feat: Phase 2 complete — 13 Phases of COBOL type classification and test benchmark

P0.6: gcov infrastructure
P1: extract_structure output expansion (11 new feature fields)
P2: Confusion group rule engine (8 pairs + contradiction + backtrack)
P3: 4-factor confidence calculation + quality gate update
P4: 33+2 COBOL program type test samples (22 files, 7 categories)
P5: parametrized/ test data generation engine
P6: japanese_data.py lookup tables
P7-10: Type-specific test suites (~159 parametrized tests)
P11: Full classification pipeline (classify_program) + orchestrator integration
P12: Documentation (module-interfaces, test-plan v3.0, coverage-matrix)

Architecture decisions:
- classification_pipeline/ merged to hina/pipeline/
- parametrized/ as independent module
- japanese_data.py as root-level file
- hina/__all__ only exports classify_program()

Co-Authored-By: Claude <noreply@anthropic.com>

cobol_testgen/.gitignore

@@ -0,0 +1,4 @@
+__pycache__/
+.pytest_cache/
+*.pyc
+test_output/

cobol_testgen/__init__.py

@@ -1,4 +1,12 @@
-"""COBOL Test Data Generator — 模块化版入口"""
+"""COBOL Test Data Generator — 模块化版入口
+
+from __future__ import annotations
+公开 API:
+  extract_structure()    — 解析 COBOL 控制流 → dict
+  generate_data()        — 生成测试数据 → list[dict]
+  incremental_supplement — 差分补充数据 → list[dict]
+  check_coverage()       — 覆盖率报告 → dict
+"""
 
 import sys
 import re
@@ -11,14 +19,25 @@ from pathlib import Path
 CONFIG = {}
 
 from .read import preprocess, extract_data_division, extract_procedure_division
-from .read import resolve_copybooks, parse_data_division, parse_file_section, scan_open_statements
+from .read import resolve_copybooks, parse_data_division, parse_file_section, scan_open_statements, parse_file_control
 from .core import build_branch_tree, classify_field_roles, _init_child_names
-from .cond import parse_single_condition, is_field
+from .cond import parse_single_condition, is_field, collect_leaves
 from .design import enum_paths, generate_records, _filter_stop
 from .output import output_json, output_input_files
-from .coverage import run_coverage, generate_coverage_index
+from .coverage import run_coverage, generate_coverage_index, check_coverage
+from japanese_data import generate_fullwidth_text, generate_halfwidth_katakana, generate_wareki_date
 
 logger = logging.getLogger(__name__)
+n__all__ = [
+    "extract_structure",
+    "generate_data",
+    "incremental_supplement",
+    "check_coverage",
+    "CONFIG",
+    "generate_fullwidth_text",
+    "generate_halfwidth_katakana",
+    "generate_wareki_date",
+]
 
 
 # ── OCCURS 展开 ──
@@ -118,7 +137,7 @@ def main():
     fh = logging.FileHandler(log_path, encoding="utf-8", mode="w")
     fh.setLevel(logging.DEBUG)
     fh.setFormatter(logging.Formatter(
-        "%(asctime)s [%(levelname)s] %(name)s: %(message)s"
+"%(asctime)s [%(levelname)s] %(name)s: %(message)s"
     ))
     sh = logging.StreamHandler()
     sh.setLevel(logging.INFO)
@@ -353,7 +372,7 @@ def extract_structure(cobol_source: str) -> dict:
     file_sec = parse_file_section(preprocessed)
     open_dir = scan_open_statements(proc_div) if proc_div else {}
 
-    from .models import BrIf, BrEval, BrSeq
+    from .models import BrIf, BrEval, BrSeq, BrPerform, Assign, CondAnd, CondOr
 
     decision_points = []
     total_branches = 0
@@ -395,19 +414,219 @@ def extract_structure(cobol_source: str) -> dict:
         if m:
             paragraphs.add(m.group(1))
 
+    # ── 新增字段: select_files ──
+    select_files = parse_file_control(preprocessed)
+
+    # ── 新增字段: open_directions_detail (与 open_directions 一致) ──
+    open_directions_detail = open_dir
+
+    # ── 新增字段: has_divide / has_inspect / has_string ──
+    has_divide = bool(re.search(r'\bDIVIDE\b', cobol_source.upper()))
+    has_inspect = bool(re.search(r'\bINSPECT\b', cobol_source.upper()))
+    has_string = bool(re.search(r'\bSTRING\b', cobol_source.upper()))
+
+    # ── 新增字段: divide_constants ──
+    divide_constants = []
+    if has_divide and proc_div:
+        for dm in re.finditer(r'\bDIVIDE\s+([\d.]+)\b', proc_div, re.IGNORECASE):
+            val = dm.group(1)
+            try:
+                divide_constants.append(float(val))
+            except ValueError:
+                pass
+
+    # ── 新增字段: perform_patterns ──
+    perform_patterns = []
+
+    def _walk_performs(node):
+        if isinstance(node, BrPerform):
+            entry = {
+                "type": node.perf_type,
+                "target": node.target,
+                "condition": node.condition,
+                "times": node.times,
+                "varying_var": node.varying_var,
+            }
+            perform_patterns.append(entry)
+            _walk_performs(node.body_seq)
+        elif isinstance(node, BrIf):
+            _walk_performs(node.true_seq)
+            _walk_performs(node.false_seq)
+        elif isinstance(node, BrEval):
+            for _, seq in node.when_list:
+                _walk_performs(seq)
+            _walk_performs(node.other_seq)
+        elif isinstance(node, BrSeq):
+            for c in node.children:
+                _walk_performs(c)
+
+    if branch_tree:
+        _walk_performs(branch_tree)
+
+    # ── 新增字段: main_loop ──
+    main_loop = None
+
+    def _find_main_loop(node, depth=0):
+        nonlocal main_loop
+        if main_loop is not None:
+            return
+        if isinstance(node, BrPerform):
+            if _perform_has_read(node):
+                main_loop = {
+                    "type": node.perf_type,
+                    "read_file": _perform_read_file(node),
+                    "has_at_end": False,
+                }
+                return
+            _find_main_loop(node.body_seq, depth + 1)
+        elif isinstance(node, BrIf):
+            _find_main_loop(node.true_seq, depth + 1)
+            _find_main_loop(node.false_seq, depth + 1)
+        elif isinstance(node, BrEval):
+            for _, seq in node.when_list:
+                _find_main_loop(seq, depth + 1)
+            _find_main_loop(node.other_seq, depth + 1)
+        elif isinstance(node, BrSeq):
+            for c in node.children:
+                _find_main_loop(c, depth + 1)
+
+    def _perform_has_read(perf_node):
+        def _walk_seq(seq):
+            if isinstance(seq, Assign):
+                if seq.source_info.get('type') == 'read_into':
+                    return True
+            elif isinstance(seq, BrSeq):
+                for ch in seq.children:
+                    if _walk_seq(ch):
+                        return True
+            return False
+        return _walk_seq(perf_node.body_seq)
+
+    def _perform_read_file(perf_node):
+        def _walk_seq(seq):
+            if isinstance(seq, Assign):
+                if seq.source_info.get('type') == 'read_into':
+                    return seq.source_info.get('file', '')
+            elif isinstance(seq, BrSeq):
+                for ch in seq.children:
+                    result = _walk_seq(ch)
+                    if result:
+                        return result
+            return None
+        return _walk_seq(perf_node.body_seq)
+
+    if branch_tree:
+        _find_main_loop(branch_tree)
+
+    # ── 新增字段: if_types ──
+    if_types = {"total": 0, "comparison": 0, "equality": 0, "compound": 0, "nested_depth": 0}
+
+    def _walk_if_types(node, depth=0):
+        if isinstance(node, BrIf):
+            if_types["total"] += 1
+            if_types["nested_depth"] = max(if_types["nested_depth"], depth)
+            ct = node.cond_tree
+            if ct:
+                leaves = collect_leaves(ct)
+                # Check compound: cond_tree is CondAnd or CondOr (not just CondLeaf)
+                if isinstance(ct, (CondAnd, CondOr)):
+                    if_types["compound"] += 1
+                for leaf in leaves:
+                    if leaf.op in ('>', '<', '>=', '<='):
+                        if_types["comparison"] += 1
+                    elif leaf.op in ('=', '<>'):
+                        if_types["equality"] += 1
+            _walk_if_types(node.true_seq, depth + 1)
+            _walk_if_types(node.false_seq, depth + 1)
+        elif isinstance(node, BrEval):
+            for _, seq in node.when_list:
+                _walk_if_types(seq, depth + 1)
+            _walk_if_types(node.other_seq, depth + 1)
+        elif isinstance(node, BrPerform):
+            _walk_if_types(node.body_seq, depth + 1)
+        elif isinstance(node, BrSeq):
+            for c in node.children:
+                _walk_if_types(c, depth + 1)
+
+    if branch_tree:
+        _walk_if_types(branch_tree)
+
+    # ── 新增字段: variable_patterns ──
+    variable_patterns = {
+        "has_prev_key": False,
+        "has_accumulator": False,
+        "has_error_flag": False,
+        "has_switch": False,
+        "has_index": False,
+        "has_save_area": False,
+        "has_counter": False,
+        "has_work": False,
+    }
+    for f in fields_dict:
+        name = f.get('name', '')
+        if re.search(r'\bWS-PREV[-_]', name, re.IGNORECASE):
+            variable_patterns["has_prev_key"] = True
+        if re.search(r'[-_]CNT\b', name, re.IGNORECASE) or re.search(r'[-_]ACCUM\b', name, re.IGNORECASE):
+            variable_patterns["has_accumulator"] = True
+        if re.search(r'[-_]ERR\b', name, re.IGNORECASE) or re.search(r'[-_]ERROR[-_]', name, re.IGNORECASE):
+            variable_patterns["has_error_flag"] = True
+        if re.search(r'[-_]SW\b', name, re.IGNORECASE) or re.search(r'[-_]FLAG\b', name, re.IGNORECASE):
+            variable_patterns["has_switch"] = True
+        if re.search(r'[-_]IDX\b', name, re.IGNORECASE) or re.search(r'[-_]INDX\b', name, re.IGNORECASE) or re.search(r'[-_]SUB\b', name, re.IGNORECASE):
+            variable_patterns["has_index"] = True
+        if re.search(r'[-_]SAVE[-_]', name, re.IGNORECASE) or re.search(r'[-_]HOLD[-_]', name, re.IGNORECASE):
+            variable_patterns["has_save_area"] = True
+        if re.search(r'[-_]CNT\b', name, re.IGNORECASE) or re.search(r'[-_]COUNT\b', name, re.IGNORECASE):
+            variable_patterns["has_counter"] = True
+        if name.startswith('WS-') and not re.search(r'(?:CNT|ERR|SW|IDX|INDX|SUB|SAVE|HOLD|PREV|ACCUM)', name, re.IGNORECASE):
+            if re.search(r'[-_]W\b|[-_]WORK\b|[-_]WK\b|^WS-W[0O]\w', name, re.IGNORECASE):
+                variable_patterns["has_work"] = True
+
+    # ── 新增字段: open_pattern ──
+    open_pattern = "sequential"
+    if proc_div:
+        proc_upper = proc_div.upper()
+        open_positions = [m.start() for m in re.finditer(r'\bOPEN\b', proc_upper)]
+        close_positions = [m.start() for m in re.finditer(r'\bCLOSE\b', proc_upper)]
+        if open_positions and close_positions:
+            # Check OPEN ... CLOSE ... OPEN sequence
+            for i, opos in enumerate(open_positions):
+                for cpos in close_positions:
+                    if cpos > opos:
+                        for opos2 in open_positions:
+                            if opos2 > cpos:
+                                open_pattern = "open-close-open"
+                                break
+                        if open_pattern == "open-close-open":
+                            break
+                if open_pattern == "open-close-open":
+                    break
+
     return {
-        "paragraphs": sorted(paragraphs) if paragraphs else [],
-        "decision_points": decision_points,
-        "branch_tree": branch_tree,
-        "file_count": len(file_sec) if file_sec else 0,
-        "open_directions": open_dir,
-        "has_search_all": any('SEARCH' in str(dp.get('label', '')) for dp in decision_points),
-        "has_evaluate": any(dp['kind'] == 'EVALUATE' for dp in decision_points),
-        "has_call": 'CALL' in cobol_source.upper(),
-        "has_break": any('KEY' in str(dp.get('label', '')).upper() for dp in decision_points),
-        "total_branches": total_branches,
-        "total_paragraphs": len(paragraphs),
-        "branch_tree_obj": branch_tree,
+"paragraphs": sorted(paragraphs) if paragraphs else [],
+"decision_points": decision_points,
+"branch_tree": branch_tree,
+"file_count": len(file_sec) if file_sec else 0,
+"open_directions": open_dir,
+"has_search_all": any('SEARCH' in str(dp.get('label', '')) for dp in decision_points),
+"has_evaluate": any(dp['kind'] == 'EVALUATE' for dp in decision_points),
+"has_call": 'CALL' in cobol_source.upper(),
+"has_break": any('KEY' in str(dp.get('label', '')).upper() for dp in decision_points),
+"total_branches": total_branches,
+"total_paragraphs": len(paragraphs),
+"branch_tree_obj": branch_tree,
+# ── 新增 8 类结构特征 ──
+"select_files": select_files,
+"open_directions_detail": open_directions_detail,
+"has_divide": has_divide,
+"divide_constants": divide_constants,
+"has_inspect": has_inspect,
+"has_string": has_string,
+"perform_patterns": perform_patterns,
+"main_loop": main_loop,
+"if_types": if_types,
+"variable_patterns": variable_patterns,
+"open_pattern": open_pattern,
     }
 
 

cobol_testgen/__main__.py

@@ -0,0 +1,4 @@
+"""允许 python -m cobol_testgen 直接运行"""
+
+from cobol_testgen import main
+main()

