segfault during preprocessing (swapping)

[MariusCausemann]

Hi,
first of all, thanks for this great meshing tool! In most cases, fTetWild works very well for me, but recently, I encountered a segfault in the preprocessing stage. It is quite a complex mesh (consisting of ~200 separate surfaces) and unfortunately, I can reproduce using a simpler example. If necessary, I can provide the input files.

Here is fTetWild output:

6 10 12 2 9 5 16 13 0 14 4 1 19 8 18 11 3 17 15 7 
TBB threads 7
[2024-08-14 10:07:12.503] [geogram] [info] Loading file results/cortical_mm3/surfaces/size+20000/dx+20_ncells+200/roi.ply...
[2024-08-14 10:07:12.504] [geogram] [info] (FP64) nb_v:8 nb_e:0 nb_f:6 nb_b:0 tri:0 dim:3
[2024-08-14 10:07:12.504] [geogram] [info] Attributes on vertices: point[3]
[2024-08-14 10:07:12.505] [geogram] [info] Loading file results/cortical_mm3/surfaces/size+20000/dx+20_ncells+200/2.ply...
[2024-08-14 10:07:12.521] [geogram] [info] (FP64) nb_v:20838 nb_e:0 nb_f:41650 nb_b:19 tri:1 dim:3
[2024-08-14 10:07:12.521] [geogram] [info] Attributes on vertices: point[3]
[2024-08-14 10:07:12.522] [geogram] [info] Loading file results/cortical_mm3/surfaces/size+20000/dx+20_ncells+200/3.ply...
[2024-08-14 10:07:12.533] [geogram] [info] (FP64) nb_v:16564 nb_e:0 nb_f:33120 nb_b:4 tri:1 dim:3
[2024-08-14 10:07:12.533] [geogram] [info] Attributes on vertices: point[3]
[2024-08-14 10:07:12.533] [geogram] [info] Loading file results/cortical_mm3/surfaces/size+20000/dx+20_ncells+200/4.ply...
[2024-08-14 10:07:12.547] [geogram] [info] (FP64) nb_v:24800 nb_e:0 nb_f:49587 nb_b:228 tri:1 dim:3
[2024-08-14 10:07:12.547] [geogram] [info] Attributes on vertices: point[3]
[2024-08-14 10:07:12.548] [geogram] [info] Loading file results/cortical_mm3/surfaces/size+20000/dx+20_ncells+200/5.ply...
[2024-08-14 10:07:12.558] [geogram] [info] (FP64) nb_v:19343 nb_e:0 nb_f:38678 nb_b:13 tri:1 dim:3
[2024-08-14 10:07:12.559] [geogram] [info] Attributes on vertices: point[3]
[2024-08-14 10:07:12.559] [geogram] [info] Loading file results/cortical_mm3/surfaces/size+20000/dx+20_ncells+200/6.ply...
[2024-08-14 10:07:12.586] [geogram] [info] (FP64) nb_v:13240 nb_e:0 nb_f:26475 nb_b:2 tri:1 dim:3
[2024-08-14 10:07:12.586] [geogram] [info] Attributes on vertices: point[3]
...

bbox_diag_length = 34575
ideal_edge_length = 1728.75
stage = 2
eps_input = 18.2627
eps = 10.1101
eps_simplification = 8.08806
eps_coplanar = 0.034575
dd = 12.1752
dd_simplification = 9.74013
[2024-08-14 10:07:59.389] [float-tetwild] [info] remove duplicates: 
[2024-08-14 10:07:59.389] [float-tetwild] [info] #v: 10072385 -> 10072385
[2024-08-14 10:07:59.389] [float-tetwild] [info] #f: 20149242 -> 20149242
[2024-08-14 10:07:59.801] [float-tetwild] [info] remove degenerate triangles and duplicated opposite-oriented triangles:
[2024-08-14 10:07:59.801] [float-tetwild] [info] (removing degenerate triangles might leave lone vertices that will be removed later)
[2024-08-14 10:07:59.801] [float-tetwild] [info] #v: 10072385 -> 10072385
[2024-08-14 10:07:59.801] [float-tetwild] [info] #f: 20149242 -> 20149242
collapsing 1410.3s
segfault

Any idea what the problem could be?
@nmnoble, I also tried your fTetWild branch, leading to the same issue. Any ideas?

