module_dftu: pass nspin and onsite_radius explicitly, removing the two test access hacks it caused

[Critsium-xy]

Follows the method established by #7921, applied to the one module outside
source_io where it actually removes the macro.

Why these two files

#define private public has two independent root causes, and only one of them
is about PARAM:

• (a) the code under test reads global PARAM itself, so the test can only
  drive it by writing PARAM.input / PARAM.sys, which are private;
• (b) the test reaches into private members of the class under test.

Passing INPUT values explicitly fixes (a) and nothing else. I classified all 69
files that carry the macro by which header the macro region actually covers:
11 are pure-(a), 22 are (a)+(b) where the macro survives either way, and 36
are pure-(b). Outside source_io, module_dftu holds the only two pure-(a)
files in the tree — so this is the complete (a)-driven macro removal available
outside that module.

What changes

dftu_nao_op.cpp read two INPUT values out of the global singleton:
PARAM.inp.nspin in the constructor and PARAM.inp.onsite_radius inside
initialize_HR(). That is why dftu_lcao_test.cpp had to write PARAM.input,
and why it carried the macro.

Both now arrive as arguments. initialize_HR() is private and called only from
the constructor, so it simply takes onsite_radius; no new member was added for
it. dftu_nao_op.cpp now has zero PARAM references and no longer includes
parameter.h.

All five call sites are updated explicitly — no default arguments were added:

• hamilt_lcao.cpp, gamma-only and multi-k branches;
• force_stress_lcao.cpp, the tmpu force/stress instance;
• the two cases in dftu_lcao_test.cpp, which pass values they already set locally.

Those first three are the composition roots for LCAO operators — the same role
relax_nsync.cpp plays in #7921. The global reads concentrate there and the leaf
operator stays clean.

dftu_pw_test.cpp needed no production change at all. It was parking its loop
variable in the singleton and reading it straight back — writing the private
half and reading the public one:

PARAM.input.nspin = c.nspin;
switch (PARAM.inp.nspin) { ... }
const double diag_coeff = PARAM.inp.nspin == 4 ? 1.0 : 0.5;

Nothing else reads it: the test's only production dependency,
source_pw/module_pwdft/dftu_base_tools.cpp, contains zero PARAM references.
c.nspin was already in scope.

Why both files are in one PR

The governance budget counts PARAM/GlobalV/GlobalC occurrences PR-wide and
blocks on added - removed > 0. Unlike #7921 — where relax_nsync.cpp already
held an injected Input_para* and the refactor was net −64 — both production
call sites here sit inside PARAM-dense functions with no local nspin in
scope, so moving the read out of the leaf costs budget at the call sites:

	Δ
dftu_nao_op.cpp — the 2 reads removed	−2
dftu_lcao_test.cpp — 3 PARAM.input writes removed	−3
dftu_pw_test.cpp — 3 PARAM lines removed	−3
3 production call sites — nspin + onsite_radius added	+6
net_delta	−2

Measured: added=6, removed=8, net_delta=-2. Split into two PRs, the second one
is +1 and CI blocks it. The dftu_pw_test cleanup is what pays for the refactor.

This is worth stating as a general point for the programme: (a)-type refactors
are budget-positive by construction — one read point becomes N call sites — so
they need either an existing injection point or a paired cleanup in the same PR.

Test expectations are unchanged

No expected value was relaxed and no result.ref was regenerated.

Both dftu_lcao_test cases already set nspin explicitly (1 and 2) and
onsite_radius in SetUp (1.0), so nothing was relying on the Input_para
defaults — the regression that cost #7921 five suites when a new struct's
defaults did not mirror input_parameter.h. I checked every case before
touching them. The fixture member onsite_radius_test = 1.0 carries the value
SetUp used to write into PARAM.

Result

	before	after
files with the macro	69	67
occurrences	97	95
PARAM/GlobalV refs in module_dftu production code	9	7

The 7 remaining are ks_solver in dftu_nao_fs_k.cpp / dftu_nao_occ.cpp and
nspin in dftu_nao_energy.cpp. None is in either test's depend_files, so
touching them buys no macro and only costs budget.

Nothing here adds an #undef private or #undef protected, and no file whose
macro survives is touched — #7921's scope rule: remove the macro, or leave the
file alone.

Verification

Linux, cmake -B build -G Ninja -DBUILD_TESTING=ON -DENABLE_LCAO=ON -DENABLE_MPI=ON -DENABLE_OPENMP=ON:

• build: 3318/3318 targets, 0 errors. Dropping the parameter.h include from
  dftu_nao_op.cpp exposed no transitive dependency.
• ctest -R dftu: 5/5 pass, including both changed tests.
• full unit suite: 29 of 338 fail — the failure set is byte-identical to the
  base commit (402aa8dab) built and run in a parallel worktree in the same
  environment. Both comm directions are empty: no new failures, no tests newly
  passing. The 29 are pre-existing and environment-related (mpirun-based
  *_para/*_parallel wrappers, plus cubic_spline, math_sphbes, dav,
  bpcg, PSI_init, ...).
• agent_governance_check.py --base origin/develop --head HEAD: 0 errors,
  exit 0. The access-hack ratchet reports nothing at all (2 removed, 0 added).
• Verified via compiled objects that all five changed files were actually built,
  rather than skipped by the local feature configuration.

Governance warnings and why: the 6 "Global dependency budget" hits are the
PARAM.inp.nspin / PARAM.inp.onsite_radius arguments added at the three call
sites, explained above; net delta is −2. The "Documentation sync review" warning
applies because no INPUT parameter behaviour changed — this is an internal
signature change only, so docs/parameters.yaml and
docs/advanced/input_files/input-main.md need no update.

What is deliberately not here

Reason-(b) files need a per-class decision in a production header — a public
const observer, an explicit friend, or turning a void method that writes a
private flag into one that returns it. That is the next tranche and is
independent of PARAM.

Reason-(a) elsewhere outside source_io no longer removes any macro: the six
source_md tests are driven entirely by five PARAM writes in setcell.h and
their integrators are already PARAM-free, but all six macros are (b);
source_estate's remaining (a) is nspin/nbands/nlocal, which is the
de-globalization programme rather than one PR.

🤖 Generated with Claude Code