cobol_testgen/cond.py

@@ -0,0 +1,258 @@
+"""条件层：COBOL条件表达式解析 + MC/DC枚举 + 约束合并"""
+
+import re
+from .models import CondLeaf, CondAnd, CondOr, CondNot, PicInfo
+
+
+# ── 条件解析 ──
+
+def _split_at_operator(text, operator):
+    """Split text on operator word, respecting parentheses."""
+    result = []
+    current = []
+    depth = 0
+    # Normalize so parentheses are space-delimited tokens
+    normalized = text.replace('(', ' ( ').replace(')', ' ) ')
+    for token in normalized.split():
+        if not token:
+            continue
+        if token == '(':
+            depth += 1
+            current.append(token)
+        elif token == ')':
+            depth -= 1
+            current.append(token)
+        elif token == operator and depth == 0:
+            result.append(' '.join(current).strip())
+            current = []
+        else:
+            current.append(token)
+    result.append(' '.join(current).strip())
+    return result
+
+
+def parse_single_condition(text, fields=None):
+    """Parse 'AMOUNT > 1000' into ('AMOUNT', '>', '1000').
+    Also handles subscripted fields: 'WS-ITEM(SUB) = 'A''.
+    Also resolves 88-level condition names (e.g. STATUS-APPROVED → WS-TRAN-STATUS = 'A').
+    Returns None if the condition contains AND/OR (compound).
+    """
+    if ' AND ' in text or ' OR ' in text:
+        return None
+    # Check if text is an 88-level condition name
+    if fields:
+        for f in fields:
+            if f.get('is_88') and f['name'] == text.upper():
+                return (f.get('parent', ''), '=', f.get('value', ''))
+    m = re.match(
+        r"^(\w[\w-]*(?:\s*\([^)]*\))?)\s*(>=|<=|<>|>|<|=)\s*(.+)$",
+        text
+    )
+    if m:
+        field = re.sub(r'\s*([(),])\s*', r'\1', m.group(1))
+        return (field, m.group(2), m.group(3).strip().strip("'").strip('"'))
+    # Try arithmetic expression: e.g. A + B > C
+    m = re.match(
+        r"^(\w[\w\s+\-*/().-]+?)\s*(>=|<=|<>|>|<|=)\s*(.+)$",
+        text
+    )
+    if m:
+        field = re.sub(r'\s*([(),])\s*', r'\1', m.group(1)).strip()
+        return (field, m.group(2), m.group(3).strip().strip("'").strip('"'))
+    return None
+
+
+def parse_compound_condition(text, fields=None):
+    """Parse a COBOL condition into a condition tree (AND/OR/LEAF).
+    Handles AND > OR precedence and parentheses.
+    """
+    text = text.strip()
+    if not text:
+        return None
+    # Normalize parentheses to be space-delimited for reliable tokenization
+    text = text.replace('(', ' ( ').replace(')', ' ) ')
+    text = re.sub(r'\s+', ' ', text).strip()
+    # Strip outer parentheses
+    if text.startswith('(') and text.endswith(')'):
+        depth = 0
+        wrapped = True
+        for i, c in enumerate(text):
+            if c == '(':
+                depth += 1
+            elif c == ')':
+                depth -= 1
+                if depth == 0 and i < len(text) - 1:
+                    wrapped = False
+                    break
+        if wrapped:
+            inner = parse_compound_condition(text[1:-1], fields)
+            if inner:
+                return inner
+    # Split on OR (lowest precedence)
+    parts = _split_at_operator(text, 'OR')
+    if len(parts) > 1:
+        node = parse_compound_condition(parts[0], fields)
+        for p in parts[1:]:
+            node = CondOr(node, parse_compound_condition(p, fields))
+        return node
+    # Split on AND
+    parts = _split_at_operator(text, 'AND')
+    if len(parts) > 1:
+        node = parse_compound_condition(parts[0], fields)
+        for p in parts[1:]:
+            node = CondAnd(node, parse_compound_condition(p, fields))
+        return node
+    # NOT prefix (highest precedence, after AND/OR splitting)
+    if text.upper().startswith('NOT '):
+        inner = parse_compound_condition(text[4:].strip(), fields)
+        return CondNot(inner) if inner else None
+    # Leaf condition
+    parsed = parse_single_condition(text, fields)
+    if parsed:
+        return CondLeaf(*parsed)
+    return None
+
+
+def collect_leaves(tree):
+    """Return list of all CondLeaf nodes in the tree."""
+    if isinstance(tree, CondLeaf):
+        return [tree]
+    elif isinstance(tree, CondNot):
+        return collect_leaves(tree.child)
+    elif isinstance(tree, (CondAnd, CondOr)):
+        return collect_leaves(tree.left) + collect_leaves(tree.right)
+    return []
+
+
+def evaluate_tree(tree, assignment):
+    """Evaluate condition tree given leaf→bool assignment dict."""
+    if isinstance(tree, CondLeaf):
+        return assignment[tree]
+    elif isinstance(tree, CondNot):
+        return not evaluate_tree(tree.child, assignment)
+    elif isinstance(tree, CondAnd):
+        return evaluate_tree(tree.left, assignment) and evaluate_tree(tree.right, assignment)
+    elif isinstance(tree, CondOr):
+        return evaluate_tree(tree.left, assignment) or evaluate_tree(tree.right, assignment)
+    return False
+
+
+def is_field(name, fields):
+    # Strip subscript: WS-ITEM-STATUS(WS-INDEX-VAR) -> WS-ITEM-STATUS
+    bare = re.sub(r'\s*\(.*\)\s*$', '', name).strip()
+    for f in fields:
+        if f['name'] == bare.upper():
+            return True
+    return False
+
+
+# ── MC/DC ──
+
+def mcdc_sets(tree, fields=None):
+    """Generate MC/DC constraint sets.
+    Returns list of (constraints_list, decision_outcome) or None for simple conditions.
+    Each constraint is (field, op, value, want_true).
+    """
+    leaves = collect_leaves(tree)
+    n = len(leaves)
+    if n <= 1:
+        return None
+    # Evaluate all 2^n truth assignments
+    all_results = []
+    for bits in range(1 << n):
+        assignment = {}
+        for i, leaf in enumerate(leaves):
+            assignment[leaf] = bool(bits & (1 << i))
+        result = evaluate_tree(tree, assignment)
+        all_results.append((assignment, result))
+    # For each leaf, find a pair showing independent effect on decision
+    needed_pairs = {}
+    for leaf in leaves:
+        for a1, r1 in all_results:
+            if leaf in needed_pairs:
+                break
+            for a2, r2 in all_results:
+                if a1[leaf] != a2[leaf] and r1 != r2:
+                    if all(a1[o] == a2[o] for o in leaves if o != leaf):
+                        needed_pairs[leaf] = (dict(a1), r1, dict(a2), r2)
+                        break
+    # Convert leaf assignments to constraint tuples
+    result = []
+    added = set()
+    for leaf, (a1, r1, a2, r2) in needed_pairs.items():
+        for assignment, decision in [(a1, r1), (a2, r2)]:
+            key = frozenset((l, assignment[l]) for l in leaves)
+            if key not in added:
+                added.add(key)
+                constraints = []
+                for l in leaves:
+                    want = assignment[l]
+                    constraints.append((l.field, l.op, l.value, want))
+                result.append((constraints, decision))
+    return result
+
+
+
+
+
+# ── 值计算 ──
+
+def satisfying_value(field_info: dict, operator: str, value, want_true: bool) -> str:
+    ftype = field_info.get('type', 'unknown')
+    digits = field_info.get('digits', 0)
+    decimal = field_info.get('decimal', 0)
+    total = digits + decimal
+
+    if ftype == 'numeric':
+        try:
+            val_str = str(value)
+            val_float = float(val_str)
+            val_int = int(val_float * (10 ** decimal) + 0.5)
+        except (ValueError, TypeError):
+            val_int = 0
+
+        if want_true:
+            if operator == '>':
+                val_int = val_int + 1
+            elif operator in ('>=', '=', '<='):
+                pass
+            elif operator == '<':
+                val_int = max(0, val_int - 1)
+            elif operator == '<>':
+                val_int = (val_int + 1) % (10 ** total)
+        else:
+            if operator in ('>', '>='):
+                val_int = 0
+            elif operator == '=':
+                val_int = (val_int + 1) % (10 ** total)
+            elif operator == '<':
+                pass
+            elif operator == '<=':
+                val_int = val_int + 1
+            elif operator == '<>':
+                pass
+
+        val_int = val_int % (10 ** total)
+        int_part = str(val_int // (10 ** decimal)).zfill(digits)
+        dec_part = str(val_int % (10 ** decimal)).zfill(decimal)
+        if decimal == 0:
+            return int_part
+        return int_part + dec_part
+
+    elif ftype in ('alphanumeric', 'alphabetic'):
+        length = field_info.get('length', 1)
+        base_chr = value[0].upper() if isinstance(value, str) and value else 'A'
+        if want_true:
+            if operator in ('=', '=='):
+                return base_chr.ljust(length, base_chr)
+            elif operator in ('<>', '!='):
+                other = chr(65 + (ord(base_chr) - 64) % 26)
+                return other.ljust(length, other)
+        else:
+            if operator in ('=', '=='):
+                other = chr(65 + (ord(base_chr) - 64) % 26)
+                return other.ljust(length, other)
+            elif operator in ('<>', '!='):
+                return base_chr.ljust(length, base_chr)
+
+    return '0'.zfill(total)

