Fix issues #208-#211, #213, #214, #216, #218-#220, #249, #251, #253-#255; remove the HIT capability (#243-#246)

[asalmgren]

Addresses nineteen open issues: ten correctness/EB fixes, five build-configuration
and documentation fixes, and the removal of the HIT capability (which closes four
more). Nothing here has been built or run yet — that is the point of opening it now,
so it can be tested on a second machine in parallel.

Issues closed by this PR

Merging this into development (the default branch) will close all nineteen
automatically:

Closes #208
Closes #209
Closes #210
Closes #211
Closes #213
Closes #214
Closes #216
Closes #218
Closes #219
Closes #220
Closes #243
Closes #244
Closes #245
Closes #246
Closes #249
Closes #251
Closes #253
Closes #254
Closes #255

It does not close #252; see "Not closed, but affected" below.

Each is described in context in the sections that follow.

Correctness and EB

Closes #208 — post_timestep: the ns.dump_plane slab was built with this level's EB
factory on a one-box BoxArray of its own, so EBFArrayBoxFactory::create indexed the
level's cell-flag FabArray with a foreign box index. Now built with the default factory.

Closes #209 — ParticleDerive: the MultiFab overload only writes the valid region, but
errorEst reads the ghost cells for GRAD/RELGRAD tags. The result is now zeroed and
FillBoundary'd when ngrow > 0.

Closes #210 — ComputeAofs: the EBFluxRegister coarse re-redistribution buffer was
handed over at component 0 for both the velocity and the scalar call, so scalar mass
accumulated into the velocity slots. Now offset by state_indx, matching the adjacent
CrseAdd/FineAdd calls.

Closes #211 — reflux: with EB, the shares that EBFluxRegister re-redistributes into
fine-covered coarse cells were deleted by the zeroing loop that ran after it. The zeroing
now runs before the advective reflux, so SyncInterp carries them. Non-EB order is
unchanged (YAFluxRegister::Reflux never touches covered cells).

Closes #213 — press_fill: aborted unconditionally on every physical-boundary fill of
Press_Type, which any avg_pressure gradient tag would hit. Replaced with the no-op
nodal functor already sketched in the file; press_bc never contains ext_dir and
GpuBndryFuncFab extrapolates nodal ghosts itself.

Closes #214 — scaleVar/rescaleVar: the RZ velocity scaling looped over sig's grown
tile box, which has no ghost cells in initialVelocityProject, so the axial inflow ghost
row was never scaled. Now uses vel's own one-cell ghost box.

Closes #216 — Diffusion::computeExtensiveFluxes: covered fluxes were overwritten with
COVERED_VAL (1e40) and then passed to the EB-unaware viscous FluxRegister. Now zero,
which is what MLEBABecLap::FFlux already returns.

Closes #218 — ConvertCheckpointGrids: the TimeAverage file was rewritten as a single
(time_avg, time_avg_fluct) pair, dropping dt_avg and every level above 0. Now copied
verbatim, which also keeps old single-pair files readable.

Closes #219 — computeNewDt: ns.fixed_dt was silently undercut by the CFL estimate,
contrary to the docs. With fixed_dt > 0 the estimate is now ignored and only the
change_max ramp applies, so init_shrink still gives the documented reduced start. The
ns.cfl and dt_cutoff documentation is corrected to match the code rather than the
reverse.

Closes #220 — create_umac_grown: the mask test compared against the unqualified
interior, which resolves to RegType::interior, not level_mask_interior. Correct only
by coincidence (both are 0).

Build configuration and documentation

Closes #249 — deleted Exec/run3d/Make.package and
Exec/square_grid_turbulence/Make.package. Nothing includes a Make.package from a build
directory, and both listed Fortran and pre-AMReX_ sources that no longer exist.

Closes #251 — removed amr.probin_file, the two extra ns.gravity values and the
unregistered velocity_average derive from inputs.3d.square_grid; removed mg.v from
inputs_ml; changed ns.smago_Cs_cst to ns.sigma_Cs_cst in inputs.3d.LES_hotspot,
which selects the Sigma model.

Closes #253 — removed the breathe/Doxygen configuration from conf.py, the breathe pip
install, the Doxygen and LaTeX apt packages, and the uninvoked make_api.py. The docs job
only runs make html. The workflow concurrency group is now
${{ github.ref }}-${{ github.head_ref }}-docs, matching the other six workflows, so a
push to any branch no longer cancels an in-flight development deploy.

Closes #254 — corrected eleven documented keys, prefixes and defaults, and documented
every key the issue listed as undocumented: num_steps, strt_time, stop_interval, the
ns.LES_* set, ns.avg_interval/ns.compute_fluctuations, ns.redistribution_type,
ns.refine_cutcells, the outflow gridding keys, the restart-compatibility keys
(ns.gradp_in_checkpoint, ns.avg_in_checkpoint), the valid prob.probtype values with
their prob.* parameters, and the registered derive names.

