GetPlayerAddress() had 7 failures. Incidently, all tests that took over 20 seconds took around 23 seconds, which smells like some fixed timeout.
Ismael, can you look into the really slow tests and check what subtest is taking most of the time?
Cheers, Kai
On Monday 11 August 2008 17:14:52 Kai Blin wrote:
test_Open() 49s test_EnumSessions() 180s test_CreatePlayer() 32s test_GetPlayerAccount() 63s, also 1 test failure on XP
Skipping these four tests, the test completes in two minutes on my box, which still is a lot of time. They probably need to be split up more.
Cheers, Kai
On 8/11/08, Kai Blin kai.blin@gmail.com wrote:
On Sunday 10 August 2008 18:34:02 Dan Kegel wrote:
http://test.winehq.org/data/tests/dplayx:dplayx.html Since your commits on the 4th of Aug, the dplayx tests have been hanging. This is getting in the way of my regression testing.
I've toyed with the tests a little and the following tests took over 30s to finish:
test_Open() 49s test_EnumSessions() 180s test_CreatePlayer() 32s test_GetPlayerAccount() 63s, also 1 test failure on XP
A bunch of other tests took > 20s, and test_GetPlayerAddress() had 7 failures. Incidently, all tests that took over 20 seconds took around 23 seconds, which smells like some fixed timeout.
That's probably the main problem, as there's no network latency (all the messages are sent to localhost) or cpu intensive operation. Actually the thread stuff is a very good idea, I'll take a look. I already tried changing the default timeout for EnumSessions, and on XP the whole test set changed from 7 to 4 minutes, but with a lot of failures.
And thanks for the info, what did you use to time the tests? Also, could you send me the log of the failed tests? They all work on my XP.
On Monday 11 August 2008 18:17:58 Ismael Barros wrote:
On 8/11/08, Kai Blin kai.blin@gmail.com wrote:
On Sunday 10 August 2008 18:34:02 Dan Kegel wrote:
http://test.winehq.org/data/tests/dplayx:dplayx.html Since your commits on the 4th of Aug, the dplayx tests have been hanging. This is getting in the way of my regression testing.
I've toyed with the tests a little and the following tests took over 30s to finish:
test_Open() 49s test_EnumSessions() 180s test_CreatePlayer() 32s test_GetPlayerAccount() 63s, also 1 test failure on XP
A bunch of other tests took > 20s, and test_GetPlayerAddress() had 7 failures. Incidently, all tests that took over 20 seconds took around 23 seconds, which smells like some fixed timeout.
That's probably the main problem, as there's no network latency (all the messages are sent to localhost) or cpu intensive operation. Actually the thread stuff is a very good idea, I'll take a look. I already tried changing the default timeout for EnumSessions, and on XP the whole test set changed from 7 to 4 minutes, but with a lot of failures.
What kind of failures?
And thanks for the info, what did you use to time the tests? Also, could you send me the log of the failed tests? They all work on my XP.
Henri pointed me at "timeit" from the Win2k3 ressource kit at http://www.microsoft.com/downloads/details.aspx?familyid=9d467a69-57ff-4ae7-...
I just ran one test at a time and noted the time the test ran. Attached is the output of the unmodified dplayx tests, running on a vanilla WinXP ProfessionalN SP2 box, without any extra dx runtime installed.
Cheers, Kai
On 8/11/08, Kai Blin kai.blin@gmail.com wrote:
On Monday 11 August 2008 18:17:58 Ismael Barros wrote:
That's probably the main problem, as there's no network latency (all the messages are sent to localhost) or cpu intensive operation. Actually the thread stuff is a very good idea, I'll take a look. I already tried changing the default timeout for EnumSessions, and on XP the whole test set changed from 7 to 4 minutes, but with a lot of failures.
What kind of failures?
A lot of tests failed for no obvious reason. I can start trying bigger timeouts to see if all the tests start working again, but it's difficult to make sure that'll work on every system...
And thanks for the info, what did you use to time the tests? Also, could you send me the log of the failed tests? They all work on my XP.
Henri pointed me at "timeit" from the Win2k3 ressource kit at http://www.microsoft.com/downloads/details.aspx?familyid=9d467a69-57ff-4ae7-...
I just ran one test at a time and noted the time the test ran. Attached is the
Godsend, I've been looking for a similar app but fallbacked to my phone's chronometer.
output of the unmodified dplayx tests, running on a vanilla WinXP ProfessionalN SP2 box, without any extra dx runtime installed.
My system is the same but with debug dx libraries. I'll probably remove those tests, they are too intensive (checking the size of received messages and suchlike), and it's funny that the GetPlayerAccount() takes a minute, it's actually a stub because that function only works with secure sessions and I still didn't manage to get them working (and they probably aren't even important for the 99% of the games)
On 8/11/08, Ismael Barros razielmine@gmail.com wrote:
Actually the thread stuff is a very good idea, I'll take a look.
I tried launching each tests in a new thread, but a lot of tests failures arise, sometimes even with segfaults or deadlocks. Looks like the original implementation of dplay doesn't like concurrency too much.
The good news are that playing with the timeout value of EnumSessions I already managed to cut the time of some tests in half.
-
Dan Kegel -
Ismael Barros -
Kai Blin -
Rob Shearman