cobol_testgen/core.py

@@ -0,0 +1,1649 @@
+"""核心层：PROCEDURE DIVISION解析 + 数据流追踪"""
+
+import re
+import logging
+from datetime import datetime
+from .models import BrSeq, BrIf, BrEval, BrPerform, BrSearch, BrSeq, CondLeaf, CondNot, ParseError, Assign, CallNode, ExitNode, GoTo
+from .cond import parse_compound_condition, parse_single_condition, collect_leaves
+
+logger = logging.getLogger(__name__)
+
+
+_COBOL_SCOPE_ENDERS = {
+    'END-IF', 'END-EVALUATE', 'END-PERFORM', 'END-EXEC', 'END-CALL',
+    'END-READ', 'END-WRITE', 'END-DELETE', 'END-REWRITE', 'END-START',
+    'END-SEARCH',
+    'ELSE', 'WHEN', 'OTHER',
+}
+
+
+def scan_paragraphs(raw_lines):
+    paragraphs = {}
+    i = 0
+    while i < len(raw_lines):
+        line = raw_lines[i].strip()
+        m = re.match(r'^([A-Z0-9][A-Z0-9-]*)\.\s*$', line)
+        sec_m = re.match(r'^([A-Z][A-Z0-9-]*)\s+SECTION\.?\s*$', line, re.IGNORECASE)
+        if m and m.group(1) not in _COBOL_SCOPE_ENDERS:
+            name = m.group(1)
+        elif sec_m:
+            name = sec_m.group(1).upper()
+        else:
+            i += 1
+            continue
+        start = i + 1
+        j = i + 1
+        while j < len(raw_lines):
+            nline = raw_lines[j].strip()
+            nm = re.match(r'^([A-Z0-9][A-Z0-9-]*)\.\s*$', nline)
+            if nm and nm.group(1) not in _COBOL_SCOPE_ENDERS:
+                break
+            if re.match(r'^[A-Z][A-Z0-9-]*\s+SECTION\.\s*$', nline, re.IGNORECASE):
+                break
+            j += 1
+        paragraphs[name] = (start, j - 1)
+        i = j
+    return paragraphs
+
+
+def build_branch_tree(proc_text, fields=None):
+    raw_lines = proc_text.split('\n')
+    paragraphs = scan_paragraphs(raw_lines)
+
+    first_para_name = None
+    first_para_idx = None
+    for i, line in enumerate(raw_lines):
+        clean = line.strip()
+        m = re.match(r'^([A-Z0-9][A-Z0-9-]*)\.\s*$', clean)
+        if m and m.group(1) in paragraphs:
+            first_para_name = m.group(1)
+            first_para_idx = i
+            break
+
+    if first_para_name:
+        before = raw_lines[:first_para_idx]
+        has_code = any(
+            l.strip() and 'PROCEDURE DIVISION' not in l
+            for l in before
+        )
+        if has_code:
+            main_raw = raw_lines[:first_para_idx]
+        else:
+            p_start, p_end = paragraphs[first_para_name]
+            main_raw = raw_lines[p_start:p_end + 1]
+    else:
+        main_raw = raw_lines
+
+    filtered = [l for l in main_raw if l.strip()]
+    assignments = {}
+    parser = _BrParser(filtered, paragraphs, raw_lines, assignments, fields)
+    tree = parser.parse_seq(terminators={'GOBACK', 'STOP RUN', 'EXIT PROGRAM'})
+    return tree, assignments
+
+
+# ── 定数 ──
+
+_FIGURATIVE_CONSTANTS = frozenset({
+    'ZERO', 'ZEROS', 'ZEROES',
+    'SPACE', 'SPACES',
+    'HIGH-VALUE', 'HIGH-VALUES',
+    'LOW-VALUE', 'LOW-VALUES',
+})
+
+
+# ── _BrParser ──
+
+class _BrParser:
+    def __init__(self, lines, paragraphs=None, raw_lines=None, assignments=None, fields=None, goto_depth=0):
+        self.lines = lines
+        self.pos = 0
+        self.paragraphs = paragraphs or {}
+        self.raw_lines = raw_lines or lines
+        # assignments is a dict[str, list[dict]] — append, never overwrite
+        self.assignments = assignments if assignments is not None else {}
+        self.fields = fields
+        self._goto_depth = goto_depth
+
+    def peek(self):
+        if self.pos < len(self.lines):
+            return self.lines[self.pos].strip()
+        return ''
+
+    def clean(self):
+        return self.peek().rstrip('.').strip()
+
+    def advance(self):
+        self.pos += 1
+
+    def parse_seq(self, end_tokens=None, end_check=None, terminators=None):
+        if end_tokens is None:
+            end_tokens = []
+        seq = BrSeq()
+        while self.pos < len(self.lines):
+            line = self.clean()
+            if self._is_end(line, end_tokens, end_check):
+                return seq
+            if terminators and line in terminators:
+                self.advance()
+                return seq
+            m_goto = re.match(r'^GO\s+TO\s+(\w[\w-]*)\s*$', line)
+            if m_goto:
+                goto_node = self._parse_goto(m_goto.group(1))
+                if goto_node:
+                    seq.add(goto_node)
+                while self.pos < len(self.lines):
+                    cl = self.clean()
+                    if self._is_end(cl, end_tokens, end_check):
+                        break
+                    if cl in _COBOL_SCOPE_ENDERS:
+                        break
+                    self.advance()
+                return seq
+            m_exit = re.match(r'^EXIT\s+(PARAGRAPH|PERFORM|SECTION)\s*$', line)
+            if m_exit:
+                self.advance()
+                seq.add(ExitNode(m_exit.group(1)))
+                while self.pos < len(self.lines):
+                    cl = self.clean()
+                    if self._is_end(cl, end_tokens, end_check):
+                        break
+                    if cl in _COBOL_SCOPE_ENDERS:
+                        break
+                    self.advance()
+                return seq
+            m = re.match(r'^IF\s+(.+?)(?:THEN)?\s*$', line)
+            if m:
+                seq.add(self._parse_if())
+                continue
+            m = re.match(r'^EVALUATE\s+(.+?)\s*$', line)
+            if m:
+                seq.add(self._parse_evaluate())
+                continue
+            m = re.match(r'^PERFORM\s+', line)
+            if m:
+                perf_node = self._parse_perform()
+                if perf_node:
+                    seq.add(perf_node)
+                continue
+            m_search = re.match(r'^SEARCH\b(?:\s+(ALL))?\s+(\w[\w-]*)(?:\s+VARYING\s+(\w[\w-]*))?', line, re.IGNORECASE)
+            if m_search:
+                seq.add(self._parse_search(m_search))
+                continue
+            m = re.match(r'^INITIALIZE\s+', line)
+            if m:
+                init_seq = self._parse_initialize()
+                if init_seq:
+                    seq.add(init_seq)
+                continue
+            m_str = re.match(r'^STRING\s+', line)
+            if m_str:
+                str_seq = self._parse_string()
+                if str_seq:
+                    seq.add(str_seq)
+                continue
+            m_unstr = re.match(r'^UNSTRING\s+', line)
+            if m_unstr:
+                unstr_seq = self._parse_unstring()
+                if unstr_seq:
+                    seq.add(unstr_seq)
+                continue
+            m = re.match(r'^CALL\s+', line)
+            if m:
+                seq.add(self._parse_call())
+                continue
+            m = re.match(
+                r'^ACCEPT\s+(\w[\w-]*)(?:\s+FROM\s+(DATE|TIME|DAY|DAY-OF-WEEK|YEAR|YYYYMMDD|HHMMSS))?\s*$',
+                line, re.IGNORECASE
+            )
+            if m:
+                tgt = m.group(1).strip().upper()
+                from_type = (m.group(2) or 'USER').upper()
+                info = {'type': 'accept', 'from': from_type}
+                self.assignments.setdefault(tgt, []).append(info)
+                seq.add(Assign(tgt, info))
+                self.advance()
+                continue
+            m = re.match(r'^READ\s+(\w[\w-]*)\s+INTO\s+(\w[\w-]*)\s*$', line, re.IGNORECASE)
+            if m:
+                tgt = m.group(2).strip().upper()
+                info = {'type': 'read_into', 'file': m.group(1).strip().upper(), 'source_vars': []}
+                self.assignments.setdefault(tgt, []).append(info)
+                seq.add(Assign(tgt, info))
+                self.advance()
+                # 跳过 READ 语句剩余行（AT END / NOT AT END / END-READ）
+                while self.pos < len(self.lines):
+                    cl = self.clean()
+                    if cl in ('END-READ', 'END-READ.'):
+                        self.advance()
+                        break
+                    self.advance()
+                continue
+            m_set_false = re.match(r'^SET\s+(\w[\w-]*)\s+TO\s+FALSE\s*$', line, re.IGNORECASE)
+            if m_set_false:
+                seq.add(self._parse_set_false(m_set_false.group(1)))
+                continue
+            m = re.match(r'^(?:WRITE|REWRITE)\s+(\w[\w-]*)(?:\s+FROM\s+(\w[\w-]*))?\s*$', line, re.IGNORECASE)
+            if m:
+                rec_name = m.group(1).strip().upper()
+                if m.group(2):
+                    tgt = m.group(2).strip().upper()
+                    info = {'type': 'write_from', 'file': rec_name, 'source_vars': [tgt]}
+                    self.assignments.setdefault(tgt, []).append(info)
+                    seq.add(Assign(tgt, info))
+                else:
+                    seq.add(Assign(rec_name, {'type': 'write_bare', 'file': rec_name}))
+                self.advance()
+                continue
+            m_set = re.match(r'^SET\s+(\w[\w-]*)\s+TO\s+TRUE\s*$', line, re.IGNORECASE)
+            if m_set:
+                seq.add(self._parse_set_true(m_set.group(1)))
+                continue
+            m_insp = re.match(r'^INSPECT\s+', line, re.IGNORECASE)
+            if m_insp:
+                info = self._parse_inspect(line)
+                if info:
+                    tgt = info.get('tgt', '')
+                    self.assignments.setdefault(tgt, []).append(info)
+                    seq.add(Assign(tgt, info))
+                self.advance()
+                continue
+            assign_node = self._record_assignment(line)
+            if assign_node:
+                seq.add(assign_node)
+            self.advance()
+        return seq
+
+    def _is_end(self, line, end_tokens, end_check):
+        if end_check and end_check(line):
+            return True
+        for tok in end_tokens:
+            if line == tok or line.startswith(tok + ' '):
+                return True
+        return False
+
+    # ── INSPECT ──
+
+    _PIC_FIG_CONV = {'ZERO': '0', 'ZEROS': '0', 'ZEROES': '0',
+                     'SPACE': ' ', 'SPACES': ' '}
+
+    @staticmethod
+    def _expand_figurative(val):
+        if val.upper() in _BrParser._PIC_FIG_CONV:
+            return _BrParser._PIC_FIG_CONV[val.upper()]
+        return val
+
+    def _parse_inspect_phrase(self, phrase):
+        m = re.match(
+            r'TALLYING\s+(\w[\w-]*)\s+FOR\s+'
+            r'(LEADING|TRAILING|CHARACTERS)'
+            r'(?:\s+([\'"])(.*?)\3)?'
+            r'(?:\s+(BEFORE|AFTER)\s+INITIAL\s+([\'"])(.*?)\6)?\s*$',
+            phrase, re.IGNORECASE
+        )
+        if m:
+            return ('tally', {
+                'count_var': m.group(1).upper(),
+                'kind': m.group(2).upper(),
+                'char': self._expand_figurative(m.group(4) or ''),
+                'before_after': (m.group(5) or '').upper(),
+                'delimiter': self._expand_figurative(m.group(7) or ''),
+            })
+        m = re.match(
+            r'REPLACING\s+'
+            r'(ALL|LEADING|FIRST|CHARACTERS)\s+'
+            r'([\'"])(.*?)\2\s+BY\s+'
+            r'([\'"])(.*?)\4'
+            r'(?:\s+(BEFORE|AFTER)\s+INITIAL\s+([\'"])(.*?)\7)?\s*$',
+            phrase, re.IGNORECASE
+        )
+        if m:
+            return ('replace', {
+                'kind': m.group(1).upper(),
+                'src': self._expand_figurative(m.group(3)),
+                'dst': self._expand_figurative(m.group(5)),
+                'before_after': (m.group(6) or '').upper(),
+                'delimiter': self._expand_figurative(m.group(8) or ''),
+            })
+        m = re.match(
+            r'CONVERTING\s+([\'"])(.*?)\1\s+TO\s+([\'"])(.*?)\3\s*$',
+            phrase, re.IGNORECASE
+        )
+        if m:
+            return ('convert', {
+                'from_chars': self._expand_figurative(m.group(2)),
+                'to_chars': self._expand_figurative(m.group(4)),
+            })
+        return None
+
+    def _parse_inspect(self, line):
+        m = re.match(r'^INSPECT\s+(\w[\w-]*)\s+(.+)$', line, re.IGNORECASE)
+        if not m:
+            return None
+        tgt = m.group(1).upper()
+        rest = m.group(2).strip()
+        phrases = re.split(r'\s+(?=(?:TALLYING|REPLACING|CONVERTING)\b)', rest, flags=re.IGNORECASE)
+        sub_ops = []
+        for phrase in phrases:
+            sub = self._parse_inspect_phrase(phrase.strip())
+            if sub:
+                sub_ops.append(sub)
+        if not sub_ops:
+            return None
+        return {
+            'type': 'inspect',
+            'tgt': tgt,
+            'source_vars': [tgt],
+            'sub_ops': sub_ops,
+        }
+
+    def _record_assignment(self, line):
+        if self.assignments is None:
+            return None
+
+        # MOVE
+        m = re.match(r'^MOVE\s+(.+?)\s+TO\s+(.+?)\s*$', line)
+        if m:
+            raw_src = m.group(1).strip()
+            tgt = m.group(2).strip()
+            # 保留下标：WS-CODE-VAL(1) → key='WS-CODE-VAL(1)'
+            m_tgt = re.match(r'^([A-Z][A-Z0-9-]*)(?:\s*\(([^)]*)\))?\s*$', tgt, re.IGNORECASE)
+            if not m_tgt:
+                return None
+            tgt_base = m_tgt.group(1).upper()
+            if m_tgt.group(2):
+                subscript = re.sub(r'\s*', '', m_tgt.group(2))
+                tgt_key = f"{tgt_base}({subscript})"
+            else:
+                tgt_key = tgt_base
+            src_clean = raw_src.strip("'").strip('"')
+            is_field_name = self.fields and any(f['name'] == src_clean for f in self.fields)
+            if is_field_name:
+                info = {'type': 'move', 'source_vars': [src_clean]}
+            else:
+                info = {'type': 'move_literal', 'literal': src_clean}
+            self.assignments.setdefault(tgt_key, []).append(info)
+            return Assign(tgt_key, info)
+
+        # COMPUTE
+        m = re.match(r'^COMPUTE\s+(.+?)(?:\s+ROUNDED)?\s*=\s*(.*)$', line)
+        if m:
+            tgt_raw = m.group(1).strip()
+            expr = m.group(2).strip()
+            m_tgt = re.match(r'^([A-Z][A-Z0-9-]*)(?:\s*\(([^)]*)\))?\s*$', tgt_raw, re.IGNORECASE)
+            tgt_key = tgt_raw
+            if m_tgt:
+                tgt_base = m_tgt.group(1).upper()
+                if m_tgt.group(2):
+                    subscript = re.sub(r'\s*', '', m_tgt.group(2))
+                    tgt_key = f"{tgt_base}({subscript})"
+                else:
+                    tgt_key = tgt_base
+            if not expr:
+                peek_pos = self.pos + 1
+                if peek_pos < len(self.lines):
+                    nxt = self.lines[peek_pos].strip().rstrip('.').strip()
+                    if nxt and not re.match(r'^(PERFORM|END-|IF|ELSE|EVALUATE|WHEN|OTHER|MOVE|COMPUTE|ADD|SUBTRACT|MULTIPLY|DIVIDE|STRING|UNSTRING|READ|WRITE|INITIALIZE|ACCEPT|CALL|GO\s*TO|GOBACK|STOP|EXIT)', nxt, re.IGNORECASE):
+                        expr = nxt
+            if expr:
+                info = self._parse_compute_expr(tgt_key, expr)
+                self.assignments.setdefault(tgt_key, []).append(info)
+                return Assign(tgt_key, info)
+
+        # ADD x TO y  → y = y + x  (支持变量和常量源)
+        m = re.match(r'^ADD\s+(\w[\w-]*)\s+TO\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line)
+        if m:
+            src = m.group(1).strip()
+            tgt = m.group(2).strip()
+            is_field = self.fields and any(f['name'] == src for f in self.fields)
+            if is_field:
+                info = {'type': 'compute', 'source_vars': [tgt, src],
+                        'op': '+', 'const': None, 'expr': f'{tgt} + {src}'}
+            else:
+                try:
+                    const = float(src)
+                    info = {'type': 'compute', 'source_vars': [tgt],
+                            'op': '+', 'const': const, 'expr': f'{tgt} + {const}'}
+                except ValueError:
+                    return None
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # ADD x TO y GIVING z  → z = y + x
+        m = re.match(r'^ADD\s+(.+?)\s+TO\s+(\w[\w-]*)\s+GIVING\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line, re.IGNORECASE)
+        if m:
+            raw_a = m.group(1).strip()
+            src_b = m.group(2).strip()
+            tgt = m.group(3).strip()
+            is_field_a = self.fields and any(f['name'] == raw_a for f in self.fields)
+            if is_field_a:
+                info = {'type': 'compute', 'source_vars': [src_b, raw_a],
+                        'op': '+', 'const': None, 'expr': f'{src_b} + {raw_a}'}
+            else:
+                try:
+                    const = float(raw_a)
+                    info = {'type': 'compute', 'source_vars': [src_b],
+                            'op': '+', 'const': const, 'expr': f'{src_b} + {const}'}
+                except ValueError:
+                    return None
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # ADD a[, b[, c...]] GIVING z  → z = a + b + c + ...
+        m = re.match(r'^ADD\s+(.+?)\s+GIVING\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line, re.IGNORECASE)
+        if m:
+            raw_parts = re.findall(r'[A-Z][A-Z0-9-]*|\d+(?:\.\d+)?', m.group(1).upper())
+            fields_only = []
+            const_sum = 0.0
+            for p in raw_parts:
+                if self.fields and any(f['name'] == p for f in self.fields):
+                    fields_only.append(p)
+                else:
+                    try:
+                        const_sum += float(p)
+                    except ValueError:
+                        pass
+            tgt = m.group(2).strip()
+            if not fields_only:
+                info = {'type': 'move_literal',
+                        'literal': str(int(const_sum)) if const_sum == int(const_sum) else str(const_sum)}
+            else:
+                info = {'type': 'compute', 'source_vars': fields_only,
+                        'op': '+', 'const': const_sum if const_sum != 0 else None,
+                        'expr': '+'.join(fields_only) + (f' + {const_sum}' if const_sum else '')}
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # SUBTRACT x FROM y  → y = y - x
+        m = re.match(r'^SUBTRACT\s+([\d.]+)\s+FROM\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line)
+        if m:
+            const = float(m.group(1))
+            tgt = m.group(2).strip()
+            info = {'type': 'compute', 'source_vars': [tgt],
+                    'op': '-', 'const': const, 'expr': f'{tgt} - {const}'}
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # SUBTRACT a FROM b GIVING z  → z = b - a
+        m = re.match(r'^SUBTRACT\s+([\d.\w-]*)\s+FROM\s+(\w[\w-]*)\s+GIVING\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line, re.IGNORECASE)
+        if m:
+            raw_a = m.group(1).strip()
+            src_b = m.group(2).strip()
+            tgt = m.group(3).strip()
+            is_field_a = self.fields and any(f['name'] == raw_a for f in self.fields)
+            if is_field_a:
+                info = {'type': 'compute', 'source_vars': [src_b, raw_a],
+                        'op': '-', 'const': None, 'expr': f'{src_b} - {raw_a}'}
+            else:
+                try:
+                    const = float(raw_a)
+                    info = {'type': 'compute', 'source_vars': [src_b],
+                            'op': '-', 'const': const, 'expr': f'{src_b} - {const}'}
+                except ValueError:
+                    return None
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # MULTIPLY x BY y  → y = y * x
+        m = re.match(r'^MULTIPLY\s+([\d.]+)\s+BY\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line)
+        if m:
+            const = float(m.group(1))
+            tgt = m.group(2).strip()
+            info = {'type': 'compute', 'source_vars': [tgt],
+                    'op': '*', 'const': const, 'expr': f'{tgt} * {const}'}
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # MULTIPLY a BY b GIVING z  → z = a * b
+        m = re.match(r'^MULTIPLY\s+(\w[\w-]*)\s+BY\s+(\w[\w-]*)\s+GIVING\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line, re.IGNORECASE)
+        if m:
+            src_a = m.group(1).strip()
+            src_b = m.group(2).strip()
+            tgt = m.group(3).strip()
+            is_field_a = self.fields and any(f['name'] == src_a for f in self.fields)
+            if is_field_a:
+                info = {'type': 'compute', 'source_vars': [src_a, src_b],
+                        'op': '*', 'const': None, 'expr': f'{src_a} * {src_b}'}
+            else:
+                try:
+                    const = float(src_a)
+                    info = {'type': 'compute', 'source_vars': [src_b],
+                            'op': '*', 'const': const, 'expr': f'{const} * {src_b}'}
+                except ValueError:
+                    return None
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # DIVIDE x INTO y  → y = y / x
+        m = re.match(r'^DIVIDE\s+([\d.]+)\s+INTO\s+(\w[\w-]*?)(?:\s+ROUNDED)?\s*$', line)
+        if m:
+            const = float(m.group(1))
+            tgt = m.group(2).strip()
+            info = {'type': 'compute', 'source_vars': [tgt],
+                    'op': '/', 'const': const, 'expr': f'{tgt} / {const}'}
+            self.assignments.setdefault(tgt, []).append(info)
+            return Assign(tgt, info)
+
+        # DIVIDE a INTO b GIVING z  → z = b / a
+        # Optional REMAINDER r → r = b - (b / a) * a
+        m = re.match(r'^DIVIDE\s+(.+?)\s+INTO\s+(\w[\w-]*)\s+GIVING\s+(\w[\w-]*?)(?:\s+ROUNDED)?(?:\s+REMAINDER\s+(\w[\w-]*))?\s*$', line, re.IGNORECASE)
+        if m:
+            raw_a = m.group(1).strip()
+            src_b = m.group(2).strip()
+            tgt = m.group(3).strip()
+            rem_tgt = m.group(4).strip().upper() if m.group(4) else None
+            is_field_a = self.fields and any(f['name'] == raw_a for f in self.fields)
+            if is_field_a:
+                info = {'type': 'compute', 'source_vars': [src_b, raw_a],
+                        'op': '/', 'const': None, 'expr': f'{src_b} / {raw_a}'}
+                rem_info = {'type': 'compute', 'source_vars': [src_b, raw_a],
+                            'op': 'rem', 'const': None, 'expr': f'REM({src_b} / {raw_a})'}
+            else:
+                try:
+                    const = float(raw_a)
+                    info = {'type': 'compute', 'source_vars': [src_b],
+                            'op': '/', 'const': const, 'expr': f'{src_b} / {const}'}
+                    rem_info = {'type': 'compute', 'source_vars': [src_b],
+                                'op': 'rem', 'const': const, 'expr': f'REM({src_b} / {const})'}
+                except ValueError:
+                    return None
+            self.assignments.setdefault(tgt, []).append(info)
+            seq = BrSeq()
+            seq.add(Assign(tgt, info))
+            if rem_tgt:
+                self.assignments.setdefault(rem_tgt, []).append(rem_info)
+                seq.add(Assign(rem_tgt, rem_info))
+            return seq
+