Closes #255 — removed the MOL time-step section and its mentions, the non-existent Euler
tutorial entry, the Average? LES? placeholder, the IAMR-Codes remote URLs, and fixed
the ViscBench2d.cpp name, the magveel figure filename, the
Chap:InputsLoadBalancing reference and the ytemplates path.

Removal of the HIT capability

Tutorials/HIT and its turbulent-forcing engine (TurbulentForcing_*, depRand, the
local NS_getForce.cpp and prob_init.cpp overrides) are deleted, together with the
AMREX_USE_TURBULENT_FORCING hook in NavierStokesBase::post_restart and the
USE_TURBULENT_FORCING/USE_FAST_FORCE block in Exec/Make.IAMR. No CI job built it and
Test/IAMR-tests.ini had no entry for it, so nothing in the test matrix changes.

This closes four issues that were confined entirely to that directory:

Closes #243 — getForce verbose min/max loops dereferencing device memory on the host.
Closes #244 — USE_FAST_FORCE coarse lattice anchored at the box corner.
Closes #245 — depRand.cpp using omp_get_thread_num() without <omp.h>.
Closes #246 — turb.force_file read by nothing.

The out-of-date Util/run_scripts SLURM templates are removed as well; run.perlmutter
referenced inputs.3d.HIT and hit_ic_4_32.dat, neither of which existed.

The build-system fact that HIT used to illustrate — a build directory's local files
override IAMR/Source — is preserved as prose in ProblemSetup.rst, since the behaviour
survives the removal. The three HIT mentions in RELEASE_NOTES are left alone: those
releases did ship it.

Not closed, but affected

#252 lists USE_TURBULENT_FORCING among the configurations CI never builds. That bullet is
now moot, but the issue's other gaps (USE_PARTICLES, Util/ConvertCheckpoint,
TaylorGreen/benchmarks, USE_OMP breadth, RZ/BDS/restart decks, no GPU job running an
executable) are untouched.

Housekeeping

A second commit adds tmp_build_dir/ and tmp_install_dir/ to the top-level
.gitignore. The make system creates a tmp_build_dir in every build directory; its
o/ and s/ subdirectories were already ignored but the parent was not, so eight of
them appeared as untracked in a clean tree. tmp_install_dir is its sibling, which the
CI whitespace check already prunes. No tracked file is affected.

Testing notes

Not yet built or run. Worth exercising:

• a plain non-EB build for post_timestep: ns.dump_plane slab MultiFab is built with this level's EB factory on a foreign BoxArray #208/ParticleDerive: particle_count derives with ngrow > 0 leave ghost cells uninitialised, so gradient tagging on them reads garbage #209/press_fill: aborts on every physical-boundary fill of Press_Type, so tagging on avg_pressure with adjacent_difference_greater kills the run #213/computeNewDt: ns.fixed_dt is silently undercut by the CFL estimate; ns.cfl is mandatory and dt_cutoff stops instead of aborting, contrary to the docs #219/create_umac_grown: mask test compares against RegType::interior instead of level_mask_interior #220;
• an EB build for ComputeAofs: EB flux-redistribution drho_as_crse for the scalar call is written into the register's velocity components #210/reflux: with EB, the part of the advective reflux that EBFluxRegister re-redistributes into fine-covered coarse cells is zeroed and never reaches the fine level #211/Diffusion::computeExtensiveFluxes: covered FABs/tiles get flux 1e40, which the EB-unaware viscous FluxRegister refluxes into the neighbouring coarse cell #216;
• Util/ConvertCheckpoint for ConvertCheckpointGrids: TimeAverage file is rewritten as a single (time_avg, time_avg_fluct) pair, dropping dt_avg and every level above 0 #218;
• make html in Docs/sphinx_documentation for docs.yml / conf.py / make_api.py: documentation build carries an unused Doxygen+breathe toolchain, and the workflow's concurrency group cancels development deploys from unrelated pushes #253/RunningProblems.rst, AlgorithmOptions.rst, ProblemSetup.rst: eleven documented input keys, prefixes or defaults do not match what the code reads #254/Docs/sphinx_documentation: stale content — removed MOL scheme, non-existent Euler tutorial, old IAMR-Codes URLs, broken cross-reference and figure name #255.

#211 is the one change here that should move an existing baseline, so
Exec/eb_run2d/regtest.2d.bubble and Exec/eb_run3d/regtest.3d.double_shear_layer are the
decks to watch. #214 changes answers only for 2D RZ with an axial inflow, which no shipped
deck has.

🤖 Generated with Claude Code