Fix issues #241, #240: remove unused Diffusion members; scale only the delta_rhs source by rho_half

[asalmgren]

Closes #241. Closes #240.

#241 — remove Diffusion members that are never exercised

None of these has a caller anywhere in Source/, Exec/, Tutorials/ or Util/:

• diffuse_Ssync (~170 lines). This is a second, independent copy of the scalar sync solve, with its own density logic (computeAlpha on current-time S, an explicit rho_flag==2 rescaling, rho_half scaling of the RHS) that was never checked against the live path. The live sync diffusion is diffuse_scalar with add_old_time_divFlux=false, called from NavierStokes::mac_sync. Its last call site was already commented out before c55efb16 (2019).
• FillBoundary(BndryRegister&, ...) — left over from the pre-MLMG boundary-register solvers.
• viscFluxReg(), maxOrder(), tensorMaxOrder() — the flux register is reached through NavierStokesBase::getViscFluxReg(), and the max_order / tensor_max_order statics are used directly. The member variables and the diffuse.max_order / diffuse.tensor_max_order ParmParse knobs are unchanged.

The comment in diffuse_tensor_velocity that cited diffuse_Ssync as the reference for RHS scaling now cites only diffuse_scalar.

computeAlpha keeps its alpha_in branch: once diffuse_Ssync is gone the remaining forwarder is diffuse_scalar, which also consumes alpha_in in its own source loop, so it is live API surface rather than dead code. Its assert now checks alpha_in->nComp() instead of the freshly defined one-component destination, which only ever admitted alpha_in_comp == 0 and never validated the source at all.

#240 — scale only the delta_rhs source by rho_half

For rho_flag==1 the source block multiplied the whole RHS by rho_half, including the old-time diffusion term that mgn.apply had already deposited there when add_old_time_divFlux is set. Only the source should be scaled. Latent today: no caller passes both delta_rhs and add_old_time_divFlux=true.

Verification

• Exec/run2d and Exec/eb_run3d build clean with USE_ASSERTION=TRUE WARN_ALL=TRUE WARN_ERROR=TRUE and the CI's EXTRACXXFLAGS.
• grep for all five removed symbols across Source/ Exec/ Tutorials/ Util/ Docs/ returns nothing.
• regtest.2d.hotspot (3 levels, temperature conduction + scal_diff_coefs, mac_sync active every step), 10 steps, compared against a binary built from development: fcompare reports PLOTFILE AGREE — exact zeros on every variable at all three levels. Expected, since nothing removed was reachable and the Diffusion::diffuse_scalar: with rho_flag==1 and delta_rhs, the old-time diffusion term is also multiplied by rho_half #240 path is latent.

🤖 Generated with Claude Code