+        # DIVIDE a BY b GIVING z  → z = a / b
+        # Optional REMAINDER r → r = a - (a / b) * b
+        m = re.match(r'^DIVIDE\s+(\w[\w-]*)\s+BY\s+(\w[\w-]*)\s+GIVING\s+(\w[\w-]*?)(?:\s+ROUNDED)?(?:\s+REMAINDER\s+(\w[\w-]*))?\s*$', line, re.IGNORECASE)
+        if m:
+            src_a = m.group(1).strip()
+            src_b = m.group(2).strip()
+            tgt = m.group(3).strip()
+            rem_tgt = m.group(4).strip().upper() if m.group(4) else None
+            info = {'type': 'compute', 'source_vars': [src_a, src_b],
+                    'op': '/', 'const': None, 'expr': f'{src_a} / {src_b}'}
+            rem_info = {'type': 'compute', 'source_vars': [src_a, src_b],
+                        'op': 'rem', 'const': None, 'expr': f'REM({src_a} / {src_b})'}
+            self.assignments.setdefault(tgt, []).append(info)
+            seq = BrSeq()
+            seq.add(Assign(tgt, info))
+            if rem_tgt:
+                self.assignments.setdefault(rem_tgt, []).append(rem_info)
+                seq.add(Assign(rem_tgt, rem_info))
+            return seq
+
+        return None
+
+    def _parse_compute_expr(self, target, expr):
+        # const OP var
+        m = re.match(r'^\s*([\d.]+)\s*([+\-*/])\s*(\w[\w-]*)\s*$', expr)
+        if m:
+            const, op, var = float(m.group(1)), m.group(2), m.group(3)
+            return {'type': 'compute', 'source_vars': [var], 'op': op, 'const': const, 'expr': expr}
+        # var OP const
+        m = re.match(r'^\s*(\w[\w-]*)\s*([+\-*/])\s*([\d.]+)\s*$', expr)
+        if m:
+            var, op, const = m.group(1), m.group(2), float(m.group(3))
+            return {'type': 'compute', 'source_vars': [var], 'op': op, 'const': const, 'expr': expr}
+        # var OP var
+        m = re.match(r'^\s*(\w[\w-]*)\s*([+\-*/])\s*(\w[\w-]*)\s*$', expr)
+        if m:
+            var1, op, var2 = m.group(1), m.group(2), m.group(3)
+            return {'type': 'compute', 'source_vars': [var1, var2], 'op': op, 'expr': expr}
+        # complex expression — extract variable names only
+        vars_in = re.findall(r'[A-Z][A-Z0-9-]*', expr.upper())
+        return {'type': 'compute', 'source_vars': list(set(vars_in)), 'op': None, 'const': None, 'expr': expr}
+
+    # ── SEARCH / SEARCH ALL ──
+
+    def _parse_search(self, m):
+        is_all = bool(m.group(1))
+        table = m.group(2).upper()
+        varying = m.group(3).upper() if m.group(3) else None
+        node = BrSearch(table, is_all=is_all, varying=varying)
+        self.advance()
+        while self.pos < len(self.lines):
+            line = self.clean()
+            if line in ('END-SEARCH', 'END-SEARCH.'):
+                self.advance()
+                return node
+            m_at = re.match(r'^AT\s+END(.+)?$', line, re.IGNORECASE)
+            if m_at:
+                self.advance()
+                rest = m_at.group(1)
+                if rest and rest.strip():
+                    self.lines.insert(self.pos, rest.strip())
+                node.at_end_seq = self.parse_seq(
+                    end_check=lambda l: re.match(r'^WHEN\b', l) or l in ('END-SEARCH',)
+                )
+                node.has_at_end = True
+                continue
+            m_when = re.match(r'^WHEN\s+(.+?)\s*$', line, re.IGNORECASE)
+            if m_when:
+                cond_upper = m_when.group(1).strip()
+                self.advance()
+                cond_tree = parse_compound_condition(cond_upper, self.fields)
+                body_seq = self.parse_seq(
+                    end_check=lambda l: re.match(r'^(WHEN|AT\s+END)\b', l) or l in ('END-SEARCH',)
+                )
+                node.when_list.append((cond_upper, body_seq))
+                node.cond_trees.append(cond_tree)
+                continue
+            self.advance()
+        return node
+
+    def _parse_if(self):
+        line = self.clean()
+        m = re.match(r'^IF\s+(.+?)(?:THEN)?\s*$', line)
+        cond_text = m.group(1).strip()
+        self.advance()
+        # Join continuation lines (multi-line IF conditions)
+        while self.pos < len(self.lines):
+            peek = self.clean()
+            if re.match(r'^(THEN|ELSE|END-IF|MOVE|IF|PERFORM|EVALUATE|COMPUTE|CALL|STRING|UNSTRING|INITIALIZE|ADD|SUBTRACT|MULTIPLY|DIVIDE|GO\b|EXIT\b)', peek, re.IGNORECASE):
+                break
+            if peek.endswith('.'):
+                cond_text += ' ' + peek.rstrip('.')
+                self.advance()
+                break
+            cond_text += ' ' + peek
+            self.advance()
+        # Consume optional THEN on its own line
+        if self.pos < len(self.lines):
+            peek = self.clean()
+            if peek == 'THEN':
+                self.advance()
+        node = BrIf(cond_text)
+        node.cond_tree = parse_compound_condition(node.condition, self.fields)
+        node.true_seq = self.parse_seq(['ELSE', 'END-IF'])
+        if self.clean() == 'ELSE':
+            self.advance()
+            node.false_seq = self.parse_seq(['END-IF'])
+        if self.clean() == 'END-IF':
+            self.advance()
+        return node
+
+    def _parse_evaluate(self):
+        line = self.clean()
+        m = re.match(r'^EVALUATE\s+(.+?)\s*$', line)
+        raw_subject = m.group(1).strip()
+        node = BrEval(raw_subject)
+        if ' ALSO ' in raw_subject:
+            node.subjects = [s.strip() for s in re.split(r'\s+ALSO\s+', raw_subject)]
+        self.advance()
+        while self.pos < len(self.lines):
+            line = self.clean()
+            if line == 'END-EVALUATE':
+                self.advance()
+                return node
+            m = re.match(r'^WHEN\s+(.+?)\s*$', line)
+            if m:
+                raw_val = m.group(1).strip().strip("'").strip('"')
+                self.advance()
+                # Capture multi-line WHEN conditions (AND/OR continuation)
+                while self.pos < len(self.lines):
+                    peek = self.clean()
+                    if re.match(r'^(?:AND|OR)\b', peek, re.IGNORECASE):
+                        raw_val += ' ' + peek
+                        self.advance()
+                    else:
+                        break
+                if raw_val == 'OTHER':
+                    node.other_seq = self.parse_seq(end_check=lambda l: l == 'END-EVALUATE')
+                    node.has_other = True
+                else:
+                    case_seq = self.parse_seq(end_check=lambda l: l.startswith('WHEN') or l == 'END-EVALUATE')
+                    if node.subjects:
+                        vals = [v.strip().strip("'").strip('"')
+                                for v in re.split(r'\s+ALSO\s+', raw_val)]
+                        node.when_list.append((vals, case_seq))
+                    else:
+                        node.when_list.append((raw_val, case_seq))
+                continue
+            self.advance()
+        return node
+
+    def _parse_perform(self):
+        line = self.clean()
+
+        m = re.match(r'^PERFORM\s+UNTIL\s+(.+?)\s*$', line)
+        if m:
+            node = BrPerform('until', condition=m.group(1).strip())
+            self.advance()
+            node.body_seq = self.parse_seq(end_check=lambda l: l == 'END-PERFORM')
+            if self.clean() == 'END-PERFORM':
+                self.advance()
+            return node
+
+        m = re.match(r'^PERFORM\s+(\w[\w-]*)\s+UNTIL\s+(.+?)\s*$', line)
+        if m:
+            target = m.group(1).strip()
+            node = BrPerform('para_until', target=target, condition=m.group(2).strip())
+            self.advance()
+            self._inline_perform(node, target)
+            return node
+
+        m = re.match(r'^PERFORM\s+(\d+)\s+TIMES\s*$', line)
+        if m:
+            node = BrPerform('times', times=int(m.group(1)))
+            self.advance()
+            return node
+
+        m = re.match(r'^PERFORM\s+(\w[\w-]*)\s+THRU\s+(\w[\w-]*)\s*$', line)
+        if m:
+            node = BrPerform('thru', target=m.group(1).strip(), thru=m.group(2).strip())
+            self.advance()
+            self._inline_perform(node, node.target, node.thru)
+            return node
+
+        m = re.match(r'^PERFORM\s+VARYING\s+(\w[\w-]*)\s+FROM\s+(\S+)\s+BY\s+(\S+)(?:\s+UNTIL\s+(.+))?\s*$', line)
+        if m:
+            varying_var = m.group(1).strip()
+            from_val = m.group(2).strip()
+            by_val = m.group(3).strip()
+            condition = m.group(4).strip() if m.group(4) else None
+            if not condition:
+                save_pos = self.pos
+                self.advance()
+                while self.pos < len(self.lines):
+                    nxt = self.clean()
+                    cm = re.match(r'^UNTIL\s+(.+)$', nxt)
+                    if cm:
+                        condition = cm.group(1).strip()
+                        self.advance()
+                        break
+                    fm = re.match(r'^FROM\s+(\S+)\s+BY\s+(\S+)$', nxt)
+                    if fm:
+                        from_val = fm.group(1).strip()
+                        by_val = fm.group(2).strip()
+                        self.advance()
+                        continue
+                    self.pos = save_pos
+                    break
+            if condition:
+                node = BrPerform('varying', condition=condition,
+                                 varying_var=varying_var,
+                                 varying_from=from_val,
+                                 varying_by=by_val)
+                # condition from regex (single-line) → advance past PERFORM line
+                # condition from while-loop (multi-line) → already advanced past FROM/BY/UNTIL
+                if m.group(4):
+                    self.advance()
+                node.body_seq = self.parse_seq(end_check=lambda l: l == 'END-PERFORM')
+                if self.clean() == 'END-PERFORM':
+                    self.advance()
+                return node
+            self.pos = save_pos
+        # PERFORM VARYING var — FROM/BY/UNTIL all on subsequent lines
+        m = re.match(r'^PERFORM\s+VARYING\s+(\w[\w-]*)\s*$', line)
+        if m:
+            varying_var = m.group(1).strip()
+            save_pos = self.pos
+            self.advance()
+            from_val = by_val = condition = None
+            while self.pos < len(self.lines):
+                nxt = self.clean()
+                fm = re.match(r'^FROM\s+(\S+)\s+BY\s+(\S+)$', nxt)
+                if fm:
+                    from_val, by_val = fm.group(1).strip(), fm.group(2).strip()
+                    self.advance()
+                    continue
+                um = re.match(r'^UNTIL\s+(.+)$', nxt)
+                if um:
+                    condition = um.group(1).strip()
+                    self.advance()
+                    break
+                break
+            if from_val and by_val and condition:
+                node = BrPerform('varying', condition=condition,
+                                 varying_var=varying_var,
+                                 varying_from=from_val,
+                                 varying_by=by_val)
+                node.body_seq = self.parse_seq(end_check=lambda l: l == 'END-PERFORM')
+                if self.clean() == 'END-PERFORM':
+                    self.advance()
+                return node
+            self.pos = save_pos
+
+        m = re.match(r'^PERFORM\s+(\w[\w-]*)\s+VARYING\s+(\w[\w-]*)\s+FROM\s+(\S+)\s+BY\s+(\S+)(?:\s+UNTIL\s+(.+))?\s*$', line)
+        if m:
+            target = m.group(1).strip()
+            varying_var = m.group(2).strip()
+            from_val = m.group(3).strip()
+            by_val = m.group(4).strip()
+            condition = m.group(5).strip() if m.group(5) else None
+            if not condition:
+                save_pos = self.pos
+                self.advance()
+                while self.pos < len(self.lines):
+                    nxt = self.clean()
+                    cm = re.match(r'^UNTIL\s+(.+)$', nxt)
+                    if cm:
+                        condition = cm.group(1).strip()
+                        self.advance()
+                        break
+                    self.pos = save_pos
+                    break
+            if condition:
+                node = BrPerform('para_varying', target=target,
+                                 condition=condition,
+                                 varying_var=varying_var,
+                                 varying_from=from_val,
+                                 varying_by=by_val)
+                self.advance()
+                self._inline_perform(node, node.target)
+                return node
+            self.pos = save_pos
+
+        m = re.match(r'^PERFORM\s+(\w[\w-]*)\s*$', line)
+        if m:
+            target = m.group(1).strip()
+            node = BrPerform('para', target=target)
+            self.advance()
+            self._inline_perform(node, target)
+            return node
+
+        self.advance()
+        return None
+
+    def _inline_perform(self, node, target, thru=None):
+        if thru:
+            if target in self.paragraphs and thru in self.paragraphs:
+                start = self.paragraphs[target][0]
+                end = self.paragraphs[thru][1]
+                all_lines = []
+                for name, (s, e) in self.paragraphs.items():
+                    if s >= start and e <= end:
+                        all_lines.extend(self.raw_lines[s:e + 1])
+                sub = _BrParser(
+                    [l for l in all_lines if l.strip()],
+                    self.paragraphs, self.raw_lines, self.assignments, self.fields
+                )
+                node.body_seq = sub.parse_seq()
+        elif target in self.paragraphs:
+            start, end = self.paragraphs[target]
+            para_lines = self.raw_lines[start:end + 1]
+            sub = _BrParser(
+                [l for l in para_lines if l.strip()],
+                self.paragraphs, self.raw_lines, self.assignments, self.fields
+            )
+            node.body_seq = sub.parse_seq()
+
+    def _parse_initialize(self):
+        line = self.clean()
+        m = re.match(r'^INITIALIZE\s+(.+?)\s*$', line)
+        if not m:
+            self.advance()
+            return None
+        rest = m.group(1).strip()
+
+        # Split off REPLACING clause
+        parts = re.split(r'\s+REPLACING\s+', rest, maxsplit=1, flags=re.IGNORECASE)
+        target_str = parts[0].strip()
+        targets = re.findall(r'[A-Z][A-Z0-9-]*', target_str)
+
+        # Parse REPLACING: (NUMERIC|ALPHANUMERIC|ALPHABETIC) DATA BY literal
+        replacing = {}
+        if len(parts) > 1:
+            pairs = re.findall(
+                r'(NUMERIC|ALPHANUMERIC-EDITED|NUMERIC-EDITED|ALPHANUMERIC|ALPHABETIC)\s+DATA\s+BY\s+(\S+)',
+                parts[1], re.IGNORECASE
+            )
+            for ptype, literal in pairs:
+                replacing[ptype.upper()] = literal.strip("'").strip('"')
+
+        seq = BrSeq()
+        for tgt in targets:
+            info = {'type': 'initialize'}
+            if replacing:
+                info['replacing'] = replacing
+            self.assignments.setdefault(tgt, []).append(info)
+            seq.add(Assign(tgt, info))
+        self.advance()
+        return seq
+
+    def _parse_string(self):
+        parts = [self.clean()]
+        self.advance()
+        while self.pos < len(self.lines):
+            cl = self.clean()
+            if cl == 'END-STRING':
+                self.advance()
+                break
+            parts.append(cl)
+            self.advance()
+        full = ' '.join(parts)
+        m = re.match(r'^STRING\s+(.+)\s+INTO\s+(\w[\w-]*)\s*$', full, re.IGNORECASE | re.DOTALL)
+        if not m:
+            return None
+        source_part = m.group(1).strip()
+        target = m.group(2).strip()
+        source_vars = re.findall(r'[A-Z][A-Z0-9-]*', source_part)
+        info = {'type': 'string_concat', 'source_vars': source_vars}
+        self.assignments.setdefault(target, []).append(info)
+        seq = BrSeq()
+        seq.add(Assign(target, info))
+        return seq
+
+    def _parse_unstring(self):
+        parts = [self.clean()]
+        self.advance()
+        while self.pos < len(self.lines):
+            cl = self.clean()
+            if cl == 'END-UNSTRING':
+                self.advance()
+                break
+            parts.append(cl)
+            self.advance()
+        full = ' '.join(parts)
+        m = re.match(r'^UNSTRING\s+(.+?)\s+INTO\s+(.+?)\s*$', full, re.IGNORECASE | re.DOTALL)
+        if not m:
+            return None
+        source_part = m.group(1).strip()
+        targets_part = m.group(2).strip()
+        source_vars = re.findall(r'[A-Z][A-Z0-9-]*', source_part)
+        targets = re.findall(r'[A-Z][A-Z0-9-]*', targets_part)
+        source_var = source_vars[0] if source_vars else ''
+        seq = BrSeq()
+        for tgt in targets:
+            info = {'type': 'unstring_split', 'source_vars': [source_var], 'index': targets.index(tgt)}
+            self.assignments.setdefault(tgt, []).append(info)
+            seq.add(Assign(tgt, info))
+        return seq
+
+    def _parse_call(self):
+        line = self.clean()
+        m = re.match(r'^CALL\s+(\S+?)(?:\s+USING\s+(.+))?\s*$', line)
+        if not m:
+            self.advance()
+            return BrSeq()
+        prog = m.group(1).strip("'\"").upper()
+        params = []
+        if m.group(2):
+            rest = m.group(2)
+            # 逐 segment 解析: BY mechanism names...
+            current = "reference"  # COBOL 默认 BY REFERENCE
+            for seg in re.split(r'\s+(?=BY\s+(?:REFERENCE|CONTENT|VALUE)\s+)',
+                                rest, flags=re.IGNORECASE):
+                seg = seg.strip()
+                m_mech = re.match(
+                    r'BY\s+(REFERENCE|CONTENT|VALUE)\s+(.*)', seg, re.IGNORECASE
+                )
+                if m_mech:
+                    current = m_mech.group(1).lower()
+                    names_text = m_mech.group(2)
+                else:
+                    names_text = seg
+                for nm in re.findall(r'\w[\w-]*', names_text):
+                    params.append({"name": nm.upper(), "mechanism": current})
+        node = CallNode(prog, using_params=params)
+        self.advance()
+        return node
+
+    def _parse_goto(self, target):
+        node = GoTo(target)
+        if self._goto_depth < 10 and target in self.paragraphs:
+            start, end = self.paragraphs[target]
+            para_lines = self.raw_lines[start:end + 1]
+            sub = _BrParser(
+                [l for l in para_lines if l.strip()],
+                self.paragraphs, self.raw_lines, self.assignments, self.fields,
+                goto_depth=self._goto_depth + 1
+            )
+            node.body_seq = sub.parse_seq()
+        self.advance()
+        return node
+
+    def _parse_set_true(self, name):
+        name = name.upper()
+        parent = None
+        value = None
+        if self.fields:
+            for f in self.fields:
+                if f.get('is_88') and f['name'] == name:
+                    parent = f.get('parent', '')
+                    value = f.get('value', '')
+                    break
+        info = {'type': 'set_true', '88_name': name, 'value': value}
+        tgt = parent or name
+        if parent:
+            self.assignments.setdefault(tgt, []).append(info)
+        self.advance()
+        return Assign(tgt, info)
+
+    def _parse_set_false(self, name):
+        name = name.upper()
+        parent = None
+        value = None
+        if self.fields:
+            for f in self.fields:
+                if f.get('is_88') and f['name'] == name:
+                    parent = f.get('parent', '')
+                    value = f.get('value', '')
+                    break
+        # FALSE 值 = 88-level VALUE 的反值
+        if value:
+            false_val = 'N' if value == 'Y' else ('Y' if value == 'N' else ' ')
+        else:
+            false_val = 'N'
+        info = {'type': 'move_literal', 'literal': false_val}
+        tgt = parent or name
+        self.assignments.setdefault(tgt, []).append(info)
+        self.advance()
+        return Assign(tgt, info)
+
+
+# ── 工具函数 ──
+
+
+def _basename(name: str) -> str:
+    """去除下标后缀，如 WS-TABLE(1) → WS-TABLE"""
+    return re.sub(r'\s*\(.*?\)\s*$', '', name).strip()
+
+
+def _init_child_names(group_name: str, fields: list) -> list:
+    """递归收集 group 下所有非 88 级子字段的扁平名列表"""
+    result = []
+    grp_level = None
+    found = False
+    for f in fields:
+        if not found and f['name'] == group_name:
+            grp_level = f.get('level', 0)
+            found = True
+            continue
+        if found:
+            if f.get('level', 0) <= grp_level or f.get('level') == 77:
+                break
+            if f.get('is_88') or f.get('redefines'):
+                continue
+            if not f.get('pic_info') or f['pic_info'].get('type') == 'unknown':
+                result.extend(_init_child_names(f['name'], fields))
+            else:
+                result.append(f['name'])
+    return result
+
+
+# ── 数据流追踪 ──
+
+def trace_to_root(field_name, assignments, fields, path_assign=None):
+    seen = set()
+    var = field_name
+    chain = []
+    while var in assignments and var not in seen:
+        seen.add(var)
+        if path_assign and var in path_assign:
+            asgn_list = path_assign[var]
+            if isinstance(asgn_list, list):
+                asgn = asgn_list[-1]
+                for a in reversed(asgn_list):
+                    sv = a.get('source_vars', [])
+                    if len(sv) == 1 and sv[0] == var:
+                        continue
+                    asgn = a
+                    break
+            else:
+                asgn = asgn_list
+        else:
+            asgn_list = assignments[var]
+            asgn = asgn_list[-1]
+            if isinstance(asgn_list, list):
+                for a in reversed(asgn_list):
+                    sv = a.get('source_vars', [])
+                    if len(sv) == 1 and sv[0] == var:
+                        continue
+                    asgn = a
+                    break
+        chain.append((var, asgn))
+        if not asgn.get('source_vars'):
+            break
+        sv = asgn['source_vars']
+        if len(sv) == 1:
+            next_var = sv[0]
+            if next_var == var:
+                break
+            var = next_var
+            if next_var not in assignments:
+                break
+        elif len(sv) >= 2 and asgn.get('op') == '+':
+            # 多源加法：取第一个源变量继续追溯
+            var = sv[0]
+        else:
+            break
+    return var, chain
+
+
+def invert_through_chain(root_var, chain, operator, value):
+    op = operator
+    try:
+        val = float(value)
+    except (ValueError, TypeError):
+        return root_var, op, value
+    for var, asgn in reversed(chain):
+        if asgn['type'] == 'move':
+            continue
+        sv = asgn.get('source_vars', [])
