adolc tests are broken

[tjhei]

The MPI and serial Jenkins build shows several (all?) adolc tests failing with

An error occurred in line <1621> of file </jenkins/workspace/dealii-mpi_PR-14568/source/differentiation/ad/ad_drivers.cc> in function
    std::enable_if_t<(dealii::Differentiation::AD::ADNumberTraits<ADNumberType>::type_code == dealii::Differentiation::AD::NumberTypes::adolc_tapeless)> dealii::Differentiation::AD::internal::configure_tapeless_mode(unsigned int) [with ADNumberType = adtl::adouble; std::enable_if_t<(dealii::Differentiation::AD::ADNumberTraits<ADNumberType>::type_code == dealii::Differentiation::AD::NumberTypes::adolc_tapeless)> = void]
The violated condition was: 
    n_live_variables == 0

see #14568 and #14576
and https://ci.tjhei.info/job/dealii-mpi/job/PR-14568/1/testReport/junit/projectroot.tests.adolc.helper_scalar_coupled_3_components_01_2debug/mpirun=1/adolc_helper_scalar_coupled_3_components_01_2_mpirun_1_debug/

I don't know enough to understand what is going on here...

[tamiko]

@tjhei Do you build adolc via candi? For reference, I have (system) AdolC enabled for Ubuntu 20.04 and 22.04: https://cdash.dealii.org/build/588 https://cdash.dealii.org/build/585 and I don't see such fallout :-|

[tamiko]

@tjhei Can I try to reproduce with a docker image? :-)

[tjhei]

Yes,candi: https://github.com/dealii/candi/blob/dealii-9.4/deal.II-toolchain/packages/adolc.package

Can I try to reproduce with a docker image? :-)
Sure, this should work:

dealii/contrib/ci/Jenkinsfile.serial

Line 48 in e4382e0

image 'tjhei/candi:v9.4.1-r1-ubuntu20.04'

dealii/contrib/ci/Jenkinsfile.serial

Lines 98 to 114 in e4382e0

	export TEST_TIME_LIMIT=1200
	export NP=`grep -c ^processor /proc/cpuinfo`
	mkdir -p /home/dealii/build
	cd /home/dealii/build
	cmake -G "Ninja" \
	-D DEAL_II_CXX_FLAGS='-Werror' \
	-D DEAL_II_CXX_FLAGS_DEBUG='-Og' \
	-D DEAL_II_EARLY_DEPRECATIONS=ON \
	-D CMAKE_BUILD_TYPE=Debug \
	-D DEAL_II_WITH_ADOLC=OFF \
	-D DEAL_II_WITH_MPI=OFF \
	-D DEAL_II_UNITY_BUILD=ON \
	$WORKSPACE/
	time ninja -j 10 # 12 gives OOM
	time ninja test # quicktests
	time ninja setup_tests
	time ctest --output-on-failure -j $NP --no-compress-output -T test

[jppelteret]

This failure is triggered by the fact that we have the --enable-traceless_refcounting flag set here. IIRC this option is new to ADOL-C 2.7.2, and I think that most package managers to not activate it by default.

I tried to reproduce this error locally, but since I'm using the docker image I can't be sure of how ADOL-C is being built. I also had a look through the code and do not see anything obviously wrong. So I'll have build it externally to try to debug the issue.

What confuses me is that I am quite certain that I developed this code with the most conservative ADOL-C build options active. Anyway, I'll try to look into this over the holiday period.

[tjhei]

Let me know if I can help somehow.

but since I'm using the docker image I can't be sure of how ADOL-C is being built

This is done using candi. Or what you do mean?