fix: apply_diff finds closest match from start_line instead of buffer midpoint (#2657)

Author: MuriloFP

src/core/diff/strategies/__tests__/multi-search-replace.spec.ts

@@ -780,6 +780,104 @@ function five() {
 }`)
 				}
 			})
+			// Tests for Issue #2657 - Strange behavior with slightly wrong start_line
+			// This test directly reproduces the bug where apply_diff finds line 21 instead of line 33
+			// when start_line is 34 (one line off from the actual target at line 33)
+			it("should find line 33 (closer match) instead of line 21 when start_line is 34", async () => {
+				const strategy = new MultiSearchReplaceDiffStrategy(1.0, 40) // Default buffer of 40 lines
+
+				// Create content with "}" at lines 21 and 33
+				const lines: string[] = []
+				for (let i = 1; i <= 50; i++) {
+					if (i === 21 || i === 33) {
+						lines.push("}")
+					} else {
+						lines.push(`line ${i}`)
+					}
+				}
+				const originalContent = lines.join("\n")
+
+				// The diff with start_line: 34 (one line off from actual line 33)
+				const diffContent = `
+<<<<<<< SEARCH
+:start_line:34
+-------
+}
+=======
+} // modified
+>>>>>>> REPLACE`
+
+				const result = await strategy.applyDiff(originalContent, diffContent)
+
+				// The issue reports that it finds line 21 instead of line 33
+				// Line 33 is closer to start_line 34 than line 21
+				expect(result.success).toBe(true)
+
+				if (result.success) {
+					const resultLines = result.content.split("\n")
+
+					// Check which line was modified
+					const line21Modified = resultLines[20] === "} // modified"
+					const line33Modified = resultLines[32] === "} // modified"
+
+					// After the fix, line33Modified should be true and line21Modified should be false
+					expect(line21Modified).toBe(false) // Fix prevents this
+					expect(line33Modified).toBe(true) // Fix causes this
+				}
+			})
+
+			// This test verifies the fix works correctly by ensuring the closest match
+			// is found when multiple identical lines exist in the file
+			it("should find closest match when multiple identical lines exist", async () => {
+				const strategy = new MultiSearchReplaceDiffStrategy(1.0, 40)
+
+				// Create content with identical lines at positions 10, 25, 30, 45
+				const lines: string[] = []
+				for (let i = 1; i <= 50; i++) {
+					if ([10, 25, 30, 45].includes(i)) {
+						lines.push("duplicate line")
+					} else {
+						lines.push(`line ${i}`)
+					}
+				}
+				const originalContent = lines.join("\n")
+
+				// Test 1: start_line 28 should find line 30 (distance 2)
+				const diffContent1 = `
+<<<<<<< SEARCH
+:start_line:28
+-------
+duplicate line
+=======
+duplicate line // modified at 30
+>>>>>>> REPLACE`
+
+				const result1 = await strategy.applyDiff(originalContent, diffContent1)
+				expect(result1.success).toBe(true)
+				if (result1.success) {
+					const resultLines = result1.content.split("\n")
+					expect(resultLines[29]).toBe("duplicate line // modified at 30") // line 30 (0-indexed: 29)
+					expect(resultLines[24]).toBe("duplicate line") // line 25 unchanged
+				}
+
+				// Test 2: start_line 12 should find line 10 (distance 2)
+				const diffContent2 = `
+<<<<<<< SEARCH
+:start_line:12
+-------
+duplicate line
+=======
+duplicate line // modified at 10
+>>>>>>> REPLACE`
+
+				const result2 = await strategy.applyDiff(originalContent, diffContent2)
+				expect(result2.success).toBe(true)
+				if (result2.success) {
+					const resultLines = result2.content.split("\n")
+					expect(resultLines[9]).toBe("duplicate line // modified at 10") // line 10 (0-indexed: 9)
+					expect(resultLines[24]).toBe("duplicate line") // line 25 unchanged
+				}
+			})
 		})
 	})
 

