Marshalling

How JavaScript values cross the WASM boundary to the Gmsh C API and back. This is handled entirely by src/runtime.mjs, driven by the generated descriptor — you do not write any of this, but understanding it explains the API shapes.

Inputs

Gmsh argument	You pass	Sent to C as
bool	boolean	int (0/1)
int, size_t	number	int / size_t
double	number	double
string	string	const char* (UTF-8, heap-allocated)
vector<int>	number[]	const int*, size_t n
vector<double>	number[]	const double*, size_t n
vector<string>	string[]	const char* const*, size_t n
vector<pair>	number[] flat	const int*, size_t n (n = flattened length)
vector<vector<int>>	number[][]	const int* const*, const size_t*, size_t nn
mesh size callback (setSizeCallback only)	(dim, tag, x, y, z, lc) => number	native function pointer, via Module.addFunction

All temporary heap allocations for inputs are freed after the call — except the mesh size callback's function-pointer table slot, which Gmsh stores and invokes later during mesh.generate(); it is released only when replaced by a later setSizeCallback call, or dropped by removeSizeCallback/finalize.

Outputs

Output parameters are allocated by the wrapper, filled by Gmsh (which mallocs the arrays), copied into JS values, and then released with gmshFree.

Gmsh output	You receive
int* / size_t* / double*	number
char**	string
int**+size_t* (vector)	number[]
char***+size_t* (vector of strings)	string[]
int***+size_t**+size_t* (vector of vectors)	number[][]

Return shape rules

For a function f:

1. No outputs, no scalar return → returns undefined.
2. Scalar return only (e.g. addPoint → a tag) → returns that number.
3. One or more output parameters → returns an object keyed by each output's name. If the function also has a scalar return, it is included as value.

// rule 2
const tag = gmsh.model.geo.addPoint(0, 0, 0);          // number

// rule 3
const { nodeTags, coord } = gmsh.model.mesh.getNodes(); // object

Memory & numeric notes

• The boundary is wasm32: pointers and size_t are 4 bytes; double is 8.
• size_t outputs are read unsigned (HEAPU32), so large tags/counts up to 2³² are represented correctly.
• Heap views are re-fetched on every access, so memory growth during a call cannot leave a stale typed-array view.
• Errors are surfaced by reading gmshLoggerGetLastError and throwing.