feat: apply_patch tool for openai models (#9127)

packages/opencode/src/patch/index.ts

@@ -177,8 +177,18 @@ export namespace Patch {
     return { content, nextIdx: i }
   }
 
+  function stripHeredoc(input: string): string {
+    // Match heredoc patterns like: cat <<'EOF'\n...\nEOF or <<EOF\n...\nEOF
+    const heredocMatch = input.match(/^(?:cat\s+)?<<['"]?(\w+)['"]?\s*\n([\s\S]*?)\n\1\s*$/)
+    if (heredocMatch) {
+      return heredocMatch[2]
+    }
+    return input
+  }
+
   export function parsePatch(patchText: string): { hunks: Hunk[] } {
-    const lines = patchText.split("\n")
+    const cleaned = stripHeredoc(patchText.trim())
+    const lines = cleaned.split("\n")
     const hunks: Hunk[] = []
     let i = 0
 
@@ -363,7 +373,7 @@ export namespace Patch {
       // Try to match old lines in the file
       let pattern = chunk.old_lines
       let newSlice = chunk.new_lines
-      let found = seekSequence(originalLines, pattern, lineIndex)
+      let found = seekSequence(originalLines, pattern, lineIndex, chunk.is_end_of_file)
 
       // Retry without trailing empty line if not found
       if (found === -1 && pattern.length > 0 && pattern[pattern.length - 1] === "") {
@@ -371,7 +381,7 @@ export namespace Patch {
         if (newSlice.length > 0 && newSlice[newSlice.length - 1] === "") {
           newSlice = newSlice.slice(0, -1)
         }
-        found = seekSequence(originalLines, pattern, lineIndex)
+        found = seekSequence(originalLines, pattern, lineIndex, chunk.is_end_of_file)
       }
 
       if (found !== -1) {
@@ -407,28 +417,75 @@ export namespace Patch {
     return result
   }
 
-  function seekSequence(lines: string[], pattern: string[], startIndex: number): number {
-    if (pattern.length === 0) return -1
+  // Normalize Unicode punctuation to ASCII equivalents (like Rust's normalize_unicode)
+  function normalizeUnicode(str: string): string {
+    return str
+      .replace(/[\u2018\u2019\u201A\u201B]/g, "'") // single quotes
+      .replace(/[\u201C\u201D\u201E\u201F]/g, '"') // double quotes
+      .replace(/[\u2010\u2011\u2012\u2013\u2014\u2015]/g, "-") // dashes
+      .replace(/\u2026/g, "...") // ellipsis
+      .replace(/\u00A0/g, " ") // non-breaking space
+  }
 
-    // Simple substring search implementation
+  type Comparator = (a: string, b: string) => boolean
+
+  function tryMatch(lines: string[], pattern: string[], startIndex: number, compare: Comparator, eof: boolean): number {
+    // If EOF anchor, try matching from end of file first
+    if (eof) {
+      const fromEnd = lines.length - pattern.length
+      if (fromEnd >= startIndex) {
+        let matches = true
+        for (let j = 0; j < pattern.length; j++) {
+          if (!compare(lines[fromEnd + j], pattern[j])) {
+            matches = false
+            break
+          }
+        }
+        if (matches) return fromEnd
+      }
+    }
+
+    // Forward search from startIndex
     for (let i = startIndex; i <= lines.length - pattern.length; i++) {
       let matches = true
-
       for (let j = 0; j < pattern.length; j++) {
-        if (lines[i + j] !== pattern[j]) {
+        if (!compare(lines[i + j], pattern[j])) {
           matches = false
           break
         }
       }
-
-      if (matches) {
-        return i
-      }
+      if (matches) return i
     }
 
     return -1
   }
 
+  function seekSequence(lines: string[], pattern: string[], startIndex: number, eof = false): number {
+    if (pattern.length === 0) return -1
+
+    // Pass 1: exact match
+    const exact = tryMatch(lines, pattern, startIndex, (a, b) => a === b, eof)
+    if (exact !== -1) return exact
+
+    // Pass 2: rstrip (trim trailing whitespace)
+    const rstrip = tryMatch(lines, pattern, startIndex, (a, b) => a.trimEnd() === b.trimEnd(), eof)
+    if (rstrip !== -1) return rstrip
+
+    // Pass 3: trim (both ends)
+    const trim = tryMatch(lines, pattern, startIndex, (a, b) => a.trim() === b.trim(), eof)
+    if (trim !== -1) return trim
+
+    // Pass 4: normalized (Unicode punctuation to ASCII)
+    const normalized = tryMatch(
+      lines,
+      pattern,
+      startIndex,
+      (a, b) => normalizeUnicode(a.trim()) === normalizeUnicode(b.trim()),
+      eof,
+    )
+    return normalized
+  }
+
   function generateUnifiedDiff(oldContent: string, newContent: string): string {
     const oldLines = oldContent.split("\n")
     const newLines = newContent.split("\n")