src/core/diff/strategies/multi-file-search-replace.ts

@@ -33,17 +33,20 @@ function getSimilarity(original: string, search: string): number {
  * Performs a "middle-out" search of `lines` (between [startIndex, endIndex]) to find
  * the slice that is most similar to `searchChunk`. Returns the best score, index, and matched text.
  */
-function fuzzySearch(lines: string[], searchChunk: string, startIndex: number, endIndex: number) {
+function fuzzySearch(lines: string[], searchChunk: string, startIndex: number, endIndex: number, targetLine?: number) {
 	let bestScore = 0
 	let bestMatchIndex = -1
 	let bestMatchContent = ""
 
 	const searchLen = searchChunk.split(/\r?\n/).length
 
-	// Middle-out from the midpoint
-	const midPoint = Math.floor((startIndex + endIndex) / 2)
-	let leftIndex = midPoint
-	let rightIndex = midPoint + 1
+	// If targetLine is provided, start from there; otherwise use midpoint
+	const searchStart =
+		targetLine !== undefined
+			? Math.max(startIndex, Math.min(targetLine, endIndex - searchLen))
+			: Math.floor((startIndex + endIndex) / 2)
+	let leftIndex = searchStart
+	let rightIndex = searchStart + 1
 
 	while (leftIndex >= startIndex || rightIndex <= endIndex - searchLen) {
 		if (leftIndex >= startIndex) {
@@ -581,7 +584,13 @@ Each file requires its own path, start_line, and diff elements.
 					bestScore,
 					bestMatchIndex,
 					bestMatchContent: midContent,
-				} = fuzzySearch(resultLines, searchChunk, searchStartIndex, searchEndIndex)
+				} = fuzzySearch(
+					resultLines,
+					searchChunk,
+					searchStartIndex,
+					searchEndIndex,
+					startLine ? startLine - 1 : undefined,
+				)
 
 				matchIndex = bestMatchIndex
 				bestMatchScore = bestScore
@@ -601,7 +610,13 @@ Each file requires its own path, start_line, and diff elements.
 					bestScore,
 					bestMatchIndex,
 					bestMatchContent: aggContent,
-				} = fuzzySearch(resultLines, aggressiveSearchChunk, searchStartIndex, searchEndIndex)
+				} = fuzzySearch(
+					resultLines,
+					aggressiveSearchChunk,
+					searchStartIndex,
+					searchEndIndex,
+					startLine ? startLine - 1 : undefined,
+				)
 
 				if (bestMatchIndex !== -1 && bestScore >= this.fuzzyThreshold) {
 					matchIndex = bestMatchIndex

src/core/diff/strategies/multi-search-replace.ts

@@ -35,17 +35,21 @@ function getSimilarity(original: string, search: string): number {
 /**
  * Performs a "middle-out" search of `lines` (between [startIndex, endIndex]) to find
  * the slice that is most similar to `searchChunk`. Returns the best score, index, and matched text.
+ * If targetLine is provided, the search starts from that line instead of the midpoint.
  */
-function fuzzySearch(lines: string[], searchChunk: string, startIndex: number, endIndex: number) {
+function fuzzySearch(lines: string[], searchChunk: string, startIndex: number, endIndex: number, targetLine?: number) {
 	let bestScore = 0
 	let bestMatchIndex = -1
 	let bestMatchContent = ""
 	const searchLen = searchChunk.split(/\r?\n/).length
 
