Validate cpp tri-comparison ledger; fix ctags/tree-sitter comparison-tooling bugs

[squid-protocol]

Summary

Sweep of all 14 cpp shapes in docs/self_scan/tri_comparison_ledger.json via the
tri-comparison-ledger-sweep skill. All 14 now status: "validated".

Comparison-tooling fixes (tests/tools/, not GitGalaxy production code, not
tree-sitter/ctags themselves) accounted for the bulk of the false-discrepancy volume:

• ctags_reader.py: out-of-class method names (Class::method) weren't being
  re-qualified with ctags' own separate scope field, and ctags' default pattern-length
  truncation was silently breaking that re-qualification on longer signatures
  (--pattern-length-limit=0 added). Resolved ~1000 of the two largest shapes' ~2170
  combined occurrences.
• ctags_reader.py: CTAGS_CLASS_KINDS["cpp"] was missing ctags' union/enum kinds
  (enum gated on scoped-vs-unscoped via the tag's own source line, since C++11's
  enum class has no ctags-side distinction from a plain enum). The identical gap
  was found and fixed for csharp in the same incidental-finding pass.
• tri_comparison_gatherer.py / tree_sitter_accuracy_audit.py: both tree-sitter
  walkers counted a bare forward declaration (class Foo;) and a plain, unscoped
  enum Foo {} as real class definitions — the existing C-only body-anchor guard was
  never extended to cpp, and there was no scoped-enum gate at all.

Six real GitGalaxy engine defects were confirmed via direct source reading during
the investigation. Five were filed rather than patched inline, since each has real
production blast radius and/or a design question needing its own scoped follow-up:

• cpp func_start misses constructors with long member-initializer-lists #2009 — func_start misses constructors with a member-initializer-list exceeding
  the regex's 500-char cap
• cpp func_start misses conversion operators with template/generic return types #2010 — func_start misses conversion operators with a template/generic return type
• cpp args: Zero-undercount for operator() and method definitions, off-by-one overcount for constructors with initializer-lists #2012 — args counting bugs (zero-undercount + constructor off-by-one overcount)
• cpp func_start: False positives from lambdas in constructors and member-initializer-lists #2013 — func_start false positive on a lambda in a constructor initializer-list
• tree_sitter_accuracy_audit.py's _get_param_count undercounts cpp functions with default parameters #2014 — tree-sitter tooling bug (not GitGalaxy), default-value parameters undercounted

The sixth, #2011 (class_start counting forward declarations as real class
definitions), was fixed in a follow-up commit in this same PR — the tree-sitter walker
fix above surfaced it as a hard CI regression in
tests/tree_sitter_accuracy_baseline_cpp.json (once the walker stopped agreeing with
GitGalaxy's own false positives, the gap became impossible to leave filed-but-open). A
naive copy of C's existing flat lookahead regex is unsafe for cpp — C++ multiple
inheritance (class Foo : public A, public B {) hits its comma stop-char before the
real {, confirmed via direct testing — so it's fixed instead with a depth-aware
scanner (_cpp_class_has_body) that correctly walks an inheritance clause's own
top-level commas and template args. Verified via 11 hand-built regression cases, the
full 122-test extraction gauntlet, and crucible_check.py against the full ~80-repo
corpus (zero golden-master diff, confirmed via direct DB query to be because that
report format doesn't expose the named-class list this fix touches — not because the
fix is a no-op).

Also adds docs/language_status/cpp.md (sections 1–8 via the language-status skill,
plus a tri-comparison capstone §9 built from this sweep's own evidence, updated to
reflect #2011's fix) and regenerates the tri-comparison chart/points-of-interest doc
for all 45 languages (only cpp/csharp's numbers actually moved).

Closes #2011.

Test plan

• python tests/ruff_audit.py --ci / mypy_audit.py --ci — no new findings
• ruff format --check clean
• 11 hand-built regression cases for _cpp_class_has_body (forward decl, plain
  def, single/multi inheritance, templated bases, type-use-in-declarator-list,
  function-parameter-default, final specifier) — all pass
• Full cpp extraction gauntlet (122 tests) + tests/core_engine/test_detector.py
  (141 tests) + C's own extraction gauntlet (128 tests, unaffected) — all pass
• crucible_check.py (full-precision + zero-dependency) against the full corpus —
  zero golden-master diff, confirmed legitimate via direct DB inspection
• tree_sitter_accuracy_audit.py --lang cpp --ci — passes cleanly against the
  regenerated baseline (class recall 87.1% → 100.0%)
• Verified --all --write chart diff touches only cpp/csharp rows + the
  expected summary-badge-count ripple
• Caught and corrected a credit_tools misapplication during review (briefly
  pushed ctags' cpp precision over 100% by double-crediting an already-mutually-
  agreeing 2-of-3 pair) before it shipped

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab