Apparent error in build 8.4-0ppa0-202301022232 [SOLVED]

[teepeeaye]

Hi Elmer users.

Today I - or rather the system - applied an update to Elmer causing it to not work anymore.

Code: Select all

Upgraded the following packages:
elmerfem-csc (8.4-0ppa0-202106021054~6dfb4824~ubuntu16.04.1) to 8.4-0ppa0-202301022232~c7b564f66~ubuntu16.04.1

So essentially same version but a more recent build.

Starting ElmerGUI can read the files, display geometry and mesh, remesh, apply equations and boundary conditions, but when I hit solve I get an error indicating that ElmerSolver is faulty. I get this message:

Code: Select all

*** Error in `ElmerSolver': double free or corruption (fasttop): 0x0000000000addd70 ***

Apparently the ElmerSolver of the build is sick.
Is it possible to remedy this or possible to download/reinstall the previous build (8.4-0ppa0-202106021054~6dfb4824~ubuntu16.04.1) as this actually worked well?

[kevinarden]

Ubuntu 16?

[teepeeaye]

Yes.

[kevinarden]

Another user reported the same. I did a fres install on Ubuntu 16 and Elmer would not work. I believe it is because Elmer is compiled on Fortran 7 and Ubuntu 16 is still on Fortran 4. Have not found a way to get it to work the OS seems too old,

[teepeeaye]

Hi kevinarden,
Thanks for reading and responding. I also saw that some other users have experienced similarly. Even after posting this thread I found some more references. The background for my hope and optimism in finding a solution is based on these facts:

• Same version (8.4) - although a sligtly older build date - has run flawlessly on the system until the unfortunate update.

• The source of the update is a Ubuntu 16 specific PPA - so it ought be compatible with the base system.

• The only remedy that needs to be done is to roll back the Elmer build pointed to by the PPA (elmerfem-csc (8.4-0ppa0-202106021054~6dfb4824~ubuntu16.04.1) to be specific)

• Everything (preparation, geometry input, meshing, postprocessing) works - except for ElmerSolver.

• The starting issue seems to be something in the memory/heap management. I would expect a malfunction of that kind to relate to a C library function rather than the FORTRAN compiler itself.

With that said ... yes, the Ubuntu is well seasoned at this stage but it is the system on the PC where I have other related software, which cannot run on newer systems.

[teepeeaye]

Also an observation: The traceback refers to ElmerSolver_mpi. It also seems that there is no ElmerSolver (i.e. without _mpi). Combined with the error "double free or corruption (fasttop)" one could suspect an incompatibility in the MPI integration - i.e. lower level libraries again.

I don't know how complicated it would be to separately compile ElmerSolver separately (i.e. without _mpi) to test.

For testing I also tried installing Elmer on Ubuntu 20.04.6 LTS (elmerfem-csc-eg (9.0-0~202306081255~ubuntu20.04.1)) but it immediately crashes, so I was quite happy with the 8.4 version on the oher system.

[teepeeaye]

Update:

I Compiled ElmerSolver and ElmerGrid from 8.4 source at github. I used the standard compiler (gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.12)) without problems. All runs well with ElmerGUI and ElmerVTK from the PPA. It seems that since /usr/local/bin is before /usr/bin in the path, the home made binaries are found first. ElmerVTK crashes if I use vector plot, but I can live with that (I mostly use ParaView anyway).

I am still convinced that the ElmerSolver from the PPA build is sick, but since I found a workaround I mark this thread as solved.