Sad state of the dashboard

Engineering
1

Hi all,

Almost all the Mac builds on the dashboard are red, and have been for weeks now.

Mine have been red since 2018-03-08. They used to build. Isn’t the idea that if one commits something one must watch the dashboards and fix any fallout?

Cheers,

Sean

2

2 out of 3 builds in the nightly expected section are fully successful. 1 2

What are the errors you are getting? Can you submit an experimental build? ctest -D Experimental

3

2 out of 3 builds in the nightly expected section are fully successful.
1 [2](https://
open.cdash.org/buildSummary.php?buildid=5308930)

Those 2 are practically the only non-red Mac builds, scroll further down…

What are the errors you are getting? Can you submit an experimental
build? ctest -D Experimental

I submit several nightly builds already, don’t see the point in your request to submit yet another, it’ll be red too.

Cheers,

4

I think this patch should help with some of the compilation errors on your build:
http://review.source.kitware.com/#/c/23277/

5

I didn’t realize that RogueResearch* are your machines, sorry.

6

Ah. I also keep forgetting that this ‘discourse’ thing strips both signatures and ‘from’ address, so yeah, the association was maybe difficult to discern.

7

Yes, we should do better with our dashboard builds and get back to a green state.

@seanm , I just merged a patch that should address your build errors.

8

That seemed to do it, thanks.

Could we move all the RogueResearch dashboards into Nightly Expected? It seems those are the ones people pay attention to.

Sean

1 Like
9

Yes, let’s do that. First, let’s get them green, though, in the spirit of ‘get it green, keep it green’ ;-).

Could you please add ITK_USE_KWSTYLE to OFF in the ctest configuration since we are having trouble suppressing the boost warnings?

Thanks,
Matt

10

That seemed to work. A few are now fully green, can they be moved now at least?

Sean

11

Fantastic!

The builds have been moved to the Expected Nightly section. Thank you for contributing them!

12

“Mac10.9-clang-dbg-x86_64-static” is also green.

“Mac10.10-clang-dbg-x86_64-static” you can remove entirely, it was renamed.

“Mac10.11-AppleClang-rel-x86_64-static” has just 2 test failures, OpenCL related. It seems only my bots even run these test at all:

https://open.cdash.org/testSummary.php?project=2&name=itkGPUBinaryThresholdImageFilterTest2D&date=2018-03-27

Sean

13

Moved.

Removed.

Fixes would be welcome.

Unfortunately, the GPU support never developed traction. A brighter future could be possible with a better abstraction like ArrayFire or VTK-M instead of OpenCL.

14

I’m afraid I don’t have time to work on the OpenCL tests. But I really want my dashboards in ‘nightly expected’. Do you want to move that one there anyway? Or shall I disable/suppress the OpenCL test first?

15

Please disable the OpenCL test for now.

16

Done. And now “Mac10.11-AppleClang-rel-x86_64-static” is green.

“Mac10.13-AppleClang-dbg-x86_64-ASanUBSan” had ~200 test failures, but I have it down to 3 today by removing some UBSan flags. The 3 remaining failures are all from ASan, and are not likely to be false positives. I looked at each, but none struck me as trivial. Anyone familiar with the code exercised by these tests:

itkHDF5ImageIOTest
itkMeshTest
netlib_test_integral

For each, ASan catches an overread of a stack-based buffer.

Sean

17

Excellent!

Mac10.11-AppleClang-rel-x86_64-static has been moved.