Segfault switching from mmg3d 5.3.6/dev 32bit to 5.7.1 64 bit

[jeandodo]

Bonjour

Notre code mailage hip marchait ss pb avec 5.3.6, en gros j'appelle

// own routine to copy mesh data to mmg
mmg_put_mesh_3d ( ..

/* Define sizes for a solution/metric field and allocate. */ if ( !MMG3D_Set_solSize( pMMesh, pMMet, MMG5_Vertex, pMMesh->np, MMG5_Scalar ) ) hip_err ( fatal, 0, "failed after MG3D_Set_solSize in adapt_mmg3d" ) ; if ( !MMG3D_Chk_meshData( pMMesh, pMMet ) ) hip_err ( fatal, 0, "failed after MMG3D_Chk_meshData in adapt_mmg3d" ) ;

hip_err ( info, 0, "before call MMG3D_doSol in adapt_mmg3d" ) ; // Initialise metric with average neighbor cell edge length status = MMG3D_doSol ( pMMesh, pMMet ); if ( !status ) { sprintf ( hip_msg, "call to MMG3D_doSol with status %d", status ) ; hip_err ( fatal, 0, hip_msg ) ; }

Basculant sur 5.7.1 master, (ayant modifie les int necessaries pour 64bit), je recois un segfault dans MMG3D_doSol.

Dans mmg/mmg3d/libmmg3d_tools.c dans la routine MMG3D_setfunc mmg declare le pointer MMG3D_doSol ou bien vers MMG3D_doSol_iso, ou bien _ani. Notre cas est _iso.

Si j'appelle directement la variante _iso, dedans mmg appele a la fin de doSol_iso MMG3D_solTruncatureForOptim et s'arrete sur le assert ( mesh->info.optim ), comme il faut. Il semble que la valeur de 0 pour mesh->info.optim est la seule possible.

Si j compile en debug, et j'appelle directement la variante _iso, mon cas simple passe ss pb, et comme avec la 5.3.

Comment pourrai-je deboguer ce cas? Cordialement, Jens

[Algiane]

Hi,

I don't work anymore on the Mmg support so I can't promise to solve this issue (or futur ones). I will try to help you to find the bug if my free time allows it.

I am not sure to understand when Mmg-7 fails :

• it fails only when Mmg is built with int64 integers or when you build it with int32 integers too?
• the issue is that the assertion on non-null value for mesh->info.optim fails?

If the issue is that mesh->info.optim is 0 inside MMG3D_solTruncatureForOptim I will remove the assert : When called from Mmg, mesh->info.optim has to be non-null in this function but called by the user it is not mandatory.

Another way to avoid this issue for you is to set (resp. unset) the mesh->info.optim value before (resp. after) the call to doSol function using the MMG<X>_Set_iparameter function and the MMG<X>_IPARAM_optim keyword.

Best, A

[jeandodo]

Hi Algiane,

nice to hear from you, and thanks for helping.

it fails only when Mmg is built with int64 integers or when you build it with int32 integers too? yes, a int32 build with cmake -DUSE_VTK=NO .. also fails: <<hip[grid_4: 1/1]> mmg isofa -f0.5 INFO: Initialisation of MMG3D MAXIMUM MEMORY AUTHORIZED (MB) 7970 INFO: before call MMG3D_doSol in adapt_mmg3d Segmentation fault (core dumped)

the issue is that the assertion on non-null value for mesh->info.optim fails? When I compile mmg in debug, then I can look inside and see that the assertion is thrown. Tracing info.optim through the source I did not spot where it is made non-zero. But assert() is no-action unless compiled with -g? So this shouldn't make any difference in non-debug mode?

When I compile w/o debug, and manually modify info.optim as

hip_err ( info, 0, "before call MMG3D_doSol in adapt_mmg3d, setting mesh->info.optim=1" ) ; pMMesh->info.optim = 1 ; // Initialise metric with average neighbor cell edge length status = MMG3D_doSol ( pMMesh, pMMet ); if ( !status ) { sprintf ( hip_msg, "call to MMG3D_doSol with status %d", status ) ; hip_err ( fatal, 0, hip_msg ) ; } hip_err ( info, 0, "after call MMG3D_doSol in adapt_mmg3d, setting mesh->info.optim=0" ) ; pMMesh->info.optim = 0 ;

in int32 I still get a segfault, so the segfault is not associated with the assertion:

<<hip[grid_4: 1/1]> mmg isofa -f0.5 INFO: Initialisation of MMG3D MAXIMUM MEMORY AUTHORIZED (MB) 7970 INFO: before call MMG3D_doSol in adapt_mmg3d, setting mesh->info.optim=1 Segmentation fault (core dumped)

What works is compiling mmg either in debug with int64, or compiling with int32 (no debug), and calling doSol_iso instead of doSol:

<<hip[grid_4: 1/1]> mmg isofa -f0.5 INFO: Initialisation of MMG3D MAXIMUM MEMORY AUTHORIZED (MB) 7970 INFO: before call MMG3D_doSol in adapt_mmg3d, setting mesh->info.optim=1 INFO: after call MMG3D_doSol in adapt_mmg3d, setting mesh->info.optim=0 INFO: metric info: max = 1.874758 min = 1.701226 avg = 1.799191

Thanks for your advice, Jens