docrev 0.9.13 → 0.9.14

package/lib/commands/section-boundaries.ts

@@ -1,82 +1,82 @@
-/**
- * Compute section boundaries in a DOCX from its real heading paragraphs.
- *
- * Given the configured `sections.yaml` and the headings extracted via
- * `extractHeadings()`, return one boundary per section file with text
- * positions in the same coordinate system as `CommentAnchorData.docPosition`.
- *
- * Matching is by heading text (primary header + aliases, case-insensitive).
- * This replaces the older keyword-search-in-body-text approach which would
- * pick up section names that happen to appear inside prose ("results across
- * countries") or in structured-abstract labels where paragraph boundaries
- * are lost in concatenation.
- */
-
-import type { DocxHeading } from '../word-extraction.js';
-import type { SectionConfig } from '../types.js';
-
-export interface SectionBoundary {
-  file: string;
-  start: number;
-  end: number;
-}
-
-export function computeSectionBoundaries(
-  sections: Record<string, SectionConfig>,
-  headings: DocxHeading[],
-  docLength?: number,
-): SectionBoundary[] {
-  const matched: SectionBoundary[] = [];
-
-  // Only consider top-level (Heading1-style) when level info is available;
-  // when level==0 (unparseable style), fall back to all headings.
-  const haveLevels = headings.some(h => h.level > 0);
-  const candidates = haveLevels ? headings.filter(h => h.level === 1) : headings;
-
-  for (const [file, cfg] of Object.entries(sections)) {
-    const targets = [cfg.header, ...(cfg.aliases || [])]
-      .filter(Boolean)
-      .map(s => s.toLowerCase().trim());
-
-    let firstMatch = -1;
-    for (const h of candidates) {
-      const text = h.text.toLowerCase().trim();
-      if (targets.includes(text)) {
-        firstMatch = h.docPosition;
-        break;
-      }
-    }
-
-    // Fallback: if no level-1 hit, allow any-level match (handles single-level docs)
-    if (firstMatch < 0 && haveLevels) {
-      for (const h of headings) {
-        const text = h.text.toLowerCase().trim();
-        if (targets.includes(text)) {
-          firstMatch = h.docPosition;
-          break;
-        }
-      }
-    }
-
-    if (firstMatch >= 0) {
-      matched.push({ file, start: firstMatch, end: Number.MAX_SAFE_INTEGER });
-    }
-  }
-
-  // Sort by start position and tighten each end to the next start.
-  // The last section's end is capped at docLength when known, otherwise
-  // left at MAX_SAFE_INTEGER. Without the cap, single-section configs
-  // produce a sectionLength of ~9e15, collapsing proportional-position
-  // math in insertCommentsIntoMarkdown to 0 and stacking every comment
-  // at the document start.
-  matched.sort((a, b) => a.start - b.start);
-  for (let i = 0; i < matched.length - 1; i++) {
-    matched[i].end = matched[i + 1].start;
-  }
-  if (matched.length > 0 && docLength !== undefined) {
-    const last = matched[matched.length - 1];
-    if (last.end > docLength) last.end = docLength;
-  }
-
-  return matched;
-}
+/**
+ * Compute section boundaries in a DOCX from its real heading paragraphs.
+ *
+ * Given the configured `sections.yaml` and the headings extracted via
+ * `extractHeadings()`, return one boundary per section file with text
+ * positions in the same coordinate system as `CommentAnchorData.docPosition`.
+ *
+ * Matching is by heading text (primary header + aliases, case-insensitive).
+ * This replaces the older keyword-search-in-body-text approach which would
+ * pick up section names that happen to appear inside prose ("results across
+ * countries") or in structured-abstract labels where paragraph boundaries
+ * are lost in concatenation.
+ */
+
+import type { DocxHeading } from '../word-extraction.js';
+import type { SectionConfig } from '../types.js';
+
+export interface SectionBoundary {
+  file: string;
+  start: number;
+  end: number;
+}
+
+export function computeSectionBoundaries(
+  sections: Record<string, SectionConfig>,
+  headings: DocxHeading[],
+  docLength?: number,
+): SectionBoundary[] {
+  const matched: SectionBoundary[] = [];
+
+  // Only consider top-level (Heading1-style) when level info is available;
+  // when level==0 (unparseable style), fall back to all headings.
+  const haveLevels = headings.some(h => h.level > 0);
+  const candidates = haveLevels ? headings.filter(h => h.level === 1) : headings;
+
+  for (const [file, cfg] of Object.entries(sections)) {
+    const targets = [cfg.header, ...(cfg.aliases || [])]
+      .filter(Boolean)
+      .map(s => s.toLowerCase().trim());
+
+    let firstMatch = -1;
+    for (const h of candidates) {
+      const text = h.text.toLowerCase().trim();
+      if (targets.includes(text)) {
+        firstMatch = h.docPosition;
+        break;
+      }
+    }
+
+    // Fallback: if no level-1 hit, allow any-level match (handles single-level docs)
+    if (firstMatch < 0 && haveLevels) {
+      for (const h of headings) {
+        const text = h.text.toLowerCase().trim();
+        if (targets.includes(text)) {
+          firstMatch = h.docPosition;
+          break;
+        }
+      }
+    }
+
+    if (firstMatch >= 0) {
+      matched.push({ file, start: firstMatch, end: Number.MAX_SAFE_INTEGER });
+    }
+  }
+
+  // Sort by start position and tighten each end to the next start.
+  // The last section's end is capped at docLength when known, otherwise
+  // left at MAX_SAFE_INTEGER. Without the cap, single-section configs
+  // produce a sectionLength of ~9e15, collapsing proportional-position
+  // math in insertCommentsIntoMarkdown to 0 and stacking every comment
+  // at the document start.
+  matched.sort((a, b) => a.start - b.start);
+  for (let i = 0; i < matched.length - 1; i++) {
+    matched[i].end = matched[i + 1].start;
+  }
+  if (matched.length > 0 && docLength !== undefined) {
+    const last = matched[matched.length - 1];
+    if (last.end > docLength) last.end = docLength;
+  }
+
+  return matched;
+}