Spike: compile the engine to WebAssembly

[HugoFara]

The prior investigation concluded "the dependency structure permits it" but stopped before any compiler ran. I ran one.

NetCDF is the only thing in the way

All 58 files in src/, with em++ -std=c++11 -fsyntax-only under docker.io/emscripten/emsdk:

	Count
Compile unmodified	10
Fail	48
Distinct first-error messages across all 48 failures	1

Every failure is 'netcdf' file not found. No second problem behind it — no missing POSIX header, no unsupported construct, no toolchain incompatibility. The ten that compile are the ones that never reach DataBroker.h:

BurningMap  FFEvent  FFPoint  FFVector  FireNodeData
ParallelData  ParallelException  SimulationParameters  Simulator  TimeTable

#include <netcdf> appears in exactly one place, src/DataBroker.h:25, and the using namespace netCDF there drags it into everything transitively.

Everything else checks out

• MPI is fully behind #ifdef MPI_COUPLING; #include <mpi.h> (CLibForeFire.cpp:15) is inside the guard, and the flag is already off for wheel builds.
• The HTTP server is three references: Command.h:25, Command.h:249, Command.cpp:3422.
• popen is one block, Command::systemExec.
• termios appears only in app/forefire/AdvancedLineEditor.cpp, which a WASM build would not compile.

One portability finding

em++ flags seven -Wvla-cxx-extension, e.g. ArrayDataLayer.h:275:

T vals[nnx*nny];

Variable-length arrays are C99, never adopted by C++; GCC and Clang accept them as an extension, MSVC does not. Not a WASM blocker, but a latent portability problem and a stack-overflow risk if those dimensions are ever large. Worth fixing independently.

What remains unknown

1. Does it link and run? -fsyntax-only only. Codegen, linking and execution untested.
2. Binary size. Unmeasured; ~34k lines with heavy templating in the layer classes.
3. Threads. Emscripten pthreads need SharedArrayBuffer plus COOP/COEP headers, which conflicts with the "just a static page" appeal. Single-threaded is fine for a demo, but the browser build and the free-threading work do not compose.

Why it is worth the weekend

The demo becomes a static page — no server, and the listenHTTP exposure stops being a deployment concern. tools/htdocs/ is already 2811 lines of working Leaflet UI, and data would come in as JS typed arrays through the same interface #168 documents for numpy. Adjacent payoff: the same FF_NO_NETCDF guard enables a slim wheel that drops the entire NetCDF/HDF5 group.

Next step: guard #include <netcdf>, the using namespace netCDF, the six NcVar declarations in DataBroker.h, the two implementation blocks in DataBroker.cpp, and loadArrivalTimeNC/saveArrivalTimeNC in FireDomain.cpp behind FF_NO_NETCDF. Then re-run the sweep and see what the second error is, if there is one.

---

Drafted by Claude Opus 5 from a codebase audit. Reviewed by a maintainer before filing.

EDIT: rewrote for human readability.