Investigate possible bug when handling non-contiguous node tags in Gmsh parser #96

Activity

1. nikoscham

   self-assigned this

   on Oct 28, 2026
2. nikoscham

   added the

   Bug

   issue type on Oct 28, 2026
3. nikoscham

   added a commit that references this issue on Oct 28, 2026

   Fix non-contiguous node tags in the Gmsh reader (#96)

   ...

   a1ef9e6
4. nikoscham commented on Oct 28, 2026

   

   nikoscham

   on Oct 28, 2026

   MemberAuthor

   More actions

   Confirmed and fixed. The Gmsh reader stored node coordinates at nodeTag - 1, so gaps in the node tags created extra nodes that no element uses. Their rows in the system matrix were all zeros: Jacobi divided by zero and returned NaN at every node, and the LU solver failed outright. The reader now detects gaps from the $Nodes header (max tag > number of nodes) and maps node tags to compact indices, including in element and boundary references. Meshes with contiguous tags, which is the usual Gmsh output, are read exactly as before.
   A new unit test (tests/unit/gmshReader.test.js) copies the test meshes with gaps in their node tags and checks that the nodes, connectivity and solution are unchanged.
5. nikoscham

   linked a pull request that will close this issueFix non-contiguous node tags in the Gmsh reader (#96) #98on Oct 28, 2026
6. nikoscham

   added a commit that references this issue on Oct 28, 2026

   Fix non-contiguous node tags in the Gmsh reader (#96) (#98)

   ...

   88787c6
7. nikoscham

   closed this as completedin #98on Oct 28, 2026