FOAM-FSI does not work in Debug mode (index out of range)

[MakisH]

Hello everyone,

When I try to run any tutorial case, fsiFoam crashes in the beginning with an error I do not understand. See the log files here. The error is always in the solid and it has the following form:

FOAM-FSI build: 3.2-5f43c0922374

Patch inner-wall	Traction boundary field: U
	nonLinear set to total Lagrangian
Patch outer-wall	Traction boundary field: U
	nonLinear set to total Lagrangian
Patch inner-wall	Traction boundary field: U
	nonLinear set to total Lagrangian
Patch outer-wall	Traction boundary field: U
	nonLinear set to total Lagrangian
[1] 
[1] 
[1] --> FOAM FATAL ERROR: 
[1] index 84 out of range 0 ... 83
[1] 
[1]     From function UList<T>::checkIndex(const label)
[1]     in file /home/tester/foam/foam-extend-3.2/src/foam/lnInclude/UListI.H at line 124.

This particular error is for the 3dTube tutorial, with preCICE. I simply run ./Allrun or ./Allrun_precice (after I run ./Allclean).

I built foam-extend and FOAM-FSI successfully, the testsuite-fsi returns no errors. The foam-extend (parallel) tutorials run successfully.

Any ideas?

[SvensenSeven]

I think that FOAM-FSI should be in /home/tester/foam/foam-extend-3.2/
If I understood David's instructions right, FOAM-FSI cannot be located outside of foam-extend.

[MakisH]

I think this is not the case here, as fsiFoam was built and starts successfully. I cannot find such an instruction in the Readme. I do set FOAM_INST_DIR correctly, however.

[SvensenSeven]

which OS do you use ? Ubuntu 16.04 ?

[MakisH]

Yes, Ubuntu 16.04 64bit. Am 26.01.2018 6:24 vorm. schrieb "Sergey" <notifications@github.com>:

…

which OS do you use ? Ubuntu 16.04 ? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#382 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEtvQ93-e-1t7JD2pxZMfJFIs6fmqQg3ks5tOWGFgaJpZM4Rskf7> .

[SvensenSeven]

There are some problems during compilation on Ubuntu 16.04.
I have the successful build on a workstation, but only on 14.04.

As I noticed, you use VirtualBox. So, if it is possible, please try to use Ubuntu 14.04.
By the way, on my home computer I also use 16.04, so I will try to compile FOAM-FSI during the next days. Maybe I will be faced with the same issue.

[MakisH]

An interesting observation: This number seems to be the number of faces shared with each other processor (as I see it in log.decomposePar) and is dependent on the number of processes used.

Any idea why the index goes out of range? I assume that it is running for everybody else, so it must be platform-specific.

Things that I also tried already, without seeing any difference:

• Reverting the effects of 2f6a745 (default fvSchemes)
• commenting-out renumberMesh in the Allrun

Some information about my system:

• I use OpenMPI 1.10.2.
• Ubuntu 16.04 64bit on a VirtualMachine (host: also Ubuntu 16.04) with 2 virtual cores.
• My CPU is an Intel Kaby Lake laptop processor with 2*2 virtual cores.
• I use foam-extend 3.2-a4e7bf3b15c0, which compiled without any errors (I have the complete log file if needed, it is 6MB). FOAM-FSI also built without errors at the end. As I describe in Use relative paths to src/include for the consistent solvers #41 (comment), I had to resume building after setting $FOAM_USER_SRC appropriately. I will try to rebuild it from scratch, but I don't expect this to be relevant.

[MakisH]

An update on things I have tried in the meanwhile:
I built it on a different computer (not a VM), again with Ubuntu 16.04. Another user told me that for him it runs normally on Ubuntu 16.04.

I get the same error.

The FOAM-FSI testsuites are running without errors. I also ran a parallel tutorial bundled with foam-extend without any errors (incompressible/simpleFoam/pitzDaily with decompoesPar and mpirun -n 4 simpleFoam -parallel). I also let the foam-extend tutorials/Alltest script run for a while, without any errors.

@davidsblom do you maybe have any idea? Have you seen this kind of error before?

[MakisH]

It looks like FOAM-FSI fails to run in Debug mode.

In the building instructions, it is stated:

With the environment variable WM_COMPILE_OPTION you can switch between debug and release builds. So for a debug build: export WM_COMPILE_OPTION=Debug and for a release build: export WM_COMPILE_OPTION=Opt

For optimal performance of the production run, it is advised to add the NDEBUG compiler flag in order to disable all assertions as follows: sed -i s/"c++OPT = -O3"/"c++OPT = -O3 -DNDEBUG"/g foam-extend-3.2/wmake/rules/linux64Gcc/c++Opt

It looks like at the moment this is not only related to performance and is a "must" instead of a "suggestion". Has anybody managed to make it running in Debug? Maybe a bigger bug is hidden below this.

This was a bit counter-intuitive to find because when one tries to find out why a software doesn't work, they should test in "Debug" mode. If there are no guesses on the cause, I will issue a Pull Request to update the building instructions.