+        if asgn['type'] == 'compute' and asgn['op'] is not None:
+            if len(sv) == 1:
+                c = asgn['const']
+                inv = {'+': '-', '-': '+', '*': '/', '/': '*'}[asgn['op']]
+                if inv == '/':
+                    val = val / c if c != 0 else val
+                elif inv == '*':
+                    val = val * c
+                elif inv == '-':
+                    val = val - c
+                elif inv == '+':
+                    val = val + c
+            elif len(sv) >= 2 and asgn['op'] == '+':
+                # 多源加法：追溯第一个源变量，值不变（忽略其他源）
+                pass
+    if val == int(val):
+        return root_var, op, str(int(val))
+    return root_var, op, str(val)
+
+
+FIGURATIVE_NUMERIC = {
+    'ZERO': 0.0, 'ZEROS': 0.0, 'ZEROES': 0.0,
+    'SPACE': 0.0, 'SPACES': 0.0,
+    'HIGH-VALUE': None, 'HIGH-VALUES': None,
+    'LOW-VALUE': 0.0, 'LOW-VALUES': 0.0,
+}
+FIGURATIVE_ALPHA = {
+    'SPACE': ' ', 'SPACES': ' ',
+    'HIGH-VALUE': chr(255), 'HIGH-VALUES': chr(255),
+    'LOW-VALUE': chr(0), 'LOW-VALUES': chr(0),
+}
+
+
+def _resolve_subscript(key, rec):
+    """将变量下标解析为具体值：WS-FIXED-KEY(WS-IDX) → WS-FIXED-KEY(1) if WS-IDX=1 in rec"""
+    m = re.match(r'^(\w[\w-]*)\((\w[\w-]*)\)$', key)
+    if m:
+        base, var = m.groups()
+        if var in rec:
+            try:
+                return f'{base}({int(rec[var])})'
+            except (ValueError, TypeError):
+                pass
+    return key
+
+
+def _apply_before_after(val, before_after, delimiter):
+    if not delimiter:
+        return val
+    if before_after == 'BEFORE':
+        idx = val.find(delimiter)
+        return val[:idx] if idx >= 0 else val
+    if before_after == 'AFTER':
+        idx = val.find(delimiter)
+        return val[idx + len(delimiter):] if idx >= 0 else ''
+    return val
+
+
+def propagate_assignments(rec, assignments, fields, file_sec=None):
+    def raw_to_float(val, pi):
+        if pi.get('type') == 'numeric':
+            digits = pi.get('digits', 0)
+            decimal = pi.get('decimal', 0)
+            total = digits + decimal
+            s = str(val)
+            neg = s.startswith('-')
+            if neg:
+                s = s[1:]
+            s = s.zfill(total)
+            int_part = s[:digits] if digits else '0'
+            dec_part = s[digits:] if decimal > 0 else '0'
+            result = float(int(int_part or '0') + int(dec_part or '0') / (10 ** decimal))
+            return -result if neg else result
+        try:
+            return float(val)
+        except (ValueError, TypeError):
+            return 0.0
+
+    def float_to_raw(val, pi):
+        if pi.get('type') == 'numeric':
+            digits = pi.get('digits', 0)
+            decimal = pi.get('decimal', 0)
+            signed = pi.get('signed', False)
+            scaled = int(round(val * (10 ** decimal)))
+            if not signed and scaled < 0:
+                scaled = 0
+            capped = abs(scaled) % (10 ** (digits + decimal))
+            int_part = str(capped // (10 ** decimal)).zfill(digits)
+            dec_part = str(capped % (10 ** decimal)).zfill(decimal)
+            result = int_part + (dec_part if decimal > 0 else '')
+            if signed and scaled < 0:
+                result = '-' + result
+            return result
+        return str(val)
+
+    def literal_to_raw(literal, pi):
+        ftype = pi.get('type', 'unknown')
+        if ftype == 'numeric':
+            key = literal.upper()
+            if key in FIGURATIVE_NUMERIC:
+                v = FIGURATIVE_NUMERIC[key]
+                if v is None:
+                    digits = pi.get('digits', 0)
+                    decimal = pi.get('decimal', 0)
+                    v = 10 ** (digits + decimal) - 1
+                return float_to_raw(v, pi)
+            try:
+                return float_to_raw(float(literal), pi)
+            except ValueError:
+                return float_to_raw(0.0, pi)
+        if ftype in ('alphanumeric', 'alphabetic'):
+            key = literal.upper()
+            if key in FIGURATIVE_ALPHA:
+                ch = FIGURATIVE_ALPHA[key]
+                return ch[0].ljust(pi.get('length', 1), ch[0])
+            return literal.ljust(pi.get('length', len(literal)))[:pi.get('length', len(literal))]
+        return literal
+
+    pi_map = {f['name']: f.get('pic_info', {}) for f in fields}
+    if file_sec is None:
+        file_sec = {}
+
+    # Flatten: {tgt: [info1, info2]} → [(tgt, info1), (tgt, info2)]
+    flat_list = []
+    for tgt, asgn_val in assignments.items():
+        if isinstance(asgn_val, list):
+            for asgn in asgn_val:
+                flat_list.append((tgt, asgn))
+        elif isinstance(asgn_val, dict):
+            flat_list.append((tgt, asgn_val))
+
+    _MAX_CONVERGE = 20
+
+    # 识别有"锚定赋值"(非自引用赋值，如 MOVE  literal 或不同字段的 MOVE) 的 target
+    _anchored = set()
+    for tgt, asgn in flat_list:
+        if asgn.get('type') != 'compute':
+            _anchored.add(tgt)
+        else:
+            sv = asgn.get('source_vars', [])
+            if not (len(sv) == 1 and sv[0] == tgt) and not (len(sv) >= 2 and tgt == sv[0]):
+                _anchored.add(tgt)
+
+    for _converge_iter in range(_MAX_CONVERGE):
+        _old = dict(rec)
+
+        # Pass 1: variable-to-variable MOVE
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'move' and asgn['source_vars']:
+                src = asgn['source_vars'][0]
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                resolved_src = _resolve_subscript(src, rec)
+                if resolved_src in rec:
+                    rec[resolved_tgt] = rec[resolved_src]
+
+        # Pass 2: literal MOVE
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'move_literal':
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                pi = pi_map.get(resolved_tgt, {})
+                rec[resolved_tgt] = literal_to_raw(asgn['literal'], pi)
+
+        # Pass 3: INITIALIZE
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'initialize':
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                pi = pi_map.get(resolved_tgt, {})
+                ftype = pi.get('type', 'unknown')
+                replacing = asgn.get('replacing', {})
+                if replacing:
+                    mapped = replacing.get(ftype.upper(), None)
+                    if mapped:
+                        rec[resolved_tgt] = literal_to_raw(mapped, pi)
+                    else:
+                        if ftype == 'numeric':
+                            rec[resolved_tgt] = float_to_raw(0.0, pi)
+                        else:
+                            rec[resolved_tgt] = literal_to_raw('SPACE', pi)
+                else:
+                    if ftype == 'numeric':
+                        rec[resolved_tgt] = float_to_raw(0.0, pi)
+                    else:
+                        rec[resolved_tgt] = literal_to_raw('SPACE', pi)
+
+        # Pass 3.5: READ INTO
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'read_into':
+                fname = asgn.get('file', '')
+                if fname in file_sec:
+                    fd_children = _init_child_names(file_sec[fname][0], fields)
+                    ws_children = _init_child_names(tgt, fields)
+                    for ws_c in ws_children:
+                        fd_candidate = ws_c
+                        if ws_c.startswith('WS-'):
+                            fd_candidate = ws_c[3:]
+                        if fd_candidate in rec:
+                            rec[ws_c] = rec[fd_candidate]
+                        else:
+                            idx = ws_children.index(ws_c)
+                            if idx < len(fd_children) and fd_children[idx] in rec:
+                                rec[ws_c] = rec[fd_children[idx]]
+                    rec[tgt] = ''.join(str(rec.get(c, '')) for c in ws_children)
+
+        # Pass 4: COMPUTE
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'compute' and asgn['source_vars'] and asgn['op'] is not None:
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                pi_tgt = pi_map.get(resolved_tgt, {})
+                if len(asgn['source_vars']) == 1:
+                    src = asgn['source_vars'][0]
+                    resolved_src = _resolve_subscript(src, rec)
+                    # 无锚定的自引用 COMPUTE（如 ADD 1 TO X）：只在第 0 轮应用一次
+                    if resolved_tgt == resolved_src and tgt not in _anchored and _converge_iter > 0:
+                        continue
+                    if resolved_src in rec:
+                        sv = raw_to_float(rec[resolved_src], pi_map.get(resolved_src, {}))
+                        c = asgn.get('const', 0)
+                        if asgn['op'] == 'rem':
+                            quotient = int(sv / c) if c != 0 else 0
+                            result = sv - quotient * c
+                        else:
+                            result = {'+': sv + c, '-': sv - c, '*': sv * c, '/': sv / c if c != 0 else sv}[asgn['op']]
+                        rec[resolved_tgt] = float_to_raw(result, pi_tgt)
+                elif len(asgn['source_vars']) == 2:
+                    v1, v2 = asgn['source_vars']
+                    resolved_v1 = _resolve_subscript(v1, rec)
+                    resolved_v2 = _resolve_subscript(v2, rec)
+                    # 无锚定的自引用 COMPUTE（如 ADD X TO Y 且 Y 无前置 MOVE）
+                    if resolved_tgt == resolved_v1 and tgt not in _anchored and _converge_iter > 0:
+                        continue
+                    if resolved_v1 in rec and resolved_v2 in rec:
+                        sv1 = raw_to_float(rec[resolved_v1], pi_map.get(resolved_v1, {}))
+                        sv2 = raw_to_float(rec[resolved_v2], pi_map.get(resolved_v2, {}))
+                        if asgn['op'] == 'rem':
+                            quotient = int(sv1 / sv2) if sv2 != 0 else 0
+                            result = sv1 - quotient * sv2
+                        else:
+                            result = {'+': sv1 + sv2, '-': sv1 - sv2, '*': sv1 * sv2, '/': sv1 / sv2 if sv2 != 0 else sv1}[asgn['op']]
+                        rec[resolved_tgt] = float_to_raw(result, pi_tgt)
+                elif len(asgn['source_vars']) >= 3 and asgn['op'] == '+':
+                    total = 0
+                    all_found = True
+                    for v in asgn['source_vars']:
+                        resolved_v = _resolve_subscript(v, rec)
+                        if resolved_v in rec:
+                            total += raw_to_float(rec[resolved_v], pi_map.get(resolved_v, {}))
+                        else:
+                            all_found = False
+                            break
+                    if all_found:
+                        rec[resolved_tgt] = float_to_raw(total, pi_tgt)
+
+        # Pass 4.5: INSPECT
+        for tgt, asgn in flat_list:
+            if asgn['type'] != 'inspect':
+                continue
+            resolved_tgt = _resolve_subscript(tgt, rec)
+            if resolved_tgt not in rec:
+                continue
+            src_val = str(rec[resolved_tgt])
+            for op_type, params in asgn.get('sub_ops', []):
+                if op_type == 'tally':
+                    cv = params['count_var'].upper()
+                    cv_pi = pi_map.get(cv, {})
+                    effective = _apply_before_after(src_val, params.get('before_after'), params.get('delimiter'))
+                    cnt = 0
+                    if params['kind'] == 'LEADING':
+                        cnt = len(effective) - len(effective.lstrip(params['char']))
+                    elif params['kind'] == 'TRAILING':
+                        cnt = len(effective) - len(effective.rstrip(params['char']))
+                    else:
+                        cnt = len(effective)
+                    if cv_pi.get('type') == 'numeric':
+                        rec[cv] = float_to_raw(float(cnt), cv_pi)
+                elif op_type == 'replace':
+                    effective = _apply_before_after(src_val, params.get('before_after'), params.get('delimiter'))
+                    if params['kind'] == 'ALL':
+                        new_val = effective.replace(params['src'], params['dst'])
+                    elif params['kind'] == 'LEADING':
+                        new_val = effective
+                        while new_val.startswith(params['src']):
+                            new_val = new_val[len(params['src']):]
+                        new_val = effective.replace(params['src'], params['dst'], 1)
+                    elif params['kind'] == 'FIRST':
+                        new_val = effective.replace(params['src'], params['dst'], 1)
+                    else:
+                        new_val = params['dst'] * len(effective)
+                    rec[resolved_tgt] = new_val
+                elif op_type == 'convert':
+                    effective = _apply_before_after(src_val, params.get('before_after'), params.get('delimiter'))
+                    table = str.maketrans(params['from_chars'], params['to_chars'])
+                    rec[resolved_tgt] = effective.translate(table)
+
+        # Pass 5: STRING / UNSTRING
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'string_concat':
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                pi = pi_map.get(resolved_tgt, {})
+                parts = []
+                for v in asgn.get('source_vars', []):
+                    resolved_v = _resolve_subscript(v, rec)
+                    if resolved_v in rec:
+                        parts.append(str(rec[resolved_v]))
+                val = ''.join(parts)
+                if pi.get('type') in ('alphanumeric', 'alphabetic'):
+                    val = val.ljust(pi.get('length', len(val)))[:pi.get('length', len(val))]
+                rec[resolved_tgt] = val
+            elif asgn['type'] == 'unstring_split':
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                pi = pi_map.get(resolved_tgt, {})
+                src_var = asgn.get('source_vars', [None])[0]
+                resolved_src = _resolve_subscript(src_var, rec) if src_var else None
+                idx = asgn.get('index', 0)
+                if resolved_src and resolved_src in rec:
+                    src_val = str(rec[resolved_src])
+                    ftype = pi.get('type', 'unknown')
+                    if idx == 0:
+                        val = src_val
+                    else:
+                        val = ' ' if ftype in ('alphanumeric', 'alphabetic') else '0'
+                    if ftype in ('alphanumeric', 'alphabetic'):
+                        val = val.ljust(pi.get('length', len(val)))[:pi.get('length', len(val))]
+                    rec[resolved_tgt] = val
+
+        # Pass 6: READ INTO / WRITE FROM
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'read_into':
+                fname = asgn.get('file', '')
+                if fname in file_sec:
+                    children = _init_child_names(file_sec[fname][0], fields)
+                    rec[tgt] = ''.join(str(rec.get(c, '')) for c in children)
+            elif asgn['type'] == 'write_from':
+                buf = tgt
+                rec_name = asgn.get('file', '')
+                children = _init_child_names(rec_name, fields)
+                if children:
+                    src = str(rec.get(buf, ''))
+                    pos = 0
+                    for c in children:
+                        pi = pi_map.get(c, {})
+                        length = pi.get('digits', 0) + pi.get('decimal', 0) or pi.get('length', 0)
+                        if length > 0:
+                            chunk = src[pos:pos + length]
+                            if not chunk:
+                                chunk = '0' if pi.get('type') == 'numeric' else ' '
+                            rec[c] = chunk.ljust(length)
+                            pos += length
+
+        # Pass 7: ACCEPT
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'accept':
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                pi = pi_map.get(resolved_tgt, {})
+                ftype = pi.get('type', 'unknown')
+                total = pi.get('digits', 0) + pi.get('decimal', 0)
+                length = pi.get('length', 0)
+                from_type = asgn.get('from', 'USER')
+                val = None
+                if from_type == 'DATE':
+                    val = '20260603'
+                elif from_type == 'TIME':
+                    val = '120000'
+                elif from_type == 'DAY':
+                    val = '2026154'
+                elif from_type == 'DAY-OF-WEEK':
+                    val = '3'
+                elif from_type == 'YEAR':
+                    val = '2026'
+                if val is not None:
+                    if ftype == 'numeric':
+                        rec[resolved_tgt] = val.zfill(total)
+                    else:
+                        rec[resolved_tgt] = val.ljust(length)[:length] if length else val
+
+        # Pass 8: SET var TO TRUE (88-level)
+        for tgt, asgn in flat_list:
+            if asgn['type'] == 'set_true':
+                resolved_tgt = _resolve_subscript(tgt, rec)
+                val = asgn.get('value', '1')
+                pi = pi_map.get(resolved_tgt, {})
+                ftype = pi.get('type', 'unknown')
+                if ftype in ('alphanumeric', 'alphabetic'):
+                    length = pi.get('length', len(str(val)))
+                    rec[resolved_tgt] = str(val)[0].ljust(length)[:length]
+                else:
+                    total = pi.get('digits', 0) + pi.get('decimal', 0)
+                    rec[resolved_tgt] = str(val).zfill(max(total, 1))
+
+        if rec == _old:
+            break
+    else:
+        logger.warning(f"propagate_assignments 未收敛（{_MAX_CONVERGE} 次迭代后仍有变化）")
+
+
+def classify_field_roles(tree, assignments, fields, source=None, proc_text=None):
+    """分析分支树和赋值记录，分类各字段的入出力角色。
+    优先级：FD/OPEN 方向 > 静态分析
+    返回 {字段名: 'input'|'output'|'inout'|'unused'}.
+    """
+    # Phase 0: FD/OPEN 方向解析
+    fd_roles = {}
+    if source and proc_text:
+        from .read import parse_file_control, parse_file_section, scan_open_statements
+        file_ctl = parse_file_control(source)
+        file_sec = parse_file_section(source)
+        open_dir = scan_open_statements(proc_text)
+        for iname, direction in open_dir.items():
+            if iname in file_sec:
+                for rec_name in file_sec[iname]:
+                    if direction == 'INPUT':
+                        fd_roles[rec_name] = 'input'
+                    elif direction == 'OUTPUT':
+                        fd_roles[rec_name] = 'output'
+                    elif direction == 'I-O':
+                        fd_roles[rec_name] = 'inout'
+        # 传播到子字段
+        for rec_name, role in list(fd_roles.items()):
+            for child in _init_child_names(rec_name, fields):
+                fd_roles[child] = role
+
+    counts = {f['name']: {'read': 0, 'write': 0} for f in fields}
+
+    def _walk(node):
+        if isinstance(node, BrIf):
+            if node.cond_tree:
+                for leaf in collect_leaves(node.cond_tree):
+                    name = _basename(leaf.field)
+                    if name in counts:
+                        counts[name]['read'] += 1
+            _walk(node.true_seq)
+            _walk(node.false_seq)
+        elif isinstance(node, BrEval):
+            name = _basename(node.subject)
+            if name in counts:
+                counts[name]['read'] += 1
+            for _, seq in node.when_list:
+                _walk(seq)
+            _walk(node.other_seq)
+        elif isinstance(node, BrPerform):
+            if node.condition:
+                parsed = parse_single_condition(node.condition)
+                if parsed:
+                    name = _basename(parsed[0])
+                    if name in counts:
+                        counts[name]['read'] += 1
+            if node.varying_var:
+                name = _basename(node.varying_var)
+                if name in counts:
+                    counts[name]['write'] += 1
+            _walk(node.body_seq)
+        elif isinstance(node, CallNode):
+            for p in node.using_params:
+                name = _basename(p.get("name", ""))
+                mechanism = p.get("mechanism", "reference")
+                if name in counts:
+                    counts[name]["read"] += 1
+                    if mechanism.lower() == "reference":
+                        counts[name]["write"] += 1
+        elif isinstance(node, Assign):
+            tgt_base = _basename(node.target)
+            atype = node.source_info.get('type')
+            if atype == 'read_into':
+                if tgt_base in counts:
+                    counts[tgt_base]['write'] += 1
+            elif atype == 'write_from':
+                if tgt_base in counts:
+                    counts[tgt_base]['read'] += 1
+            elif atype == 'set_true':
+                if tgt_base in counts:
+                    counts[tgt_base]['write'] += 1
+            else:
+                if tgt_base in counts:
+                    counts[tgt_base]['write'] += 1
+                for v in node.source_info.get('source_vars', []):
+                    v_base = _basename(v)
+                    if v_base in counts:
+                        counts[v_base]['read'] += 1
+                if atype == 'initialize' and tgt_base in counts:
+                    for child in _init_child_names(tgt_base, fields):
+                        if child in counts:
+                            counts[child]['write'] += 1
+        elif isinstance(node, BrSeq):
+            for c in node.children:
+                _walk(c)
+
+    _walk(tree)
+
+    # Phase extra: ACCEPT / DISPLAY (proc_text 扫描)
+    if proc_text:
+        for m in re.finditer(r'ACCEPT\s+(\w[\w-]*)', proc_text):
+            name = _basename(m.group(1).upper())
+            if name in counts:
+                counts[name]['write'] += 1
+        for m in re.finditer(r'DISPLAY\s+(\w[\w-]*)', proc_text):
+            name = _basename(m.group(1).upper())
+            if name in counts:
+                counts[name]['read'] += 1
+
+    # LINKAGE 字段默认 input（未使用时不改变）
+    for f in fields:
+        if f.get('section') == 'LINKAGE':
+            name = f['name']
+            if name in counts and counts[name]['read'] == 0 and counts[name]['write'] == 0:
+                counts[name]['read'] = 1
+
+    result = {}
+    for name, c in counts.items():
+        if name in fd_roles:
+            result[name] = fd_roles[name]
+            continue
+        if c['read'] > 0 and c['write'] > 0:
+            result[name] = 'inout'
+        elif c['write'] > 0:
+            result[name] = 'output'
+        elif c['read'] > 0:
+            result[name] = 'input'
+        else:
+            result[name] = 'unused'
+    # 确保 FD 记录字段也出现（即使不在 fields 中—应不会）
+    for name, role in fd_roles.items():
+        if name not in result:
+            result[name] = role
+    return result

