fix(groovy): func_start no longer mistakes MarkupBuilder/DSL calls for declarations

[squid-protocol]

Closes #2530.

Root cause

func_start's zero-prefix branch (needed to match real bare constructors,
MyClass(String arg) {) is syntactically indistinguishable from Groovy's
MarkupBuilder/NodeBuilder DSL idiom — a method call with a named-argument map plus
a trailing closure (div(class: "x") { ... }, button(name: "clear", type: "submit") { ... }). Pervasive in Jenkins's own .groovy view templates (which
replaced Jelly) and Jenkins Pipeline DSL steps.

Fix

A real Groovy declaration's parameter list can never contain a key: token (that
syntax is call-site named-argument sugar only), so branch 2 now rejects a
candidate whose parenthesized argument list contains an identifier immediately
followed by : with no intervening whitespace — distinguishing it from a
ternary's cond ? a : b, which conventionally has a space before the colon.

Scope (per the issue's own framing)

The issue explicitly asked to investigate "whether branch 2 can be tightened to
exclude the named-argument-map-first-parameter shape... without also rejecting
legitimate bare constructors" and to "re-run the recall audit against the full
crucible corpus." Did both:

• Full corpus differential scan (galaxyscope --db-only, 201 files, before
  vs. after): 44 of the 57 misdetections in jenkins_view_groovy/ resolved, zero
  removals anywhere else in the corpus — no legitimate bare constructor lost
  recall, including edge cases I specifically probed for (a Map-literal default
  = [:], a ternary default value).
• ReDoS-scaling probed the new lookahead (linear to n=16000, both an unclosed
  plain paren and a colon-heavy pathological input).
• 13 residual FPs remain, explicitly out of scope: single-positional-arg DSL
  calls (stage('Build') { }, node('x') { }, dir(x) { }) have no key: token
  to key off of — the issue itself flagged this as needing "a curated
  builder-tag-name denylist" that's "fragile and incomplete by construction," so I
  didn't try to force it into this fix.

A second, unrelated defect this unmasked

Fixing the button(...) false positive in one file also unmasked a second,
independent func_start bug: raw _("Dismiss") (a paren-less call, Groovy's _()
i18n-helper convention) was already being misdetected by branch 1 (a completely
different mechanism — treating the bare identifier raw as a modifier/type
prefix), but was silently absorbed as "nested inside button" until button
stopped being wrongly detected as an enclosing function. Confirmed via git stash
that this raw-regex match pre-dates this PR. Filed separately as
#2558 rather than
bundled in here — different branch, different root cause, same reasoning the
original issue used to defer itself from #2532.

Docs / data updated

• docs/language_status/groovy.md §5/§9 — corrected FP counts, cross-referenced
  groovy func_start: paren-less builder call (raw _("x")) misdetected as function when no longer masked by a nested FP #2558, noted the residual.
• docs/self_scan/manual_verification.json — groovy function.total 718 → 675
  (verified stays 661, all previously-correct matches unaffected).
• docs/self_scan/tri_comparison_chart.svg / tri_comparison_ledger.json —
  regenerated per repo convention (--all --write), groovy precision badge now
  reads 661/675** instead of 661/718**.

Verification

• Reproduced the exact FP locally and confirmed the fix.
• Added cases to tests/extraction/languages/test_groovy.py (invalid-match cases,
  plus a positive case for the Map-literal-default edge case) and
  test_groovy_strict.py (dedicated regression test covering the DSL-call
  rejections and the ternary/map-literal preservation).
• test_groovy.py + test_groovy_strict.py: 150 passed.
• tests/core_engine/test_language_standards_strict.py: 21 passed.
• Full pytest tests/: 7154 passed, 2 skipped, 9 xfailed, 3 xpassed (unrelated
  pre-existing).
• tests/tools/audit_check.py: clean (ruff/mypy/dead-key/ast-accuracy).
• tests/tools/tree_sitter_accuracy_audit.py --all --ci: clean (groovy correctly
  stays excluded as gg_only, per fix(groovy): tri-comparison manual verification -- 3 real engine defects fixed #2532).
• tests/tools/crucible_check.py (both full_precision and zero_dependency
  modes, full corpus): re-blessed after confirming every one of the 631 diffs
  traces to either the direct fix (43 net, isolated to jenkins_view_groovy/) or
  the standard topological/aggregate ripple class CLAUDE.md's Differential Scan
  protocol pre-approves (588) — zero unexplained/off-target diffs, verified via a
  full uncapped golden_diff.deep_compare() categorization, not just the
  pytest-truncated summary. Both modes PASS clean after re-bless.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab