Fix line numbers

PythonScripts/audit_translations/auditor.py

@@ -291,11 +291,84 @@ def resolve_issue_line(rule: RuleInfo, kind: str, token: Optional[str] = None) -
     return lines[0] if lines else rule.line_number
 
 
+def structure_token_occurrence_index(tokens: List[str], position: int) -> Optional[int]:
+    """
+    Return which occurrence of a token appears at a given absolute token position.
+
+    Example: for ["test:", "if:", "test:"], position 2 returns 1.
+    """
+    if position < 0 or position >= len(tokens):
+        return None
+    token = tokens[position]
+    return sum(1 for current in tokens[:position] if current == token)
+
+
+def resolve_structure_issue_lines(diff: RuleDifference) -> Optional[Tuple[int, int]]:
+    """
+    Resolve stable line anchors for a structural rule difference.
+
+    Strategy:
+    - Use position-aware token occurrence matching when possible.
+    - For insert/delete cases (one side missing token), anchor to the previous
+      shared structural token; if unavailable, anchor to `replace:`.
+    """
+    en_tokens = extract_structure_elements(diff.english_rule.data)
+    tr_tokens = extract_structure_elements(diff.translated_rule.data)
+    en_token, tr_token, mismatch_pos = first_structure_mismatch(en_tokens, tr_tokens)
+
+    if mismatch_pos < 0:
+        return None
+
+    # Insertion/deletion: anchor to the previous shared token if possible.
+    if en_token is None or tr_token is None:
+        anchor_pos = mismatch_pos - 1
+        if (
+            anchor_pos >= 0
+            and anchor_pos < len(en_tokens)
+            and anchor_pos < len(tr_tokens)
+            and en_tokens[anchor_pos] == tr_tokens[anchor_pos]
+        ):
+            anchor_token = en_tokens[anchor_pos]
+            en_occ = structure_token_occurrence_index(en_tokens, anchor_pos)
+            tr_occ = structure_token_occurrence_index(tr_tokens, anchor_pos)
+            if en_occ is not None and tr_occ is not None:
+                line_en = resolve_issue_line_at_position(diff.english_rule, "structure", anchor_token, en_occ)
+                line_tr = resolve_issue_line_at_position(diff.translated_rule, "structure", anchor_token, tr_occ)
+                if line_en is not None and line_tr is not None:
+                    return line_en, line_tr
+
+        # Fallback: anchor both sides to replace, which is the rule body entrypoint.
+        line_en = resolve_issue_line(diff.english_rule, "structure", "replace:") or diff.english_rule.line_number
+        line_tr = resolve_issue_line(diff.translated_rule, "structure", "replace:") or diff.translated_rule.line_number
+        return line_en, line_tr
+
+    # Exact token available on both sides: resolve by occurrence index at mismatch.
+    en_occ = structure_token_occurrence_index(en_tokens, mismatch_pos)
+    tr_occ = structure_token_occurrence_index(tr_tokens, mismatch_pos)
+    if en_occ is not None and tr_occ is not None:
+        line_en = resolve_issue_line_at_position(diff.english_rule, "structure", en_token, en_occ)
+        line_tr = resolve_issue_line_at_position(diff.translated_rule, "structure", tr_token, tr_occ)
+        if line_en is not None and line_tr is not None:
+            return line_en, line_tr
+
+    line_en = resolve_issue_line(diff.english_rule, "structure", en_token)
+    line_tr = resolve_issue_line(diff.translated_rule, "structure", tr_token)
+    if line_en is None or line_tr is None:
+        return None
+    return line_en, line_tr
+
+
 def collect_issues(
     result: ComparisonResult,
     file_name: str,
     language: str,
 ) -> List[dict]:
+    """
+    Flatten a ComparisonResult into one normalized dictionary per issue.
+
+    This is the canonical bridge from parser/diff objects to serializable
+    records consumed by JSONL output, snapshot tests, and line-level assertions.
+    """
     issues = []
 
     for rule in result.missing_rules:
@@ -345,23 +418,10 @@ def collect_issues(
         rule = diff.english_rule
         issue = issue_base(rule, file_name, language)
         if diff.diff_type == "structure":
-            en_tokens = extract_structure_elements(diff.english_rule.data)
-            tr_tokens = extract_structure_elements(diff.translated_rule.data)
-            en_token, tr_token, mismatch_pos = first_structure_mismatch(en_tokens, tr_tokens)
-
-            # Skip reporting when tokens are misaligned (both exist but differ)
-            # This avoids misleading line numbers when entire blocks are missing/added
-            # We only report when one is None (clear case of missing element)
-            if en_token is not None and tr_token is not None and en_token != tr_token:
-                continue
-
-            issue_line_en = resolve_issue_line(diff.english_rule, "structure", en_token)
-            issue_line_tr = resolve_issue_line(diff.translated_rule, "structure", tr_token)
-
-            # Skip reporting structure differences where we can't find both tokens
-            # This avoids misleading line numbers when blocks are missing
-            if issue_line_en is None or issue_line_tr is None:
+            structure_lines = resolve_structure_issue_lines(diff)
+            if structure_lines is None:
                 continue
+            issue_line_en, issue_line_tr = structure_lines
         else:
             issue_line_en = resolve_issue_line(diff.english_rule, diff.diff_type)
             issue_line_tr = resolve_issue_line(diff.translated_rule, diff.diff_type)
@@ -438,20 +498,10 @@ def add_issue(rule: RuleInfo, issue_type: str, payload: Dict[str, Any]) -> None:
 
     for diff in result.rule_differences:
         if diff.diff_type == "structure":
-            en_tokens = extract_structure_elements(diff.english_rule.data)
-            tr_tokens = extract_structure_elements(diff.translated_rule.data)
-            en_token, tr_token, mismatch_pos = first_structure_mismatch(en_tokens, tr_tokens)
-
-            # Skip reporting when tokens are misaligned (both exist but differ)
-            # This avoids misleading line numbers when entire blocks are missing/added
-            if en_token is not None and tr_token is not None and en_token != tr_token:
-                continue
-
-            line_en = resolve_issue_line(diff.english_rule, "structure", en_token)
-            line_tr = resolve_issue_line(diff.translated_rule, "structure", tr_token)
-            # Skip structure diffs where we can't find both tokens
-            if line_en is None or line_tr is None:
+            structure_lines = resolve_structure_issue_lines(diff)
+            if structure_lines is None:
                 continue
+            line_en, line_tr = structure_lines
         else:
             line_en = resolve_issue_line(diff.english_rule, diff.diff_type)
             line_tr = resolve_issue_line(diff.translated_rule, diff.diff_type)

PythonScripts/audit_translations/tests/fixtures/repro/README.md

@@ -0,0 +1,7 @@
+# Repro Fixtures
+
+This folder contains minimal fixtures from snapshots of the Rules at some point in time.
+
+The first use of this is comparing `per-fraction` between the English and Norwegian rules as of 2026-02-17, which had a bug of the wrong lines being shown.
+- `en/per_fraction.yaml`
+- `nb/per_fraction.yaml`

PythonScripts/audit_translations/tests/fixtures/repro/en/per_fraction.yaml

@@ -0,0 +1,12 @@
+- name: per-fraction
+  tag: fraction
+  match:
+  - "BaseNode(*[1])[contains(@data-intent-property, ':unit') or"
+  - "               ( self::m:mrow and count(*)=3 and"  # maybe a bit paranoid checking the structure...
+  - "                 *[1][self::m:mn] and *[2][.='\u2062'] and BaseNode(*[3])[contains(@data-intent-property, ':unit')] ) ] and"
+  - "BaseNode(*[2])[contains(@data-intent-property, ':unit')] "
+  replace:
+  - x: "*[1]"
+  - t: "per"      # phrase('5 meters 'per' second)
+  - x: "*[2]"
+

