Fix issues #221, #223, #225, #228, #231, #232, #234, #235, #237, #238…

[asalmgren]

Closes #221
Closes #223
Closes #225
Closes #228
Closes #231
Closes #232
Closes #234
Closes #235
Closes #237
Closes #238
Closes #242
Closes #247
Closes #248

Correctness

• ViscBench: plotfile components are matched to the exact solution by position, not by name #248 ViscBench: fill plotfile components by name rather than by position, so a plotfile written with a non-default amr.plot_vars ordering still gives the right error norms (and a missing name aborts instead of reading past PlotVarNames()).
• ConvertCheckpointGrids::ConvertData: refine path takes interpolation slopes from ghost cells that still hold the setVal(10.) sentinel at coarse-fine and physical boundaries #247 ConvertCheckpointGrids::ConvertData: extrapolate into the ghost cells the periodic FillBoundary does not reach -- coarse-fine boundaries on levels > 0 and non-periodic physical boundaries -- which otherwise still held the setVal(10.) sentinel when the interpolation slopes were computed. Also build the target MultiFabs on the source DistributionMapping, since the loops index source and target through a single MFIter (latent under USE_MPI=TRUE).
• prob_initData: prob.velocity_factor silently overwrites prob.velocity_ic[0] for every probtype, and the Taylor-Green check accepts velocity_ic in its place #234 prob_initData: read prob.velocity_factor only for probtype == 11, and reset IC.v_x first, so prob.velocity_ic is neither silently overwritten by it nor able to stand in for it.
• post_timestep_particle: particles.timestamp_indices is never range-checked, so an entry >= NUM_STATE reads past the state FAB #223 post_timestep_particle: validate particles.timestamp_indices against NUM_STATE at first use, where NUM_STATE is final.

BC / AMR sync

• set_outflow_bcs: a fine level that only partially covers an outflow face is caught by AMREX_ASSERT alone, so release builds silently give its outflow nodes phi = 0 while the coarse level gets the hydrostatic value #235 Projection::set_outflow_bcs: abort unconditionally on a level that only partially covers an outflow face, instead of only asserting in a debug build.
• set_x_vel_bc and friends: the six-entry BC tables in NS_BC.H are indexed by an unvalidated phys_bc, so ns.lo_bc = 6 (or any value outside 0..5) reads past the tables #232 Initialize_bcs: range-check the integer ns.lo_bc/ns.hi_bc path; those values index the six-entry tables in NS_BC.H.
• mlmg_mac_solve: the MAC sync solve on a level above 0 never calls setCoarseFineBC, so MLMG assumes a coarse/fine ratio of 2 #237 mlmg_mac_solve: register the coarse/fine ratio on every level above 0, not only when a coarse phi is supplied. The sync solve passes null, so MLMG was falling back to a ratio of 2. Bit-identical for amr.ref_ratio = 2; corrects amr.ref_ratio = 4.
• MacProj::set_outflow_bcs: a no-op (it rewrites zeros that mac_project just set) that still aborts with "Cannot yet handle partially refined outflow" #238 Remove MacProj::set_outflow_bcs, which only rewrote zeros that mac_project had just set but could still abort with "Cannot yet handle partially refined outflow", along with grids_on_side_of_domain, the then-unused have_divu argument and the now-dead mac_proj.do_outflow_bcs key.

EB

• NavierStokesBase::calc_mut_LES: with EB, mu_LES on covered FABs/tiles is computed from uninitialized velocity gradients #242 calc_mut_LES: zero the velocity-gradient FluxBoxes before compVelGrad, which skips covered FABs.

Inputs / diagnostics

• variableSetUp: five shipped decks request the derived variable diveru, which no longer exists and is silently dropped from the plotfiles #231 Drop the stale derive name diveru from the tutorial inputs files.
• set_plot_coveredCell_val: documented as an ns. input but never read from ParmParse #228 NavierStokes::Initialize: read ns.set_plot_coveredCell_val, which was documented but never queried.
• set_state_in_checkpoint: the abort text tells the user to "try setting both to 0", which corrupts the restart when only one state type is missing #225 set_state_in_checkpoint: correct the abort text -- the flags must mark exactly the missing state types; the old "if unsure, try setting both to 0" advice corrupts the restart when only one type is absent.
• umacFill: unused fill functor in NavierStokesBase.cpp with a stale comment about u_mac BCs #221 Remove the unused umacFill functor and its stale comment about u_mac BCs.

Testing

Builds clean: 2D, 3D, 2D+EB, 3D+EB, 2D+particles, Util/ConvertCheckpoint, Tutorials/TaylorGreen/benchmarks.

• 2D hotspot (3 levels, walls + outflow), 3D Taylor-Green, EB flow-past-cylinder (gravity + outflow + max_level=1) and EB bubble all run normally; none hit the new aborts.
• Both new guards fire when provoked: prob.velocity_ic alone for Taylor-Green, and ns.lo_bc = 6.
• ns.set_plot_coveredCell_val=0 now yields a different plotfile than =1, confirming the key is read.
• For ConvertCheckpointGrids::ConvertData: refine path takes interpolation slopes from ghost cells that still hold the setVal(10.) sentinel at coarse-fine and physical boundaries #247, a no-extrapolation build was used to diff the converted checkpoints: every cell-centred state type differs at every level while nodal pressure is byte-identical, and the spurious density overshoot next to the walls drops from 1.000144 to 1.000090.

🤖 Generated with Claude Code