[nmnobre]

Hi @MariusCausemann, I'm afraid I don't have any ideas, it's been a while since I've used fTetWild...
Have you tried using a debugger, like gdb or lldb, to help you narrow down the problematic code?

[MariusCausemann]

I've tried to narrow down the problem with a debugger now. Using gdb:

Thread 1 "FloatTetwild_bi" received signal SIGSEGV, Segmentation fault.
floatTetWild::get_angle_cos (p=..., p1=..., p2=...) at /home/mariusca/emimesh/fTetWild/fTetWild-master/src/Simplification.cpp:875
875         Vector3 v1 = p1 - p;
(gdb) print p
$1 = (const floatTetWild::Vector3 &) <error reading variable: Cannot access memory at address 0x1555452abff8>
(gdb) print p1
$2 = (const floatTetWild::Vector3 &) @0x15554710b518: {<Eigen::PlainObjectBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<Eigen::MatrixBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<Eigen::DenseBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<Eigen::DenseCoeffsBase<Eigen::Matrix<double, 3, 1, 0, 3, 1>, 3>> = {<Eigen::DenseCoeffsBase<Eigen::Matrix<double, 3, 1, 0, 3, 1>, 1>> = {<Eigen::DenseCoeffsBase<Eigen::Matrix<double, 3, 1, 0, 3, 1>, 0>> = {<Eigen::EigenBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<No data fields>}, <No data fields>}, <No data fields>}, <No data fields>}, <No data fields>}, <No data fields>}, m_storage = {m_data = {array = {4676.6656289838811, 3589.3853991366313, 
          1557.1230833073887}}}}, <No data fields>}
(gdb) print p2
$3 = (const floatTetWild::Vector3 &) @0x155547131a50: {<Eigen::PlainObjectBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<Eigen::MatrixBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<Eigen::DenseBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<Eigen::DenseCoeffsBase<Eigen::Matrix<double, 3, 1, 0, 3, 1>, 3>> = {<Eigen::DenseCoeffsBase<Eigen::Matrix<double, 3, 1, 0, 3, 1>, 1>> = {<Eigen::DenseCoeffsBase<Eigen::Matrix<double, 3, 1, 0, 3, 1>, 0>> = {<Eigen::EigenBase<Eigen::Matrix<double, 3, 1, 0, 3, 1> >> = {<No data fields>}, <No data fields>}, <No data fields>}, <No data fields>}, <No data fields>}, <No data fields>}, m_storage = {m_data = {array = {4693.8347660668605, 3581.6003856561606, 
          1508.7043087844927}}}}, <No data fields>}
(gdb) up
#1  0x0000555555670db7 in floatTetWild::swapping (input_vertices=..., input_faces=..., tree=..., params=..., v_is_removed=..., f_is_removed=..., conn_fs=...)
    at /home/mariusca/emimesh/fTetWild/fTetWild-master/src/Simplification.cpp:546
546             Scalar cos_a1 = get_angle_cos(input_vertices[n_v_ids[1]], input_vertices[v1_id], input_vertices[v2_id]);
(gdb) down
#0  floatTetWild::get_angle_cos (p=..., p1=..., p2=...) at /home/mariusca/emimesh/fTetWild/fTetWild-master/src/Simplification.cpp:875
875         Vector3 v1 = p1 - p;
(gdb) up

So it seems p is out of memory when get_angle_cos is being called in line 546 in Simplification.cpp. But I'm not familiar enough with the algorithm to figure out what could be going wrong...

[MariusCausemann]

The issue can be reproduced with:

FloatTetwild_bin --csg csgtree_roi.json --max-threads 8 --level 2 -e 0.004

The files are available here:
https://drive.google.com/file/d/1cRMugFdne2H4Mi7GOxlD-sga4rv7oODW/view?usp=sharing
I'd appreciate if someone could take a look! (@danielepanozzo ?)

[MariusCausemann]

Hi @teseoch , I've seen that you added a few fixes to the simplification process in wildmeshing/wildmeshing-toolkit#788. Do you think those would also address this issue?