PythonScripts/audit_translations/tests/fixtures/repro/nb/per_fraction.yaml

@@ -0,0 +1,16 @@
+- name: per-fraction
+  tag: fraction
+  match:
+  - "BaseNode(*[1])[contains(@data-intent-property, ':unit') or"
+  - "               ( self::m:mrow and count(*)=3 and"  # maybe a bit paranoid checking the structure...
+  - "                 *[1][self::m:mn] and *[2][.='\u2062'] and BaseNode(*[3])[contains(@data-intent-property, ':unit')] ) ] and"
+  - "BaseNode(*[2])[contains(@data-intent-property, ':unit') or (contains(@data-intent-property, ':unit') and .='t')] "
+  replace:
+  - x: "*[1]"
+  - T: "per"      # phrase('5 meters 'per' second)
+  - test: 
+      if: "*[2]='t'"
+      then: [T: "time"]
+      else: 
+       - x: "*[2]"
+

PythonScripts/audit_translations/tests/golden/jsonl/de.json

@@ -115,8 +115,8 @@
     "rule_name": "struct-rule",
     "rule_tag": "mi",
     "rule_key": "struct-rule|mi",
-    "issue_line_en": 9,
-    "issue_line_tr": 1,
+    "issue_line_en": 7,
+    "issue_line_tr": 7,
     "rule_line_en": 1,
     "rule_line_tr": 1,
     "issue_type": "rule_difference",
@@ -161,16 +161,34 @@
     "english_snippet": "$Verbosity!='Terse', $Setting = 'Value', parent::m:minus, *[2][.='2']",
     "translated_snippet": "$Setting = 'Value', parent::m:minus, *[2][.='2']",
     "untranslated_texts": [],
-    "_explanation": "structure_misaligned.yaml: English has extra test block causing misalignment. Fix filters out misleading structure differences but reports condition difference."
+    "_explanation": "structure_misaligned.yaml: condition difference remains reported."
+  },
+  {
+    "language": "de",
+    "file": "structure_misaligned.yaml",
+    "rule_name": "misaligned-structure",
+    "rule_tag": "root",
+    "rule_key": "misaligned-structure|root",
+    "issue_line_en": 11,
+    "issue_line_tr": 11,
+    "rule_line_en": 1,
+    "rule_line_tr": 1,
+    "issue_type": "rule_difference",
+    "diff_type": "structure",
+    "description": "Rule structure differs (test/if/then/else blocks)",
+    "english_snippet": "replace: test: if: then: test: if: then: test: if: then: else: test: if: then: else:",
+    "translated_snippet": "replace: test: if: then: test: if: then: else: test: if: then: else:",
+    "untranslated_texts": [],
+    "_explanation": "structure_misaligned.yaml: structure substitutions/realignments are now reported with position-aware anchors."
   },
   {
     "language": "de",
     "file": "structure_missing_else.yaml",
     "rule_name": "missing-else-block",
     "rule_tag": "root",
     "rule_key": "missing-else-block|root",
-    "issue_line_en": 8,
-    "issue_line_tr": 1,
+    "issue_line_en": 7,
+    "issue_line_tr": 7,
     "rule_line_en": 1,
     "rule_line_tr": 1,
     "issue_type": "rule_difference",

PythonScripts/audit_translations/tests/golden/rich/cli_calculus_verbose.golden

@@ -43,7 +43,7 @@
                   en: $Verbosity!='Terse', not(IsNode(*[1], 'leaf'))
                   tr: not(IsNode(*[1], 'leaf'))
           Structure Differences [1]
-              • (line 38 en, 18 tr)
+              • (line 40 en, 25 tr)
                   Rule structure differs (test/if/then/else blocks)
                   en: replace: test: if: then: test: if: then:
                   tr: replace: test: if: then:

PythonScripts/audit_translations/tests/golden/rich/structure_diff_nonverbose.golden

@@ -7,5 +7,5 @@
   ≠ Rule Issues [1] (grouped by rule and issue type)
       • struct-rule (mi)
           Structure Differences [1]
-              • (line 9 en, 1 tr)
+              • (line 7 en, 7 tr)
                   Rule structure differs (test/if/then/else blocks)

PythonScripts/audit_translations/tests/golden/rich/structure_diff_verbose.golden

@@ -7,7 +7,7 @@
   ≠ Rule Issues [1] (grouped by rule and issue type)
       • struct-rule (mi)
           Structure Differences [1]
-              • (line 9 en, 1 tr)
+              • (line 7 en, 7 tr)
                   Rule structure differs (test/if/then/else blocks)
                   en: replace: test: if: then: else:
                   tr: replace: test: if: then: