flappa-doormal 2.16.2 → 2.16.3

package/dist/index.mjs

@@ -1521,7 +1521,7 @@ const JOINER_PREFIX_LENGTHS = [
 	8,
 	6
 ];
-const STOP_CHARACTERS = /[\s\n.,;!?؛،۔]/;
+const STOP_CHARACTERS = /[\s\n.,;!?؛،۔۝۞]/;
 /**
 * Maximum allowed deviation between expected and actual boundary positions (characters).
 * Matches outside this range are rejected unless `ignoreDeviation` is active.
@@ -1833,6 +1833,46 @@ const estimateStartOffsetInCurrentPage = (remainingContent, currentFromIdx, page
 	}
 	return 0;
 };
+const estimateStartOffsetInCurrentPageFromEnd = (remainingContent, currentFromIdx, pageIds, normalizedPages) => {
+	const currentPageData = normalizedPages.get(pageIds[currentFromIdx]);
+	if (!currentPageData) return 0;
+	const remPrefix = remainingContent.slice(0, 500).trimStart();
+	if (!remPrefix) return 0;
+	const maxNeedleLen = Math.min(30, remPrefix.length);
+	for (let len = maxNeedleLen; len >= 5; len -= 5) {
+		const needle = remPrefix.slice(0, len);
+		const idx = currentPageData.content.lastIndexOf(needle);
+		if (idx >= 0) return idx;
+	}
+	if (remPrefix.length >= 3) {
+		const needle = remPrefix.slice(0, 3);
+		const idx = currentPageData.content.lastIndexOf(needle);
+		if (idx >= 0) return idx;
+	}
+	return 0;
+};
+const selectStartOffsetInCurrentPage = (segmentContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, logger) => {
+	const first = estimateStartOffsetInCurrentPage(segmentContent, fromIdx, pageIds, normalizedPages);
+	const last = estimateStartOffsetInCurrentPageFromEnd(segmentContent, fromIdx, pageIds, normalizedPages);
+	const candidates = [...new Set([first, last])];
+	if (candidates.length <= 1 || fromIdx + 1 > toIdx) return candidates[0] ?? 0;
+	const rawBoundary = cumulativeOffsets[fromIdx + 1] !== void 0 && cumulativeOffsets[fromIdx] !== void 0 ? Math.max(0, cumulativeOffsets[fromIdx + 1] - cumulativeOffsets[fromIdx]) : void 0;
+	if (rawBoundary === void 0) return candidates[0] ?? 0;
+	let best = candidates[0] ?? 0;
+	let bestScore = Number.POSITIVE_INFINITY;
+	for (const candidate of candidates) {
+		const expectedBoundary = Math.max(0, rawBoundary - candidate);
+		const pos = findPageStartNearExpectedBoundary(segmentContent, fromIdx + 1, expectedBoundary, pageIds, normalizedPages, logger);
+		if (pos > 0) {
+			const score = Math.abs(pos - expectedBoundary);
+			if (score < bestScore) {
+				bestScore = score;
+				best = candidate;
+			}
+		}
+	}
+	return best;
+};
 /**
 * Attempts to find the start position of a target page within remainingContent,
 * anchored near an expected boundary position to reduce collisions.
@@ -1898,6 +1938,22 @@ const selectBestAnchor = (candidates, expectedBoundary) => {
 		return Math.abs(curr.pos - expectedBoundary) + (curr.isNewline ? 0 : NON_NEWLINE_PENALTY) < bestScore ? curr : best;
 	});
 };
+/**
+* Finds the start position of a target page after a minimum position.
+* Used to avoid duplicate earlier matches when content repeats.
