Skip to content

Save and reload a model

Persist a meshed model to a native model.h5 and bring it back later without re-running geometry or meshing. Reach for this to checkpoint a long build, hand a model to a colleague, or resume into composition on another day.

The recipe

Give the session a save_to= path and apeGmsh autosaves the neutral zone on context-exit. Reload it as a bare snapshot with FEMData.from_h5, or as a resumable chain-phase session with apeGmsh.from_h5.

from apeGmsh import apeGmsh, FEMData

# --- Save: autosave on exit -------------------------------------------
with apeGmsh(model_name="plate", save_to="plate.h5", overwrite=True) as g:
    g.model.geometry.add_box(0, 0, 0, 1.0, 1.0, 0.1, label="body")
    g.physical.add_volume("body", name="body")
    g.mesh.generation.generate(dim=3)
    g.save()                       # optional explicit checkpoint mid-session
# neutral zone is written here, in __exit__ (i.e. end()), before gmsh.finalize()

# --- Reload (A): a bare snapshot, integrity-checked -------------------
fem = FEMData.from_h5("plate.h5")  # raises on a tampered/corrupt file
print(fem.info.summary())          # query nodes/PGs/loads with no live gmsh

# --- Reload (B): a resumable chain-phase session (no gmsh) ------------
g2 = apeGmsh.from_h5("plate.h5")   # skips the gmsh build entirely
g2.compose("bolt.h5", label="bolt", translate=(0.5, 0.5, 0.0))
g2.save("assembly.h5")

Exact signatures (verified against src/):

apeGmsh(..., *, save_to: str | Path | None = None, overwrite: bool = True)  # _core.py:84
apeGmsh.save(self, path: str | Path | None = None) -> Path                  # _core.py:251
apeGmsh.from_h5(cls, path, *, model_name=None, verbose=False) -> apeGmsh    # _core.py:142
FEMData.to_h5(self, path, *, model_name="", apegmsh_version="", ndf=0)      # FEMData.py:1603
FEMData.from_h5(cls, path, *, root="/") -> FEMData                          # FEMData.py:1548

Notes / gotchas

  • Autosave fires on end(), not eagerly. Nothing is written until __exit__ / g.end() runs — if the process dies mid-build, the file never lands. Call g.save() explicitly when a checkpoint matters.
  • save_to=None (the default) disables autosave. With a path set, overwrite=False against an existing file raises FileExistsError; g.save() with neither an argument nor a save_to= raises RuntimeError.
  • Autosave catches-and-warns on write failure so gmsh still finalizes. A silently-warned failure can lose data — prefer an explicit g.save() when persistence is the point.
  • Neutral zone only. g.save() / FEMData.to_h5 write the solver-agnostic mesh (nodes, elements, PGs, labels, constraints, loads, masses, per-node ndf) — no /opensees/ zone. To persist the OpenSees deck alongside it, use apeSees(fem).h5("model.h5"), which writes both the neutral and opensees zones.
  • from_h5 is fail-loud. /meta/snapshot_id is re-verified against the recomputed hash on read; a mutated neutral zone raises MalformedH5Error, a missing /meta raises MalformedH5Error, and a wrong schema major raises SchemaVersionError. No silent corruption.
  • Two different from_h5 methods — don't confuse them: FEMData.from_h5 returns a snapshot you can query and emit; apeGmsh.from_h5 returns a chain-phase session with no gmsh state, so g.model.* and g.mesh.generation.* will fail. Only compose(...), save(...), and the chain-phase interface constraints/loads/masses work there.
  • Composed/results files nest the layout. A standalone model.h5 reads at root="/"; a composed results.h5 carries the same layout under /model/, so pass FEMData.from_h5("results.h5", root="/model").
  • Bandwidth is not stored — it is recomputed from connectivity on read, so never rely on a persisted value.

See also

Next: Compose modules into one model.