cobol_testgen/design.py

@@ -0,0 +1,894 @@
+"""设计层：路径枚举 + 值生成 + 约束应用"""
+
+import re
+import logging
+from .models import BrSeq, BrIf, BrEval, BrPerform, BrSearch, Assign, CallNode, CondNot, CondLeaf, ExitNode, GoTo
+from .cond import parse_single_condition, parse_compound_condition, is_field, collect_leaves, mcdc_sets, satisfying_value
+from .core import trace_to_root, invert_through_chain, propagate_assignments, _basename
+
+logger = logging.getLogger(__name__)
+
+_STOP = ('__STOP__', '', None, True)
+_MAX_PATHS = 10000
+
+
+def _filter_stop(cons):
+    return [c for c in cons if c is not _STOP]
+
+
+def _cap_paths(paths):
+    if len(paths) > _MAX_PATHS:
+        return paths[:_MAX_PATHS]
+    return paths
+
+
+def _cap_paths_fair(new_active, child_paths):
+    """两阶段公平截断：每个前置路径至少保留一条子路径，再填充剩余配额。"""
+    if len(new_active) <= _MAX_PATHS:
+        return new_active
+    k = len(child_paths)
+    if k <= 1:
+        return new_active[:_MAX_PATHS]
+    # 分离 STOP 路径（不参与组合，直接保留）
+    stop_paths = [(p, a) for p, a in new_active if any(c is _STOP for c in p)]
+    combined = [(p, a) for p, a in new_active if not any(c is _STOP for c in p)]
+    n_pred = len(combined) // k
+    result = list(stop_paths)
+    if n_pred <= 1:
+        result.extend(combined[:_MAX_PATHS - len(result)])
+        return result[:_MAX_PATHS]
+    remaining_quota = _MAX_PATHS - len(result)
+    # Phase 1: 每个前置至少保留一条子路径（轮询分配不同子路径索引）
+    quota = min(n_pred, remaining_quota)
+    selected = set()
+    for p_idx in range(quota):
+        c_idx = p_idx % k
+        idx = p_idx * k + c_idx
+        selected.add(idx)
+        result.append(combined[idx])
+    if len(result) >= _MAX_PATHS:
+        return result[:_MAX_PATHS]
+    # Phase 2: 用剩余配额填充其余组合
+    remaining = _MAX_PATHS - len(result)
+    for idx in range(len(combined)):
+        if idx not in selected:
+            result.append(combined[idx])
+            remaining -= 1
+            if remaining <= 0:
+                break
+    return result[:_MAX_PATHS]
+
+
+# ── 路径枚举 ──
+
+def enum_paths(node, fields):
+    """枚举路径，每条路径返回 (constraints, assignments).
+    返回 list[tuple[list[tuple], dict]].
+    """
+    if isinstance(node, Assign):
+        return [([], {node.target: [node.source_info]})]
+
+    if isinstance(node, BrSeq):
+        if not node.children:
+            return [([], {})]
+        paths = [([], {})]
+        for child in node.children:
+            child_paths = _cap_paths(enum_paths(child, fields))
+            new_active = []
+            for p_cons, p_assign in paths:
+                if any(c is _STOP for c in p_cons):
+                    new_active.append((p_cons, p_assign))
+                    continue
+                for cp_cons, cp_assign in child_paths:
+                    merged = {}
+                    for d in (p_assign, cp_assign):
+                        for k, v in d.items():
+                            merged.setdefault(k, []).extend(v if isinstance(v, list) else [v])
+                    merged_cons = p_cons + list(cp_cons)
+                    new_active.append((merged_cons, merged))
+            paths = _cap_paths_fair(new_active, child_paths)
+        return paths
+
+    elif isinstance(node, BrIf):
+        parsed = parse_single_condition(node.condition, fields)
+        if parsed and is_field(parsed[0], fields):
+            field, op, val = parsed
+            paths = []
+            true_sub = _cap_paths(enum_paths(node.true_seq, fields))
+            for sp_cons, sp_assign in (true_sub or [([], {})]):
+                paths.append(([(field, op, val, True)] + sp_cons, sp_assign))
+            false_sub = _cap_paths(enum_paths(node.false_seq, fields))
+            for fp_cons, fp_assign in (false_sub or [([], {})]):
+                paths.append(([(field, op, val, False)] + fp_cons, fp_assign))
+            return paths
+        # CondNot wrapping a single leaf (e.g., IF NOT WS-AMOUNT > 1000)
+        if node.cond_tree and isinstance(node.cond_tree, CondNot):
+            child = node.cond_tree.child
+            if isinstance(child, CondLeaf) and is_field(child.field, fields):
+                paths = []
+                true_sub = _cap_paths(enum_paths(node.true_seq, fields))
+                for sp_cons, sp_assign in (true_sub or [([], {})]):
+                    paths.append(([(child.field, child.op, child.value, False)] + sp_cons, sp_assign))
+                false_sub = _cap_paths(enum_paths(node.false_seq, fields))
+                for fp_cons, fp_assign in (false_sub or [([], {})]):
+                    paths.append(([(child.field, child.op, child.value, True)] + fp_cons, fp_assign))
+                return paths
+        if node.cond_tree:
+            leaves = collect_leaves(node.cond_tree)
+            if leaves and all(is_field(l.field, fields) for l in leaves):
+                sets = mcdc_sets(node.cond_tree, fields)
+                if sets:
+                    paths = []
+                    for constraints, decision in sets:
+                        body = _cap_paths(enum_paths(
+                            node.true_seq if decision else node.false_seq, fields
+                        ))
+                        for sp_cons, sp_assign in (body or [([], {})]):
+                            paths.append((constraints + sp_cons, sp_assign))
+                    return paths
+            # CondLeaf fallback: 单 leaf（含 88-level 解析后的条件树）MC/DC 不适用
+            if len(leaves) == 1:
+                leaf = leaves[0]
+                paths = []
+                true_sub = _cap_paths(enum_paths(node.true_seq, fields))
+                for sp_cons, sp_assign in (true_sub or [([], {})]):
+                    paths.append(([(leaf.field, leaf.op, leaf.value, True)] + sp_cons, sp_assign))
+                false_sub = _cap_paths(enum_paths(node.false_seq, fields))
+                for fp_cons, fp_assign in (false_sub or [([], {})]):
+                    paths.append(([(leaf.field, leaf.op, leaf.value, False)] + fp_cons, fp_assign))
+                return paths
+        # Fallback: parsed condition but non-field (e.g. arithmetic expr)
+        if parsed:
+            field, op, val = parsed
+            paths = []
+            true_sub = enum_paths(node.true_seq, fields)
+            for sp_cons, sp_assign in (true_sub or [([], {})]):
+                paths.append(([(field, op, val, True)] + sp_cons, sp_assign))
+            false_sub = enum_paths(node.false_seq, fields)
+            for fp_cons, fp_assign in (false_sub or [([], {})]):
+                paths.append(([(field, op, val, False)] + fp_cons, fp_assign))
+            return paths
+        return [([], {})]
+
+    elif isinstance(node, BrEval):
+        if node.subjects:
+            paths = []
+            prior_false_cons = []
+            for values, seq in node.when_list:
+                sub = _cap_paths(enum_paths(seq, fields))
+                for sp_cons, sp_assign in (sub or [([], {})]):
+                    when_cons = [(node.subjects[i], '=', values[i], True)
+                                 for i in range(len(node.subjects))]
+                    constraints = list(prior_false_cons) + when_cons + sp_cons
+                    paths.append((constraints, sp_assign))
+                for i in range(len(node.subjects)):
+                    prior_false_cons.append((node.subjects[i], '=', values[i], False))
+            if node.has_other:
+                sub = _cap_paths(enum_paths(node.other_seq, fields))
+                for sp_cons, sp_assign in (sub or [([], {})]):
+                    paths.append((list(prior_false_cons) + sp_cons, sp_assign))
+            return paths
+        if node.subject == 'TRUE':
+            paths = []
+            prior_false_sets = []  # list[list[Constraint]]
+            for value, seq in node.when_list:
+                cond = parse_compound_condition(value, fields)
+                if cond and isinstance(cond, CondLeaf) and is_field(cond.field, fields):
+                    sub = _cap_paths(enum_paths(seq, fields))
+                    for sp_cons, sp_assign in (sub or [([], {})]):
+                        constraints = [c for pf in prior_false_sets for c in pf]
+                        constraints.append((cond.field, cond.op, cond.value, True))
+                        paths.append((constraints + sp_cons, sp_assign))
+                    prior_false_sets.append([(cond.field, cond.op, cond.value, False)])
+                elif cond:
+                    leaves = collect_leaves(cond)
+                    if leaves and all(is_field(l.field, fields) for l in leaves):
+                        sets = mcdc_sets(cond, fields)
+                        if sets:
+                            sub = _cap_paths(enum_paths(seq, fields))
+                            new_false_sets = []
+                            for cs, decision in sets:
+                                if decision:
+                                    if not prior_false_sets:
+                                        for sp_cons, sp_assign in (sub or [([], {})]):
+                                            paths.append((list(cs) + sp_cons, sp_assign))
+                                    else:
+                                        for pf_set in prior_false_sets:
+                                            for sp_cons, sp_assign in (sub or [([], {})]):
+                                                paths.append((list(pf_set) + list(cs) + sp_cons, sp_assign))
+                                else:
+                                    new_false_sets.append(cs)
+                            if not new_false_sets:
+                                prior_false_sets = []
+                                break
+                            combined = []
+                            for pf_set in prior_false_sets:
+                                for nf_set in new_false_sets:
+                                    combined.append(list(pf_set) + list(nf_set))
+                            prior_false_sets = combined
+                        else:
+                            prior_false_sets = []
+                            break
+                    else:
+                        prior_false_sets = []
+                        break
+                else:
+                    prior_false_sets = []
+                    break
+            if node.has_other:
+                sub = _cap_paths(enum_paths(node.other_seq, fields))
+                for sp_cons, sp_assign in (sub or [([], {})]):
+                    constraints = [c for pf in prior_false_sets for c in pf]
+                    paths.append((constraints + sp_cons, sp_assign))
+            return paths
+        if not is_field(node.subject, fields):
+            return [([], {})]
+        paths = []
+        for value, seq in node.when_list:
+            sub = _cap_paths(enum_paths(seq, fields))
+            for sp_cons, sp_assign in (sub or [([], {})]):
+                paths.append(([(node.subject, '=', value, True)] + sp_cons, sp_assign))
+        if node.has_other:
+            case_vals = [v for v, _ in node.when_list]
+            sub = _cap_paths(enum_paths(node.other_seq, fields))
+            for sp_cons, sp_assign in (sub or [([], {})]):
+                paths.append(([(node.subject, 'not_in', case_vals, True)] + sp_cons, sp_assign))
+        return paths
+
+    elif isinstance(node, BrSearch):
+        return _enum_search_paths(node, fields)
+
+    elif isinstance(node, BrPerform):
+        if node.perf_type in ('para', 'thru'):
+            if node.body_seq:
+                return enum_paths(node.body_seq, fields)
+            return [([], {})]
+        elif node.perf_type in ('until', 'para_until', 'varying', 'para_varying'):
+            # 尝试单条件（现有逻辑）
+            parsed = parse_single_condition(node.condition, fields)
+            if parsed and is_field(parsed[0], fields):
+                field, op, val = parsed
+                paths = []
+                false_sub = _cap_paths(enum_paths(node.body_seq, fields))
+                for sp_cons, sp_assign in (false_sub or [([], {})]):
+                    # PERFORM VARYING: 将 FROM 值作为 MOVE 赋值加入 Enter 路径
+                    if node.varying_from and node.varying_var:
+                        is_fld = any(f['name'] == node.varying_from for f in fields) if fields else False
+                        from_asgn = {'type': 'move', 'source_vars': [node.varying_from]} if is_fld else {'type': 'move_literal', 'literal': node.varying_from}
+                        from_assign = {node.varying_var: [from_asgn]}
+                        merged = {}
+                        for d in (from_assign, sp_assign):
+                            for k, v in d.items():
+                                merged.setdefault(k, []).extend(v if isinstance(v, list) else [v])
+                        sp_assign = merged
+                    paths.append(([(field, op, val, False)] + sp_cons, sp_assign))
+                paths.append(([(field, op, val, True)], {}))
+                return paths
+            # 尝试复合条件（AND/OR）
+            cond_tree = parse_compound_condition(node.condition, fields)
+            if cond_tree:
+                leaves = collect_leaves(cond_tree)
+                if leaves and all(is_field(l.field, fields) for l in leaves):
+                    sets = mcdc_sets(cond_tree, fields)
+                    if sets:
+                        paths = []
+                        false_sub = _cap_paths(enum_paths(node.body_seq, fields))
+                        for sp_cons, sp_assign in (false_sub or [([], {})]):
+                            # PERFORM VARYING: 将 FROM 值作为 MOVE 赋值加入 Enter 路径
+                            if node.varying_from and node.varying_var:
+                                is_fld = any(f['name'] == node.varying_from for f in fields) if fields else False
+                                from_asgn = {'type': 'move', 'source_vars': [node.varying_from]} if is_fld else {'type': 'move_literal', 'literal': node.varying_from}
+                                from_assign = {node.varying_var: [from_asgn]}
+                                merged = {}
+                                for d in (from_assign, sp_assign):
+                                    for k, v in d.items():
+                                        merged.setdefault(k, []).extend(v if isinstance(v, list) else [v])
+                                sp_assign = merged
+                            for constraints, decision in sets:
+                                if not decision:
+                                    paths.append((list(constraints) + sp_cons, sp_assign))
+                        for constraints, decision in sets:
+                            if decision:
+                                paths.append((list(constraints), {}))
+                        if paths:
+                            return paths
+        return [([], {})]
+
+    elif isinstance(node, CallNode):
+        return [([], {})]
+
+    elif isinstance(node, ExitNode):
+        return [([_STOP], {})]
+
+    elif isinstance(node, GoTo):
+        paths = enum_paths(node.body_seq, fields)
+        return [([_STOP] + c, a) for c, a in paths]
+
+    return [([], {})]
+
+
+# ── 值生成 ──
+
+def seq_numeric(seq_num: int, total_digits: int) -> str:
+    val = seq_num % (10 ** total_digits)
+    if val == 0:
+        val = 10 ** total_digits - 1
+    return str(val).zfill(total_digits)
+
+
+def seq_alpha(seq_num: int, length: int) -> str:
+    letter = chr(65 + (seq_num - 1) % 26)
+    return letter * length
+
+
+def seq_date(seq_num: int) -> str:
+    from datetime import datetime, timedelta
+    base = datetime(2000, 1, 1)
+    d = base + timedelta(days=seq_num - 1)
+    return d.strftime('%Y%m%d')
+
+
+def _is_date_field(name: str) -> bool:
+    patterns = [r'DATE', r'YYMMDD', r'YYYYMM', r'YEAR', r'MONTH', r'DAY']
+    for p in patterns:
+        if re.search(p, name.upper()):
+            return True
+    return False
+
+
+_SPECIAL_VALUES = {
+    'ZERO': '0', 'ZEROS': '0', 'ZEROES': '0',
+    'SPACE': ' ', 'SPACES': ' ',
+    'HIGH-VALUE': '\xff', 'HIGH-VALUES': '\xff',
+    'LOW-VALUE': '\x00', 'LOW-VALUES': '\x00',
+    'QUOTE': "'", 'QUOTES': "'",
+    'ALL': '',
+}
+
+
+def _apply_value(field: dict, rec: dict) -> bool:
+    """尝试应用 VALUE 子句的初始值。返回 True 表示已处理。"""
+    raw = field.get('value')
+    if raw is None:
+        return False
+    val = str(raw).strip("'\"").strip()
+    name = field['name']
+    pi = field.get('pic_info', {})
+
+    # 处理 COBOL 特殊值
+    if val.upper() in _SPECIAL_VALUES:
+        val = _SPECIAL_VALUES[val.upper()]
+
+    ftype = pi.get('type', 'unknown')
+    if ftype == 'numeric':
+        digits = pi.get('digits', 0) + pi.get('decimal', 0)
+        if digits:
+            rec[name] = val.zfill(digits)
+        else:
+            rec[name] = val
+    else:
+        length = pi.get('length', 0) or 1
+        rec[name] = val.ljust(length)[:length]
+    return True
+
+
+def _children_of(group_name: str, fields: list) -> list:
+    """返回组项目 group_name 在 fields 中的直属子字段列表（按声明顺序）。
+    终止条件：遇到同/更高级别（sibling/组边界）或 77 级（独立字段）。
+    """
+    result = []
+    group_level = None
+    found = False
+    for f in fields:
+        if not found and f['name'] == group_name:
+            group_level = f['level']
+            found = True
+            continue
+        if found:
+            if f['level'] <= group_level or f['level'] == 77:
+                break
+            # 88-level 是条件名，不计为子字段
+            if f.get('is_88'):
+                continue
+            result.append(f)
+    return result
+
+
+def _make_numeric_value(idx: int, record_num: int, total_digits: int) -> str:
+    for step in (100, 10, 1):
+        val = idx * step + record_num
+        if val < 10 ** total_digits:
+            return str(val).zfill(total_digits)
+    return str(record_num).zfill(total_digits)
+
+
+def _make_alpha_value(idx: int, record_num: int, length: int) -> str:
+    if length == 1:
+        ch = chr(65 + (idx + record_num - 2) % 26)
+        return ch
+    letter = chr(65 + (idx - 1) % 26)
+    return letter + str(record_num).zfill(length - 1)
+
+
+def make_base_record(seq_num: int, fields: list) -> dict:
+    rec = {}
+    redefines_map = {}       # 标量 REDEFINES:  parent_name → [child_names]
+    group_redefines = []     # 组 REDEFINES:    [(redef_name, target_name)]
+    filler_key_counter = 0
+    numeric_idx = 0
+    alpha_idx = 0
+    record_num = seq_num
+
+    for f in fields:
+        name = f['name']
+
+        if f.get('is_88'):
+            continue
+
+        if f.get('redefines'):
+            parent = f['redefines']
+            if f.get('pic'):
+                # 标量 REDEFINES（有 PIC，如 WS-AMOUNT-DISP REDEFINES WS-AMOUNT PIC X(9)）
+                redefines_map.setdefault(parent, []).append(name)
+                continue
+            else:
+                # 组 REDEFINES（无 PIC，如 CUST-ADDR2 REDEFINES CUST-ADDR）
+                group_redefines.append((name, parent))
+                # 不 continue — 组本身无 PIC 会在下方"组项目跳过"处理
+                # 其子字段作为独立字段正常走循环