+*/
+const findPageStartAfterPosition = (remainingContent, targetPageIdx, minPos, pageIds, normalizedPages) => {
+	const targetPageData = normalizedPages.get(pageIds[targetPageIdx]);
+	if (!targetPageData) return -1;
+	const targetTrimmed = targetPageData.content.trimStart();
+	for (const len of WINDOW_PREFIX_LENGTHS) {
+		const prefix = targetTrimmed.slice(0, Math.min(len, targetTrimmed.length)).trim();
+		if (!prefix) continue;
+		const after = findAnchorCandidates(remainingContent, prefix, Math.max(0, minPos), remainingContent.length).filter((c) => c.pos > minPos);
+		if (after.length > 0) return selectBestAnchor(after, minPos).pos;
+	}
+	return -1;
+};
 const buildBoundaryPositionsFastPath = (segmentContent, fromIdx, toIdx, pageCount, cumulativeOffsets, logger) => {
 	const boundaryPositions = [0];
 	logger?.debug?.("[breakpoints] Using fast-path for large segment in buildBoundaryPositions", {
@@ -1956,7 +2012,7 @@ const buildBoundaryPositionsAccurate = (segmentContent, fromIdx, toIdx, pageCoun
 		pageCount,
 		toIdx
 	});
-	let startOffsetInFromPage = estimateStartOffsetInCurrentPage(segmentContent, fromIdx, pageIds, normalizedPages);
+	let startOffsetInFromPage = selectStartOffsetInCurrentPage(segmentContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, logger);
 	let didInferStartOffset = false;
 	for (let i = fromIdx + 1; i <= toIdx; i++) {
 		const rawBoundary = cumulativeOffsets[i] !== void 0 && cumulativeOffsets[fromIdx] !== void 0 ? Math.max(0, cumulativeOffsets[i] - cumulativeOffsets[fromIdx]) : void 0;
@@ -1964,7 +2020,12 @@ const buildBoundaryPositionsAccurate = (segmentContent, fromIdx, toIdx, pageCoun
 		startOffsetInFromPage = resolved.startOffsetInFromPage;
 		didInferStartOffset = didInferStartOffset || resolved.didInferStartOffset;
 		const prevBoundary = boundaryPositions[boundaryPositions.length - 1];
-		if (isBoundaryPositionValid(resolved.pos, prevBoundary, resolved.expectedBoundary, segmentContent.length)) boundaryPositions.push(resolved.pos);
+		let resolvedPos = resolved.pos;
+		if (resolvedPos <= prevBoundary) {
+			const afterPos = findPageStartAfterPosition(segmentContent, i, prevBoundary + 1, pageIds, normalizedPages);
+			if (afterPos > prevBoundary) resolvedPos = afterPos;
+		}
+		if (isBoundaryPositionValid(resolvedPos, prevBoundary, resolved.expectedBoundary, segmentContent.length)) boundaryPositions.push(resolvedPos);
 		else {
 			const estimate = Math.max(prevBoundary + 1, resolved.expectedBoundary);
 			boundaryPositions.push(Math.min(estimate, segmentContent.length));
@@ -2001,7 +2062,8 @@ const buildBoundaryPositionsAccurate = (segmentContent, fromIdx, toIdx, pageCoun
 */
 const buildBoundaryPositions = (segmentContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, logger) => {
 	const pageCount = toIdx - fromIdx + 1;
-	if (pageCount >= FAST_PATH_THRESHOLD) return buildBoundaryPositionsFastPath(segmentContent, fromIdx, toIdx, pageCount, cumulativeOffsets, logger);
+	const expectedLength = (cumulativeOffsets[toIdx + 1] ?? 0) - (cumulativeOffsets[fromIdx] ?? 0);
+	if (pageCount >= FAST_PATH_THRESHOLD && segmentContent.length === expectedLength) return buildBoundaryPositionsFastPath(segmentContent, fromIdx, toIdx, pageCount, cumulativeOffsets, logger);
 	return buildBoundaryPositionsAccurate(segmentContent, fromIdx, toIdx, pageCount, pageIds, normalizedPages, cumulativeOffsets, logger);
 };
 /**