-	// Middle-out from the midpoint
-	const midPoint = Math.floor((startIndex + endIndex) / 2)
-	let leftIndex = midPoint
-	let rightIndex = midPoint + 1
+	// If targetLine is provided, start from there; otherwise use midpoint
+	const searchStart =
+		targetLine !== undefined
+			? Math.max(startIndex, Math.min(targetLine, endIndex - searchLen))
+			: Math.floor((startIndex + endIndex) / 2)
+	let leftIndex = searchStart
+	let rightIndex = searchStart + 1
 
 	while (leftIndex >= startIndex || rightIndex <= endIndex - searchLen) {
 		if (leftIndex >= startIndex) {
@@ -378,7 +382,7 @@ Only use a single line of '=======' between search and replacement content, beca
 
 		let matches = [
 			...diffContent.matchAll(
-				/(?:^|\n)(?<!\\)<<<<<<< SEARCH\s*\n((?:\:start_line:\s*(\d+)\s*\n))?((?:\:end_line:\s*(\d+)\s*\n))?((?<!\\)-------\s*\n)?([\s\S]*?)(?:\n)?(?:(?<=\n)(?<!\\)=======\s*\n)([\s\S]*?)(?:\n)?(?:(?<=\n)(?<!\\)>>>>>>> REPLACE)(?=\n|$)/g,
+				/(?:^|\n)(?<!\\)<<<<<<< SEARCH\s*\n((?:\:start_line:\s*(-?\d+)\s*\n))?((?:\:end_line:\s*(-?\d+)\s*\n))?((?<!\\)-------\s*\n)?([\s\S]*?)(?:\n)?(?:(?<=\n)(?<!\\)=======\s*\n)([\s\S]*?)(?:\n)?(?:(?<=\n)(?<!\\)>>>>>>> REPLACE)(?=\n|$)/g,
 			),
 		]
 
@@ -464,8 +468,11 @@ Only use a single line of '=======' between search and replacement content, beca
 
 			// Validate and handle line range if provided
 			if (startLine) {
+				// Clamp startLine to valid range (1 to resultLines.length)
+				const clampedStartLine = Math.max(1, Math.min(startLine, resultLines.length))
+
 				// Convert to 0-based index
-				const exactStartIndex = startLine - 1
+				const exactStartIndex = clampedStartLine - 1
 				const searchLen = searchLines.length
 				const exactEndIndex = exactStartIndex + searchLen - 1
 
@@ -478,8 +485,11 @@ Only use a single line of '=======' between search and replacement content, beca
 					bestMatchContent = originalChunk
 				} else {
 					// Set bounds for buffered search
-					searchStartIndex = Math.max(0, startLine - (this.bufferLines + 1))
-					searchEndIndex = Math.min(resultLines.length, startLine + searchLines.length + this.bufferLines)
+					searchStartIndex = Math.max(0, clampedStartLine - (this.bufferLines + 1))
+					searchEndIndex = Math.min(
+						resultLines.length,
+						clampedStartLine + searchLines.length + this.bufferLines,
+					)
 				}
 			}
 
@@ -489,7 +499,13 @@ Only use a single line of '=======' between search and replacement content, beca
 					bestScore,
 					bestMatchIndex,
 					bestMatchContent: midContent,
-				} = fuzzySearch(resultLines, searchChunk, searchStartIndex, searchEndIndex)
+				} = fuzzySearch(
+					resultLines,
+					searchChunk,
+					searchStartIndex,
+					searchEndIndex,
+					startLine ? Math.max(0, Math.min(startLine - 1, resultLines.length - 1)) : undefined,
+				)
 				matchIndex = bestMatchIndex
 				bestMatchScore = bestScore
 				bestMatchContent = midContent
@@ -509,7 +525,13 @@ Only use a single line of '=======' between search and replacement content, beca
 					bestScore,
 					bestMatchIndex,
 					bestMatchContent: aggContent,
-				} = fuzzySearch(resultLines, aggressiveSearchChunk, searchStartIndex, searchEndIndex)
+				} = fuzzySearch(
+					resultLines,
+					aggressiveSearchChunk,
+					searchStartIndex,
+					searchEndIndex,
+					startLine ? startLine - 1 : undefined,
+				)
 				if (bestMatchIndex !== -1 && bestScore >= this.fuzzyThreshold) {
 					matchIndex = bestMatchIndex
 					bestMatchScore = bestScore