+
+        if f.get('is_filler'):
+            if name in rec:
+                filler_key_counter += 1
+                name = f'FILLER_{filler_key_counter + 1}'
+            rec[name] = 'x' * (f.get('pic_info', {}).get('length', 0) or 1)
+            continue
+
+        # Pass 0: VALUE 子句初始值优先
+        if _apply_value(f, rec):
+            continue
+
+        # 组项目（无 PIC）跳过
+        if not f.get('pic'):
+            continue
+
+        pi = f.get('pic_info', {})
+        ftype = pi.get('type', 'unknown')
+        digits = pi.get('digits', 0)
+        decimal = pi.get('decimal', 0)
+        length = pi.get('length', 0)
+
+        if ftype == 'numeric':
+            if _is_date_field(name):
+                rec[name] = seq_date(record_num)
+            else:
+                numeric_idx += 1
+                rec[name] = _make_numeric_value(numeric_idx, record_num, digits + decimal)
+        elif ftype in ('alphanumeric', 'alphabetic'):
+            alpha_idx += 1
+            rec[name] = _make_alpha_value(alpha_idx, record_num, length or 1)
+        elif ftype == 'numeric-edited':
+            numeric_idx += 1
+            raw = _make_numeric_value(numeric_idx, record_num, digits + decimal)
+            rec[name] = raw.rjust(length)
+        else:
+            alpha_idx += 1
+            rec[name] = _make_alpha_value(alpha_idx, record_num, 8)
+
+    # Pass 2a: 标量 REDEFINES 复制
+    for parent_name, child_names in redefines_map.items():
+        if parent_name in rec:
+            for child_name in child_names:
+                rec[child_name] = rec[parent_name]
+
+    # Pass 2b: 组 REDEFINES 按位置递归复制子字段
+    for redef_name, target_name in group_redefines:
+        redef_kids = _children_of(redef_name, fields)
+        tgt_kids = _children_of(target_name, fields)
+        tgt_idx = 0
+        for i, rk in enumerate(redef_kids):
+            if tgt_idx >= len(tgt_kids):
+                break
+            if i == len(redef_kids) - 1 and len(redef_kids) < len(tgt_kids):
+                # 最后一个 REDEFINES 子字段，且目标更多 → 拼接剩余所有目标值
+                parts = [rec.get(tk['name'], '') for tk in tgt_kids[tgt_idx:]]
+                rec[rk['name']] = ''.join(parts)
+            elif i == len(redef_kids) - 1 and len(redef_kids) > len(tgt_kids):
+                # REDEFINES 子字段更多 → 最后一个 REDEFINES 子字段取最后目标值
+                rec[rk['name']] = rec.get(tgt_kids[-1]['name'], '')
+            else:
+                rec[rk['name']] = rec.get(tgt_kids[tgt_idx]['name'], '')
+            tgt_idx += 1
+
+    return rec
+
+
+# ── 约束应用 ──
+
+def _check_constraint_satisfied(rec, field_name, operator, value, want_true, fields):
+    """检查 field_name 当前值是否满足该约束。满足返回 True。"""
+    for f in fields:
+        if f['name'] == field_name:
+            pi = f.get('pic_info', {})
+            ftype = pi.get('type', 'unknown')
+            val = rec.get(field_name)
+            if val is None:
+                return False
+            if operator == 'not_in':
+                cases = value if isinstance(value, list) else []
+                return str(val) not in cases
+            if ftype == 'numeric':
+                try:
+                    num_val = int(float(str(val)))
+                    num_target = int(float(str(value)))
+                except (ValueError, TypeError):
+                    return False
+                if operator in ('>=', '>', '<', '<=', '=', '<>'):
+                    if operator == '>=':   ok = num_val >= num_target
+                    elif operator == '>':  ok = num_val > num_target
+                    elif operator == '<':  ok = num_val < num_target
+                    elif operator == '<=': ok = num_val <= num_target
+                    elif operator == '=':  ok = num_val == num_target
+                    elif operator == '<>': ok = num_val != num_target
+                    return ok == want_true
+                return True
+            else:
+                s_val = str(val).strip().upper()
+                s_target = str(value).strip().upper()
+                eq = s_val == s_target
+                if operator == '=':
+                    return eq == want_true
+                elif operator == '<>':
+                    return (not eq) == want_true
+                return True
+    return False
+
+
+_ARITH_BOUNDS = {
+    'left_big_ops':   {'>', '>=', '<>'},
+    'left_small_ops': {'<', '<='},
+}
+
+def _arith_pic_info(field_name, fields):
+    for f in fields:
+        if f['name'] == field_name.upper():
+            return f.get('pic_info', {})
+    return {}
+
+def _arith_numeric_pick(field_name, want_big, fields):
+    """为字段选一个大值或小值，返回字符串。"""
+    pi = _arith_pic_info(field_name, fields)
+    if pi.get('type') != 'numeric':
+        return None
+    digits = pi.get('digits', 0)
+    decimal = pi.get('decimal', 0)
+    total = digits + decimal
+    max_val = 10 ** total - 1
+    if want_big:
+        pick = int(max_val * 0.7)
+    else:
+        pick = 1
+    int_part = str(pick // (10 ** decimal)).zfill(digits)
+    dec_part = str(pick % (10 ** decimal)).zfill(decimal)
+    if decimal == 0:
+        return int_part
+    return int_part + dec_part
+
+def _apply_arith_constraint(rec, field_name, operator, value, want_true, fields):
+    """对算术表达式条件进行字段值 steering。
+
+    例如 A + B > C (want_true=True):
+      - 左值字段（A, B）设大 → 右值字段（C）设小
+    例如 A + B <= C (want_true=True):
+      - 左值字段设小 → 右值字段设大
+
+    这是启发式 steering，不是精确求解。
+    主要目标是保证分支可达，不保证边界值精确。
+    """
+    # 1. 提取左值表达式中的所有字段名（大写）
+    tokens = re.findall(r'\b[A-Z][A-Z0-9-]*(?:\([^)]*\))?\b', field_name.upper())
+    left_fields = [t for t in tokens if any(f['name'] == t for f in fields)]
+
+    # 2. 右值是否也为字段
+    right_field = value if any(f['name'] == value for f in fields) else None
+
+    if not left_fields:
+        logger.debug(f"算术表达式无法提取字段: {field_name}")
+        return
+
+    # 3. 确定方向：want_true 时左值应大还是小
+    if operator in _ARITH_BOUNDS['left_big_ops']:
+        left_big = want_true
+    elif operator in _ARITH_BOUNDS['left_small_ops']:
+        left_big = not want_true
+    else:
+        left_big = want_true
+
+    # 4. 设置左值字段
+    for lf in left_fields:
+        pick = _arith_numeric_pick(lf, left_big, fields)
+        if pick is not None:
+            rec[lf] = pick
+
+    # 5. 设置右值字段（如果有）
+    if right_field:
+        pick = _arith_numeric_pick(right_field, not left_big, fields)
+        if pick is not None:
+            rec[right_field] = pick
+
+
+def apply_constraint(rec, field_name, operator, value, want_true, fields, assignments=None, path_assign=None):
+    # 标准化字段名：去除括号内空格（WS-CELL ( 1, 1 ) → WS-CELL(1,1)）
+    field_name = re.sub(r'\s*([(),])\s*', r'\1', field_name)
+    # 变量下标解析：WS-FIXED-VALUE(WS-IDX) → WS-FIXED-VALUE(1)
+    vm = re.match(r'^(\w[\w-]*)\((\w[\w-]*)\)$', field_name)
+    if vm:
+        base_var, subscript_var = vm.groups()
+        if subscript_var in rec:
+            try:
+                resolved_name = f'{base_var}({int(rec[subscript_var])})'
+                if any(f['name'] == resolved_name for f in fields):
+                    apply_constraint(rec, resolved_name, operator, value, want_true, fields, assignments, path_assign)
+                    return
+            except (ValueError, TypeError):
+                pass
+    # 下标传播：无下标约束 → 应用到所有下标变体
+    base = _basename(field_name)
+    subscripted = [f for f in fields if f['name'] != base and _basename(f['name']) == base]
+    if subscripted and field_name == base:
+        for sf in subscripted:
+            apply_constraint(rec, sf['name'], operator, value, want_true, fields, assignments, path_assign)
+        return
+
+    # REDEFINES 字段的约束重定向到父字段（共享存储）
+    for f in fields:
+        if f['name'] == field_name:
+            if f.get('is_filler'):
+                return
+            if f.get('redefines'):
+                parent_name = f['redefines']
+                logger.debug(f"REDEFINES 约束重定向: {field_name} → {parent_name}")
+                apply_constraint(rec, parent_name, operator, value, want_true, fields, assignments, path_assign)
+                return
+            break
+    if assignments:
+        root_var, chain = trace_to_root(field_name, assignments, fields, path_assign)
+        if root_var != field_name:
+            new_field_name, new_op, new_val = invert_through_chain(root_var, chain, operator, value)
+            if any(f['name'] == new_field_name for f in fields):
+                field_name, operator, value = new_field_name, new_op, new_val
+
+    # 如果当前值已满足该约束，跳过覆盖（保持先前约束的一致性）
+    if _check_constraint_satisfied(rec, field_name, operator, value, want_true, fields):
+        return
+
+    if operator == 'not_in':
+        for f in fields:
+            if f['name'] == field_name:
+                pi = f.get('pic_info', {})
+                cases = value if isinstance(value, list) else []
+                ftype = pi.get('type', 'unknown')
+                if ftype in ('alphanumeric', 'alphabetic'):
+                    for c in 'ABCDEFGHIJKLMNOPQRSTUVWXYZ':
+                        if c not in cases:
+                            rec[field_name] = c.ljust(pi.get('length', 1), c)
+                            return
+                else:
+                    for n in range(1, 100):
+                        if str(n) not in cases:
+                            rec[field_name] = str(n).zfill(pi.get('digits', 0) + pi.get('decimal', 0))
+                            return
+        return
+    # 字段间比较（值侧也是字段名）
+    if any(f['name'] == value for f in fields):
+        if re.search(r'[+\-*/]', field_name):
+            _apply_arith_constraint(rec, field_name, operator, value, want_true, fields)
+        else:
+            logger.debug(f"字段间比较约束跳过：{field_name} {operator} {value}")
+        return
+    for f in fields:
+        if f['name'] == field_name:
+            pi = f.get('pic_info', {})
+            val = satisfying_value(pi, operator, value, want_true)
+            rec[field_name] = val
+            return
+
+
+# ── 记录生成入口 ──
+
+def sync_redefined_fields(rec, fields):
+    """赋值/约束后同步 REDEFINES 字段：父字段的值拷贝到所有 REDEFINES 子字段。"""
+    redefines_map = {}
+    group_redefines = []
+    for f in fields:
+        if f.get('is_88') or f.get('is_filler'):
+            continue
+        if f.get('redefines') and f.get('pic'):
+            redefines_map.setdefault(f['redefines'], []).append(f['name'])
+        elif f.get('redefines') and not f.get('pic'):
+            group_redefines.append((f['name'], f['redefines']))
+    for parent_name, child_names in redefines_map.items():
+        if parent_name in rec:
+            for child_name in child_names:
+                rec[child_name] = rec[parent_name]
+    for redef_name, target_name in group_redefines:
+        redef_kids = _children_of(redef_name, fields)
+        tgt_kids = _children_of(target_name, fields)
+        tgt_idx = 0
+        for i, rk in enumerate(redef_kids):
+            if tgt_idx >= len(tgt_kids):
+                break
+            if i == len(redef_kids) - 1 and len(redef_kids) < len(tgt_kids):
+                parts = [rec.get(tk['name'], '') for tk in tgt_kids[tgt_idx:]]
+                rec[rk['name']] = ''.join(parts)
+            elif i == len(redef_kids) - 1 and len(redef_kids) > len(tgt_kids):
+                rec[rk['name']] = rec.get(tgt_kids[-1]['name'], '')
+            else:
+                rec[rk['name']] = rec.get(tgt_kids[tgt_idx]['name'], '')
+            tgt_idx += 1
+
+
+def apply_occurs_depending(rec, fields):
+    """根据 OCCURS DEPENDING ON 变量的当前值，清零超范围的下标字段。"""
+    for f in fields:
+        dep_var = f.get('occurs_depending')
+        if not dep_var:
+            continue
+        name = f['name']
+        m = re.search(r'\((\d+)\)$', name)
+        if not m:
+            continue
+        sub = int(m.group(1))
+        max_val = int(rec.get(dep_var, 0))
+        if sub <= max_val:
+            continue
+        pi = f.get('pic_info', {})
+        ftype = pi.get('type', 'unknown')
+        length = pi.get('length', 0) or 1
+        if ftype == 'numeric':
+            rec[name] = '0' * (pi.get('digits', 0) + pi.get('decimal', 0))
+        elif ftype in ('alphanumeric', 'alphabetic'):
+            rec[name] = ' ' * length
+        else:
+            rec[name] = '0' * length
+
+
+def _non_match_for(cond_leaf, fields):
+    if not fields or not cond_leaf:
+        return None
+    base = re.sub(r'\s*\(.*?\)\s*$', '', cond_leaf.field)
+    for f in fields:
+        if re.sub(r'\s*\(.*?\)\s*$', '', f['name']) == base:
+            pic = f.get('pic_info', {})
+            if pic.get('type') == 'numeric':
+                return '0'
+            return ' '
+    return None
+
+
+def _enum_search_paths(node, fields):
+    # 从条件字段名推断 OCCURS 数；如 WS-CODE-VAL(WS-IDX) → 查 WS-CODE-VAL(j) 最大 j
+    occurs_count = 1
+    if node.when_list and node.cond_trees and node.cond_trees[0]:
+        ct = node.cond_trees[0]
+        if isinstance(ct, CondLeaf):
+            base = re.sub(r'\s*\(.*?\)\s*$', '', ct.field)
+            for f in fields:
+                m = re.match(rf'^{re.escape(base)}\((\d+)\)$', f['name'])
+                if m:
+                    occurs_count = max(occurs_count, int(m.group(1)))
+            if occurs_count <= 1:
+                # 再查父组名下各字段的后缀
+                parent = node.table_name
+                for f in fields:
+                    m = re.match(rf'^{re.escape(parent)}\((\d+)\)$', f['name'])
+                    if m:
+                        occurs_count = max(occurs_count, int(m.group(1)))
+
+    paths = []
+    for i, (cond_text, body_seq) in enumerate(node.when_list):
+        cond_tree = node.cond_trees[i] if i < len(node.cond_trees) else None
+        sub = _cap_paths(enum_paths(body_seq, fields))
+        if not sub:
+            sub = [([], {})]
+
+        extra_assign = {}
+        if cond_tree and isinstance(cond_tree, CondLeaf):
+            base = re.sub(r'\s*\(.*?\)\s*$', '', cond_tree.field)
+            matching_val = cond_tree.value
+            elem_key = f'{base}({i + 1})'
+            extra_assign[elem_key] = [{'type': 'move_literal', 'literal': matching_val}]
+            non_match = _non_match_for(cond_tree, fields) or ' '
+            for j in range(i):
+                prev_key = f'{base}({j + 1})'
+                extra_assign[prev_key] = [{'type': 'move_literal', 'literal': non_match}]
+
+        for sp_cons, sp_assign in (sub or [([], {})]):
+            merged_assign = dict(extra_assign)
+            for k, v in sp_assign.items():
+                merged_assign.setdefault(k, []).extend(v if isinstance(v, list) else [v])
+            paths.append((sp_cons, merged_assign))
+
+    if node.has_at_end:
+        sub = _cap_paths(enum_paths(node.at_end_seq, fields))
+        for sp_cons, sp_assign in (sub or [([], {})]):
+            extra_assign = {}
+            non_match = ' '
+            if node.when_list:
+                ct = node.cond_trees[0]
+                if ct and isinstance(ct, CondLeaf):
+                    non_match = _non_match_for(ct, fields) or ' '
+                    base = re.sub(r'\s*\(.*?\)\s*$', '', ct.field)
+                    for j in range(max(occurs_count, 1)):
+                        extra_assign[f'{base}({j + 1})'] = [{'type': 'move_literal', 'literal': non_match}]
+            merged_assign = dict(extra_assign)
+            for k, v in sp_assign.items():
+                merged_assign.setdefault(k, []).extend(v if isinstance(v, list) else [v])
+            paths.append((sp_cons, merged_assign))
+
+    return paths
+
+
+def generate_records(branch_paths_with_assigns, data_fields, base_assignments=None, file_sec=None):
+    """生成测试数据记录。
+    branch_paths_with_assigns: list of (constraints, path_assignments).
+    base_assignments: 全局 assignments dict (用于 trace_to_root).
+    返回: (records, kept_path_cons) — kept_path_cons 是与 records 一一对应的约束。
+    """
+    records = []
+    kept_path_cons = []
+    if branch_paths_with_assigns:
+        for seq, (path_cons, path_assign) in enumerate(branch_paths_with_assigns, start=1):
+            path_cons = _filter_stop(path_cons)
+            rec = make_base_record(seq, data_fields)
+            # Pass A: 先传播赋值（MOVE/COMPUTE/READ INTO 等），模拟到决策点前的程序状态
+            if isinstance(path_assign, dict):
+                propagate_assignments(rec, path_assign, data_fields, file_sec=file_sec)
+            # Pass A.5: 检查约束是否经过链追溯到字面量截断（不可能路径）
+            skip_impossible = False
+            if base_assignments and isinstance(path_assign, dict):
+                for c in path_cons:
+                    if len(c) == 4 and not skip_impossible:
+                        field, op, val, want = c
+                        root_var, chain = trace_to_root(field, base_assignments, data_fields, path_assign)
+                        if root_var != field:
+                            new_fn, new_op, new_val = invert_through_chain(root_var, chain, op, val)
+                            if any(f['name'] == new_fn for f in data_fields):
+                                asgn_val = path_assign.get(root_var)
+                                if asgn_val is not None:
+                                    asgn_list = asgn_val if isinstance(asgn_val, list) else [asgn_val]
+                                    if asgn_list and asgn_list[-1]['type'] == 'move_literal' and root_var in rec:
+                                        if not _check_constraint_satisfied(rec, root_var, new_op, new_val, want, data_fields):
+                                            skip_impossible = True
+                                            break
+            if skip_impossible:
+                continue
+            # Pass B: 约束覆盖（确保决策条件满足，覆盖 MOVE 带来的值）
+            for c in path_cons:
+                if len(c) == 4:
+                    field, op, val, want = c
+                    apply_constraint(rec, field, op, val, want, data_fields, base_assignments, path_assign)
+            # Pass B.5: 前向再传播变量间MOVE，保持约束修改后的链一致性
+            if isinstance(path_assign, dict):
+                forward = {}
+                for tgt, asgn_val in path_assign.items():
+                    asgn_list = asgn_val if isinstance(asgn_val, list) else [asgn_val]
+                    filtered = [a for a in asgn_list if a['type'] == 'move' and a.get('source_vars')]
+                    if filtered:
+                        forward[tgt] = filtered
+                if forward:
+                    propagate_assignments(rec, forward, data_fields, file_sec=file_sec)
+            # Pass C: 同步 REDEFINES（确保共享存储一致）
+            sync_redefined_fields(rec, data_fields)
+            # Pass D: OCCURS DEPENDING ON — 清零超范围的下标字段
+            apply_occurs_depending(rec, data_fields)
+
+            records.append(rec)
+            kept_path_cons.append(path_cons)
+    if not records:
+        rec = make_base_record(1, data_fields)
+        if base_assignments:
+            propagate_assignments(rec, base_assignments, data_fields, file_sec=file_sec)
+        records.append(rec)
+        kept_path_cons.append([])
+    return records, kept_path_cons

