Bound Mode A's per-function args search to the label's own statement

[squid-protocol]

Summary

_calculate_block_metrics's generic per-function args_count derivation searched the entire greedy block when no args_search_text was passed. For Mode A (_slice_by_labels), that block legitimately spans from one func_start match to the next -- often many unrelated statements -- since Mode A has no brace-bounded body the way Mode B does.

For the 3 Mode A languages with no dedicated args_count_override (cobol, fortran, dockerfile), this let an unrelated, later occurrence get misattributed as the current label's own parameter count:

• A dockerfile RUN could pick up an unrelated ARG line swept into its greedy span.
• A cobol paragraph could pick up an unrelated CALL ... USING statement, or the file's own PROCEDURE DIVISION USING/RETURNING header -- confirmed on real corpus data: a paragraph literally named TIMESTAMP was showing 13 "arguments."

Fix

Adds _mode_a_args_window_end, which bounds the args search to the matched label's own statement span by following real line-continuation syntax (\ for dockerfile, & for fortran -- including Dockerfile heredocs and Fortran's own inline-comment/blank-line/cpp-directive interruptions, since WRF's real subroutines can have 200+ parameters spread across 90+ continuation lines) rather than a blind character-count window.

Scoped explicitly to primary_lang_id in (cobol, fortran, dockerfile) -- deliberately NOT args_count_override is None, which also fires for abap/agc_assembly/assembly whenever their own override legitimately resolves to None (e.g. an ABAP interface-implemented method with no DEFINITION-section entry). This distinction turned out to matter in practice: an earlier version of this fix gated on the override alone and silently applied an untuned bound to real ABAP methods, regressing their correct args counts (zcl_abapgit_http_client.clas.abap's check_http_200/send_receive: real 1, regressed to 0) -- caught by this PR's own crucible_check.py run before it shipped, then fixed to gate on language explicitly instead.

One known, narrow, separately-tracked residual

fortran's init_domain (module_initialize_real.F) drops from an accidental 1 to 0. Root cause: its own func_start match's match.start() lands on an unrelated earlier statement (INTEGER :: internal_time_loop, several blank lines before the real SUBROUTINE init_domain line) due to a separate, pre-existing bug in fortran's func_start regex -- filed as #1982. The old unbounded search happened to still reach the real signature further into an inflated span; a correctly bounded search anchored at the same (wrong) match start can't. Will self-resolve once #1982 lands.

Verification

• pytest across dockerfile/cobol/fortran/abap extraction + strict tests, plus test_detector.py: 672 passed.
• ruff_audit.py --ci / mypy_audit.py --ci: clean against baseline.
• Real corpus before/after (galaxyscope --db-only against language-crucible/data/{dockerfile,cobol,fortran,abap}):
  • dockerfile: all functions now correctly show 0 args (was: several RUNs showing spurious 2).
  • cobol: 3 spurious nonzero-args paragraphs -> 0 (was: TIMESTAMP=13, ACCOUNT-OVERDRAFT-COUNT=1, LOCAL-STORAGE=1; all clearly bogus for paragraphs with no formal parameter concept).
  • fortran: 184/185 real functions unchanged (including WRF's lsm/lsm_mosaic, 247/291 real dummy arguments correctly preserved); 1 known residual traced to fortran func_start's optional TYPE-prefix can bridge across an unrelated statement to a distant SUBROUTINE/FUNCTION #1982.
  • abap: confirmed unaffected/correct (check_http_200=1, send_receive=1, set_header=4, matching the pre-existing golden master).
• crucible_check.py (full_precision + zero_dependency): real, expected drift only in dockerfile/cobol/fortran files plus global-aggregate ripple; both golden masters re-blessed; final re-run: PASS/PASS.

This is #1973 in the same dockerfile tri-comparison manual-verification session as #1976 (merged) -- found while re-verifying the dockerfile func_start fix's own downstream effects, generalized once cobol/fortran turned out to share the exact same unbounded-search code path.

Fixes #1973.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

Severity	Count
🔴 Critical	0
🟠 High	0
🟡 Medium	0
🟢 Low	0
ℹ️ Info	3

ℹ️ Info Findings

[checkov] Ensure that a user for the container has been created

File: /docs/language_status/dockerfile.md:1
Rule: CKV_DOCKER_3
Ensure that a user for the container has been created

[checkov] Ensure that HEALTHCHECK instructions have been added to container images

File: /docs/language_status/dockerfile.md:1
Rule: CKV_DOCKER_2
Ensure that HEALTHCHECK instructions have been added to container images

[checkov] Ensure the base image uses a non latest version tag

File: /docs/language_status/dockerfile.md:157
Rule: CKV_DOCKER_7
Ensure the base image uses a non latest version tag

🐦‍⬛ Powered by Muninn · Skald Lab