Add HLASM / IBM Z Assembly (hlasm), resolving the .asm collision with assembly (#2503)

[squid-protocol]

Closes #2503 (parent #2516).

A new hlasm profile (.asm/.hlasm/.mac; positional_anchored through prism's bms mode — bms is HLASM macro source, so #2505's mode gate widens to lang_id in ("bms", "hlasm") at both call sites) wiring 44 of 54 rule keys and covering #2503's x/y/z asks: func_start CSECT/RSECT/START, ipc_rpc_bridges DFHEIENT/DFHEIRET + the CICS/SQL/DLI bridge vocabulary (cobol/pli's #2990 sharing, verbatim), and macros MACRO/MEND + the conditional-assembly family.

The .asm conflict is resolved as the issue asked, via discriminators plus the #2511 mechanisms:

• .asm joins COLLISION_FREQUENCIES so Tier 1 never locks it on extension alone.
• hlasm declares mainframe-sibling discriminators (.mac/.jcl/.cbl/.cob/.cpy/.pli/.pl1/.bms) for Tier 1.5 ecosystem gravity, x86-artifact disqualifiers (.nasm/.masm/.s/.S/.ld/CMakeLists.txt), and an internal_discriminator (CSECT/DSECT/RSECT/AMODE/RMODE/USING/LTORG/MEND/DFHEIENT/DFHEIRET in operation-field position) for Tier 2 content locks. Verified end-to-end through the real LanguageDetector: HLASM content in .asm locks hlasm; NASM content in .asm stays assembly.

Deliberate deviation from the issue text, contract-cited and pinned by a test (the #2511 documented-deviation shape): DSECT is class_start, not func_start — a dummy section is a named storage layout that assembles no executable logic, #2856's record/struct family, reached by USING name,reg. Also deliberate: no invocation_model: positional, diverging from bms/jcl — L 15,=V(name) + BALR, the CALL macro and USING reach hlasm's extracted units by name, so the #2866 census applies (and hlasm joins the _visibility_export_list pinned family: ENTRY A,B names many entries at once, #2823's list form).

Wiring beyond the definition:

• prism.py: bms-mode gate widened to hlasm (both call sites; */.* column-1 comments, no column-7 check, no inline split — the prism.py: ABAP's positional comment stripper reuses Fortran/COBOL's column-1 anchor set, erasing real code as comments #1898 C-name survival pinned a third time).
• detector.py: Mode A routing (the bms/bms falls through to Mode_B_Braces: functions_found 0 on the rosetta shell (13 raw DFHMDI matches, none reach function_data) #3077 slot — no braces, sections never nest) + named class_start extraction (DSECT names).
• analysis_lens.py: strictness row (all-False) + the systems ecosystem beside cobol/pli/db2_sql.
• docs: language_status/hlasm.md + index row.

tests/extraction/languages/test_hlasm_strict.py (the issue's named deliverable): 145 cases — per-rule positive/negative coverage, schema completeness, collision routing through the real detector, prism family checks, Mode A extraction, operation-field boundary audits (WAIT=YES on a continuation line, remarks after operands), and a 35-payload ReDoS detonation sweep whose payloads include the issue's "long macros" case (an unclosed 32k-char MACRO prototype). The adversarial pass caught two real defects pre-merge, both fixed: dead_code fired on * DC POWER SUPPLY NOTES banner prose, and _visibility_export_list could capture a blanks-only region.

Golden masters regenerated in both environments. All 148 fixture line changes per env are Artifact Identity metadata (Lock Tier / Identity Proof) on the crucible's 74 x86/ARM .asm files, whose extension locks became Ecosystem Consensus locks (80–100% local dominance) now that .asm is contested — the #3090 .sql shape verbatim. Verified via golden_diff.deep_compare: 0 off-target files, 0 language flips, 0 signal-count changes. The corpus carries no z/Architecture sources, so hlasm's own positive case is covered by the keyword-rosetta shell instead.

Real-scan proof (the #2806 class): galaxyscope CLI on a fresh HLASM tree → file_data reads language=hlasm, function_count=4, struct_branch=3, arch_io=3 — matching the control-shell manifest exactly.

The keyword-rosetta data/hlasm/ control folder ships as a companion PR (corpus/hlasm2503, 98 gate assertions PASS); merge this PR first, then the corpus one (the db2_sql order). The fidelity_table.py FIDELITY_PROVENANCE re-pin remains owed post-merge (the audit's standing warning, accumulated since #2511).

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab