@createiq/htmldiff 1.2.0-beta.7 → 1.2.0-beta.9

package/dist/HtmlDiff.mjs

@@ -1333,13 +1333,31 @@ function collectInsertionsKeyedByEnd(d) {
 	return out;
 }
 /**
-* Emit any insertions at boundary `b`. When both authors inserted at
-* the same boundary AND the inserted token sequences are textually
-* identical, the insertion is treated as agreed and emitted unmarked.
-* Otherwise each side's insertion is emitted with author attribution.
+* Emit any insertions at boundary `b`. Three cases:
 *
-* The CP-then-Me ordering for disagreement is arbitrary but consistent;
-* callers don't depend on it.
+*   1. One side inserted, the other didn't → emit that side's tokens
+*      with author attribution.
+*   2. Both sides inserted the EXACT same sequence → settled, emit
+*      unmarked.
+*   3. Both sides inserted overlapping but different sequences (the
+*      common case: one author accepted the other's insertion and
+*      edited it, so e.g. cp's "X Y Z" overlaps me's "X Y a Z" with
+*      "a" being a one-author-only addition). Run an LCS sub-diff
+*      between the two insertion sequences and emit:
+*        - tokens in BOTH → settled (equal segment)
+*        - tokens only in cp → ins-cp
+*        - tokens only in me → ins-me
+*      The order of emission preserves the natural reading flow of
+*      the merged insertion — common tokens read where they appear,
+*      with author-only deltas inserted in their LCS-determined
+*      positions.
+*
+* Without this sub-alignment, real-world flows like "Me added 'add
+* more things here', CP accepted minus 'things'" would render as two
+* full redundant insertions (`<ins cp>add more here</ins><ins me>add
+* more things here</ins>`) rather than the obvious single shared
+* insertion with a me-only "things" word — confusing to read and a
+* regression vs Word's track-changes UX.
 */
 function emitBoundary(b, cpInsAt, meInsAt, _cpDiffWords, _meDiffWords, segments) {
 	const cpIns = cpInsAt.get(b);
@@ -1347,18 +1365,34 @@ function emitBoundary(b, cpInsAt, meInsAt, _cpDiffWords, _meDiffWords, segments)
 	const hasCp = !!cpIns && cpIns.length > 0;
 	const hasMe = !!meIns && meIns.length > 0;
 	if (!hasCp && !hasMe) return;
-	if (hasCp && hasMe && tokenArraysEqual(cpIns, meIns)) {
+	if (!hasCp) {
+		appendSegment(segments, {
+			kind: "ins",
+			author: "me"
+		}, meIns);
+		return;
+	}
+	if (!hasMe) {
+		appendSegment(segments, {
+			kind: "ins",
+			author: "cp"
+		}, cpIns);
+		return;
+	}
+	if (tokenArraysEqual(cpIns, meIns)) {
 		appendSegment(segments, { kind: "equal" }, cpIns);
 		return;
 	}
-	if (hasCp) appendSegment(segments, {
+	const alignment = lcsAlign(cpIns, meIns);
+	for (const a of alignment) if (a.oldIdx !== null && a.newIdx !== null) appendSegment(segments, { kind: "equal" }, [cpIns[a.oldIdx]]);
+	else if (a.oldIdx !== null) appendSegment(segments, {
 		kind: "ins",
 		author: "cp"
-	}, cpIns);
-	if (hasMe) appendSegment(segments, {
+	}, [cpIns[a.oldIdx]]);
+	else if (a.newIdx !== null) appendSegment(segments, {
 		kind: "ins",
 		author: "me"
-	}, meIns);
+	}, [meIns[a.newIdx]]);
 }
 function tokenArraysEqual(a, b) {
 	if (a.length !== b.length) return false;
@@ -2751,6 +2785,17 @@ var HtmlDiff = class HtmlDiff {
 				curr = next;
 				continue;
 			}
+			let allTags = true;
+			for (let i = curr.startInNew; i < curr.endInNew; i++) if (!Utils_default.isTag(wordsForDiffNew[i])) {
+				allTags = false;
+				break;
+			}
+			if (allTags) {
+				yield curr;
+				prev = curr;
+				curr = next;
+				continue;
+			}
 			let oldDistanceInChars = 0;
 			for (let i = prev.endInOld; i < next.startInOld; i++) oldDistanceInChars += wordsForDiffOld[i].length;
 			let newDistanceInChars = 0;