Cleaning up the dashboard

dzenanz · December 12, 2017, 6:49pm

I reported the dynamic analysis reporting failure on CDash’s GitHub here.

We can use this topic for other dashboard-related discussion.

blowekamp · December 12, 2017, 9:21pm

Yes! The dashboard does need some attention!

I address a couple of the hidden Valgrind failure here and here. They were both timeout related. The remaining one, the itkDiscreteHessianGaussianImageFunctionTest, is also timing out. What is off is that in my CTest script I am setting CTEST_TEST_TIMEOUT to 5000, but when the tests are being run (and in the cache) it appears as 1200. I found the following in ITK:

github.com

InsightSoftwareConsortium/ITK/blob/ed8fbff0e6b05d25c1f569674dea2ddbb7163613/Modules/ThirdParty/HDF5/src/itkhdf5/CMakeLists.txt#L657-L659


set (DART_TESTING_TIMEOUT 1200
    CACHE INTEGER
    "Timeout in seconds for each test (default 1200=20minutes)"

I wonder if the “DART” value is clobbering the CTEST value?
This change could be the reason these timeout have recently started to show.

blowekamp · December 13, 2017, 4:19pm

The dashboard looks a lot better to day! Thanks to all those who contributed

So when is the next RC?

fbudin · December 13, 2017, 5:54pm

Today I want to have a look at why the Python wrapping only works on
Linux first though…

dzenanz · December 13, 2017, 6:00pm

Ideally, we want to merge this before tagging too.

jhlegarreta · December 13, 2017, 6:10pm

These patches to solve the valgrind defect notifications whose source was a mystery for a long time are definitely awesome !

As for the itkDiscreteHessianGaussianImageFunctionTest, I’ve tried the gerrit patch:
http://review.source.kitware.com/#/c/22799/1
locally from a fresh master and now it turns out that the 220 s threshold that I chose after some trial and error does not hold at every attempt.

I’ve tried several times with other timeouts (250, 500) and the test passes consistently reporting an execution time of around 213 s. So less than 220 s…

So I’m not sure whether Brad is correct: that the DART_TESTING_TIMEOUT flag reducing the timeouts from 1500 to 1200 by Modules/ThirdParty/HDF5/src/itkhdf5/CMakeLists.txt may be affecting this.

But neither am I sure about why being 213 less than 220 the latter seems no longer to be valid.

The only solution I can think of is printing the variable in CMake with a message to ensure that the variable is changing the timeout for this or other tests, and setting the timer in HDF5 to 1500 and running valgrind on some Linux box anyone may have to see whether that causes the issue.

My environment:
Win 7 64 bit; 2.93 GHz; 16 GB RAM; MSVC 2010

JON HAITZ