change diff format

action.yml

@@ -161,30 +161,37 @@ inputs:
       $summary
       ```
 
-      Here is the content of file `$filename` -
+      Here is the content of file `$filename` with 
+      line numbers -
       ```
       $file_content
       ```
 
-      Format for changes and review comments (if any) 
-      on line ranges (line numbers in new file) -
-      <start_line-end_line>: 
-      ```diff
-      <diff_hunk>
+      Format for changes and review comments (if any) -
+
+      old_hunk:
+      ```
+      <old code with line numbers>
+      ```
+      new_hunk:
+      ```
+      <new code with line numbers>
       ```
       ```text
       <existing review comments>
       ```
       ---
-      <start_line-end_line>: 
-      ```diff
-      <diff_hunk>
+      old_hunk:
+      ```
+      <old code with line numbers>
+      ```
+      new_hunk:
+      ```
+      <new code with line numbers>
       ```
       ---
       ...
 
-      The <diff_hunk> is in the unidiff format.
-
       Changes for review -
       $patches
 
@@ -193,13 +200,16 @@ inputs:
       line ranges and review comments applicable for that line range. 
       Any other commentary outside of this format will be ignored 
       and will not be read by the parser -
-      <review_start_line-review_end_line>:
+
+      <start_line_number_new_code>-<end_line_number_new_code>:
       <your review>
       ---
       ...
 
-      Make sure <review_start_line> and <review_end_line> for each review 
-      must be within line ranges in the changes above. Don't echo back the 
+      It's important that <start_line_number_new_code> and 
+      <end_line_number_new_code> for each review must be within 
+      line ranges of new_hunks as you cannot make comments 
+      on line numbers of old_hunks. Don't echo back the 
       code provided to you as the line number range is sufficient to map 
       your comment to the relevant code section in GitHub. Your responses 
       will be recorded as multi-line review comments on the GitHub pull 

src/commenter.ts

@@ -253,8 +253,8 @@ ${COMMENT_REPLY_TAG}
     return comments.filter(
       (comment: any) =>
         comment.path === path &&
-        comment.start_line === start_line &&
-        comment.line === end_line &&
+        comment.start_line >= start_line &&
+        comment.line <= end_line &&
         comment.body !== ''
     )
   }

src/review.ts

@@ -120,14 +120,28 @@ export const codeReview = async (
       const patches: [number, number, string][] = []
       for (const patch of split_patch(file.patch)) {
         const patch_lines = patch_start_end_line(patch)
-        if (
-          !patch_lines ||
-          patch_lines.start_line === -1 ||
-          patch_lines.end_line === -1
-        ) {
+        if (!patch_lines) {
+          continue
+        }
+        const hunks = parse_hunk(patch)
+        if (!hunks) {
           continue
         }
-        patches.push([patch_lines.start_line, patch_lines.end_line, patch])
+        const hunks_str = `old_hunk:
+\`\`\`
+${hunks.old_hunk}
+\`\`\`
+---
+new_hunk:
+\`\`\`
+${hunks.new_hunk}
+\`\`\`
+`
+        patches.push([
+          patch_lines.new_hunk.start_line,
+          patch_lines.new_hunk.end_line,
+          hunks_str
+        ])
       }
       if (patches.length > 0) {
         return [file.filename, file_content, file_diff, patches]
@@ -319,7 +333,17 @@ ${
         if (
           file_content_tokens < options.heavy_token_limits.extra_content_tokens
         ) {
+          // rewrite file_content to preprend line numbers and colon before each line
+          const lines = file_content.split('\n')
+          let line_number = 1
+          file_content = ''
+          for (const line of lines) {
+            file_content += `${line_number}: ${line}
+`
+            line_number += 1
+          }
           ins.file_content = file_content
+          core.info(`file_content: ${file_content}`)
         } else {
           core.info(
             `skip sending content of file: ${ins.filename} due to token count: ${file_content_tokens}`
@@ -348,6 +372,17 @@ ${
           } else {
             comment_chain = ''
           }
+          // check comment_chain tokens and skip if too long
+          const comment_chain_tokens = tokenizer.get_token_count(comment_chain)
+          if (
+            comment_chain_tokens >
+            options.heavy_token_limits.extra_content_tokens
+          ) {
+            core.info(
+              `skip sending comment chain of file: ${ins.filename} due to token count: ${comment_chain_tokens}`
+            )
+            comment_chain = ''
+          }
         } catch (e: unknown) {
           if (e instanceof ChatGPTError) {
             core.warning(
@@ -356,10 +391,7 @@ ${
           }
         }
         ins.patches += `
-${start_line}-${end_line}:
-\`\`\`diff
 ${patch}
-\`\`\`
 `
         if (comment_chain !== '') {
           ins.patches += `
@@ -484,21 +516,69 @@ const split_patch = (patch: string | null | undefined): string[] => {
 
 const patch_start_end_line = (
   patch: string
-): {start_line: number; end_line: number} | null => {
+): {
+  old_hunk: {start_line: number; end_line: number}
+  new_hunk: {start_line: number; end_line: number}
+} | null => {
   const pattern = /(^@@ -(\d+),(\d+) \+(\d+),(\d+) @@)/gm
   const match = pattern.exec(patch)
   if (match) {
-    const begin = parseInt(match[4])
-    const diff = parseInt(match[5])
+    const old_begin = parseInt(match[2])
+    const old_diff = parseInt(match[3])
+    const new_begin = parseInt(match[4])
+    const new_diff = parseInt(match[5])
     return {
-      start_line: begin,
-      end_line: begin + diff - 1
+      old_hunk: {
+        start_line: old_begin,
+        end_line: old_begin + old_diff - 1
+      },
+      new_hunk: {
+        start_line: new_begin,
+        end_line: new_begin + new_diff - 1
+      }
     }
   } else {
     return null
   }
 }
 
+const parse_hunk = (
+  hunk: string
+): {old_hunk: string; new_hunk: string} | null => {
+  const hunkInfo = patch_start_end_line(hunk)
+  if (!hunkInfo) {
+    return null
+  }
+
+  const old_hunk_lines: string[] = []
+  const new_hunk_lines: string[] = []
+
+  let old_line = hunkInfo.old_hunk.start_line
+  let new_line = hunkInfo.new_hunk.start_line
+
+  const lines = hunk.split('\n').slice(1) // Skip the @@ line
+
+  lines.forEach(line => {
+    if (line.startsWith('-')) {
+      old_hunk_lines.push(`${old_line}: ${line.substring(1)}`)
+      old_line++
+    } else if (line.startsWith('+')) {
+      new_hunk_lines.push(`${new_line}: ${line.substring(1)}`)
+      new_line++
+    } else {
+      old_hunk_lines.push(`${old_line}: ${line}`)
+      new_hunk_lines.push(`${new_line}: ${line}`)
+      old_line++
+      new_line++
+    }
+  })
+
+  return {
+    old_hunk: old_hunk_lines.join('\n'),
+    new_hunk: new_hunk_lines.join('\n')
+  }
+}
+
 type Review = {
   start_line: number
   end_line: number