cobol_testgen/grammar.lark

@@ -0,0 +1,35 @@
+start: data_div_content
+data_div_content: (file_section | working_storage | linkage)*
+file_section: "FILE" "SECTION" DOT fd+
+fd: "FD" NAME FD_SUFFIX data_item+
+FD_SUFFIX: /(?:"[^"]*"|'[^']*'|[^.])*\./
+working_storage: "WORKING-STORAGE" "SECTION" DOT data_item*
+linkage: "LINKAGE" "SECTION" DOT data_item*
+data_item: level_num (NAME | "FILLER") clause* DOT
+level_num: LEVEL
+clause: pic_clause | value_clause | occurs_clause | redefines_clause | usage_clause
+      | "SYNC" | "SYNCHRONIZED"
+      | "JUSTIFIED" "RIGHT"?
+      | "BLANK" "WHEN" "ZERO"
+      | "GLOBAL" | "EXTERNAL"
+pic_clause: "PIC" "IS"? PICTURE_STRING
+value_clause: "VALUE" "IS"? value_literal+
+value_literal: INT | SIGNED_NUMBER | STRING | SQSTRING
+             | "ZERO" | "ZEROS" | "ZEROES"
+             | "SPACE" | "SPACES"
+             | "HIGH-VALUE" | "HIGH-VALUES"
+             | "LOW-VALUE" | "LOW-VALUES"
+SQSTRING: /'[^']*'/
+redefines_clause: "REDEFINES" NAME
+occurs_clause: "OCCURS" INT "TIMES"? ("DEPENDING" "ON" NAME)?
+usage_clause: USAGE_VAL
+USAGE_VAL: "COMP" | "COMP-3" | "COMP-5" | "BINARY" | "PACKED-DECIMAL" | "DISPLAY"
+LEVEL: /0[1-9]|[1-4][0-9]|49|77|88/
+NAME: /[A-Z][A-Z0-9-]*/
+PICTURE_STRING: /[0-9A-Z()+,\-*\/V]+/i
+INT: /[0-9]+/
+DOT: /\./
+%import common.SIGNED_NUMBER
+%import common.ESCAPED_STRING -> STRING
+%import common.WS
+%ignore WS

cobol_testgen/models.py

@@ -0,0 +1,163 @@
+"""COBOL数据模型 — 所有层共享，无外部依赖"""
+
+from dataclasses import dataclass, field
+
+
+# ── 字段定义 ──
+
+@dataclass
+class PicInfo:
+    type: str = 'unknown'               # "numeric" | "alphanumeric" | "alphabetic"
+    digits: int = 0
+    decimal: int = 0
+    length: int = 0
+    signed: bool = False
+
+
+@dataclass
+class FieldDef:
+    name: str
+    level: int
+    pic: str | None = None
+    pic_info: PicInfo | None = None
+    is_filler: bool = False
+    occurs_count: int = 0
+    occurs_depending: str | None = None
+    redefines: str | None = None
+    usage: str | None = None           # "COMP" | "COMP-3" | "BINARY" | "PACKED-DECIMAL" | ...
+    value: str | None = None
+    values: list[str] | None = None
+    is_88: bool = False
+    parent: str | None = None
+    section: str | None = None
+
+
+# ── 分支树 ──
+
+class BrSeq:
+    def __init__(self):
+        self.children = []
+
+    def add(self, child):
+        self.children.append(child)
+
+
+class BrIf:
+    def __init__(self, condition):
+        self.condition = condition
+        self.cond_tree = None           # 由 core.py 在解析时赋值
+        self.true_seq = BrSeq()
+        self.false_seq = BrSeq()
+
+
+class BrEval:
+    def __init__(self, subject):
+        self.subject = subject
+        self.subjects = []        # ALSO 多主体: ['WS-A', 'WS-B']，空=普通模式
+        self.when_list = []
+        self.other_seq = BrSeq()
+        self.has_other = False
+
+
+class BrPerform:
+    def __init__(self, perf_type, condition=None, target=None, thru=None, times=None,
+                 varying_var=None, varying_from=None, varying_by=None):
+        self.perf_type = perf_type
+        self.condition = condition
+        self.target = target
+        self.thru = thru
+        self.times = times
+        self.varying_var = varying_var
+        self.varying_from = varying_from
+        self.varying_by = varying_by
+        self.body_seq = BrSeq()
+
+
+class Assign:
+    """赋值节点：MOVE/COMPUTE/ADD/SUBTRACT/MULTIPLY/DIVIDE"""
+    def __init__(self, target: str, source_info: dict):
+        self.target = target
+        self.source_info = source_info
+
+
+class CallNode:
+    """CALL 子程序调用节点（黑盒模式）"""
+    def __init__(self, program_name: str, using_params: list = None):
+        self.program_name = program_name
+        self.using_params = using_params or []
+        # using_params: [{"name": "WS-A", "mechanism": "reference"}, ...]
+        # mechanism: "reference" | "content" | "value"
+
+
+# ── 条件树 ──
+
+class CondLeaf:
+    def __init__(self, field, op, value):
+        self.field = field
+        self.op = op
+        self.value = value
+
+
+class CondNot:
+    def __init__(self, child):
+        self.child = child
+
+
+class CondAnd:
+    def __init__(self, left, right):
+        self.left = left
+        self.right = right
+
+
+class CondOr:
+    def __init__(self, left, right):
+        self.left = left
+        self.right = right
+
+
+class BrSearch:
+    """SEARCH / SEARCH ALL 表查找"""
+    def __init__(self, table_name, is_all=False, varying=None):
+        self.table_name = table_name
+        self.is_all = is_all
+        self.varying = varying.upper() if varying else None
+        self.at_end_seq = BrSeq()
+        self.when_list = []       # [(condition_text, BrSeq)]
+        self.cond_trees = []      # [cond_tree, ...]
+        self.has_at_end = False
+
+
+class GoTo:
+    """GO TO 节点：无条件跳转到指定段落"""
+    def __init__(self, target: str, body_seq: 'BrSeq' = None):
+        self.target = target
+        self.body_seq = body_seq or BrSeq()
+
+
+class ExitNode:
+    """控制流退出节点：EXIT PARAGRAPH / EXIT PERFORM / EXIT SECTION / EXIT PROGRAM"""
+    def __init__(self, exit_type: str):
+        self.exit_type = exit_type
+
+
+# ── 约束路径 ──
+
+Constraint = tuple   # (field, op, value, want_true)
+Path = list[Constraint]
+
+
+# ── 解析错误 ──
+
+@dataclass
+class ParseError:
+    line: int
+    message: str
+    severity: str = 'warning'
+
+
+@dataclass
+class ProcParseResult:
+    tree: BrSeq | None = None
+    assignments: dict = field(default_factory=dict)
+    errors: list[ParseError] = field(default_factory=list)
+    fallback_to_ai: bool = False

cobol_testgen/output.py

@@ -0,0 +1,118 @@
+"""输出层：JSON输出（按文件分组入出力 + 工作存储区分）"""
+
+import json
+from pathlib import Path
+
+
+_INVERSE_OP = {'>': '<=', '<': '>=', '=': '<>', '>=': '<', '<=': '>'}
+
+
+def _scenario_text(path_cons):
+    parts = []
+    for c in path_cons:
+        if len(c) != 4:
+            continue
+        field, op, val, want = c
+        if op == 'not_in':
+            desc = f"{field} not in {val}" if want else f"{field} in {val}"
+        elif not want:
+            desc = f"{field} {_INVERSE_OP.get(op, '?' + op)} {val}"
+        else:
+            desc = f"{field} {op} {val}"
+        parts.append(desc)
+    return ', '.join(parts)
+
+
+def output_json(records, outpath, roles=None, fd_fields=None, field_to_fd=None,
+                open_dir=None, path_cons_list=None):
+    outpath.parent.mkdir(parents=True, exist_ok=True)
+    if not roles:
+        with open(outpath, 'w', encoding='utf-8') as f:
+            json.dump(records, f, ensure_ascii=False, indent=2)
+        return
+
+    # FD direction lookup
+    out = []
+    for i, rec in enumerate(records):
+        inp = {}
+        out_exp = {}
+        ws = {}
+
+        # Group by FD
+        if fd_fields and field_to_fd:
+            for fd_name, fds_set in fd_fields.items():
+                direction = (open_dir or {}).get(fd_name, '')
+                inp_block = {}
+                out_block = {}
+                for fname in fds_set:
+                    if fname not in rec:
+                        continue
+                    r = roles.get(fname, 'unused')
+                    val = rec[fname]
+                    if direction in ('INPUT', 'I-O') and r in ('input', 'inout'):
+                        inp_block[fname] = val
+                    if direction in ('OUTPUT', 'I-O') and r in ('output', 'inout'):
+                        out_block[fname] = val
+                if inp_block:
+                    inp[fd_name] = inp_block
+                if out_block:
+                    out_exp[fd_name] = out_block
+
+        # Working-storage: not belonging to any FD
+        for name, val in rec.items():
+            if not field_to_fd or name not in field_to_fd:
+                ws[name] = val
+
+        entry = {
+            'input': inp,
+            'expected_output': out_exp,
+            'working_storage': ws,
+        }
+
+        if path_cons_list and i < len(path_cons_list):
+            text = _scenario_text(path_cons_list[i])
+            if text:
+                entry['scenario'] = text
+
+        out.append(entry)
+
+    with open(outpath, 'w', encoding='utf-8') as f:
+        json.dump(out, f, ensure_ascii=False, indent=2)
+
+
+def output_input_files(records, outdir, stem, roles, fd_fields, field_to_fd, open_dir):
+    """按 FD 名拆分出力入力 JSON 文件。
+    每个 INPUT / I-O 方向 FD 生成一个文件：{stem}_{fd_name}.json
+    内容为路径数 × 记录，每条只含该 FD 的入力字段值。
+    """
+    input_fds = {}
+    for fd_name, fds_set in fd_fields.items():
+        direction = (open_dir or {}).get(fd_name, '')
+        if direction not in ('INPUT', 'I-O'):
+            continue
+        has_input = any(roles.get(fname, 'unused') in ('input', 'inout') for fname in fds_set)
+        if not has_input:
+            continue
+        input_fds[fd_name] = fds_set
+
+    if not input_fds:
+        return
+
+    outdir.mkdir(parents=True, exist_ok=True)
+
+    for fd_name, fds_set in input_fds.items():
+        fd_records = []
+        direction = (open_dir or {}).get(fd_name, '')
+        for rec in records:
+            fd_rec = {}
+            for fname in fds_set:
+                r = roles.get(fname, 'unused')
+                if direction in ('INPUT', 'I-O') and r in ('input', 'inout'):
+                    if fname in rec:
+                        fd_rec[fname] = rec[fname]
+            if fd_rec:
+                fd_records.append(fd_rec)
+
+        outpath = outdir / f'{stem}_{fd_name}.json'
+        with open(outpath, 'w', encoding='utf-8') as f:
+            json.dump(fd_records, f, ensure_ascii=False, indent=2)

cobol_testgen/read.py

@@ -388,17 +388,32 @@ def parse_data_division(data_div_text: str) -> list[FieldDef]:
 
 
 def parse_file_control(source: str) -> dict:
-    """?? FILE-CONTROL??? {?????: ?????}"""
+    """Parse FILE-CONTROL paragraph.
+
+    Returns dict:
+        {filename: {"assign_to": str, "organization": str | None}}
+    """
     m = re.search(r'FILE-CONTROL\.(.*?)(?=DATA\s+DIVISION|\Z)', source, re.DOTALL | re.IGNORECASE)
     if not m:
         return {}
     fc = m.group(1)
     result = {}
-    for m in re.finditer(
+    for sel_m in re.finditer(
         r'SELECT\s+(\w[\w-]*)\s+[^.]*?\bASSIGN\s+TO\s+(["\'])(.*?)\2',
         fc, re.IGNORECASE
     ):
-        result[m.group(1).upper()] = m.group(3).upper()
+        fname = sel_m.group(1).upper()
+        assign_to = sel_m.group(3).upper()
+        # Extract ORGANIZATION clause within this SELECT statement
+        org_m = re.search(
+            r'ORGANIZATION\s+(?:IS\s+)?(\w[\w-]*)',
+            sel_m.group(0), re.IGNORECASE
+        )
+        org = org_m.group(1).upper() if org_m else None
+        result[fname] = {
+            "assign_to": assign_to,
+            "organization": org,
+        }
     return result