PR-Codex init

README.md

@@ -20,3 +20,5 @@ PR-Codex is currently free.
 ---
 
 Made by [dlabs](https://dlabs.app)
+
+<!-- wip -->

app/github/route.ts

@@ -1,4 +1,5 @@
 import { handleGithubAuth } from "@lib/handleGithubAuth"
+import { replyIssueComment } from "@lib/replyIssueComment"
 import { summarizePullRequest } from "@lib/summarizePullRequest"
 import { NextRequest, NextResponse } from "next/server"
 
@@ -9,9 +10,17 @@ export async function POST(req: NextRequest) {
 
   try {
     if (payload.action == "opened" || payload.action == "synchronize") {
+      // If a PR is opened or updated, summarize it
       const octokit = await handleGithubAuth(payload)
 
       await summarizePullRequest(payload, octokit)
+    } else if (payload.action == "created") {
+      if (payload.comment.body.includes("/ask-codex")) {
+        // If a comment is created, reply to it
+        const octokit = await handleGithubAuth(payload)
+
+        await replyIssueComment(payload, octokit)
+      }
     }
 
     return NextResponse.json("ok")

lib/joinStringsUntilMaxLength.ts

@@ -1,22 +1,24 @@
 export function joinStringsUntilMaxLength(
   parsedFiles: string[],
   maxLength: number
-): string {
-  let combinedString = ""
+) {
+  let codeDiff = ""
   let currentLength = 0
+  let maxLengthExceeded = false
 
   for (const file of parsedFiles) {
     const fileLength = file.length
 
     if (currentLength + fileLength <= maxLength) {
-      combinedString += file
+      codeDiff += file
       currentLength += fileLength
     } else {
+      maxLengthExceeded = true
       const remainingLength = maxLength - currentLength
-      combinedString += file.slice(0, remainingLength)
+      codeDiff += file.slice(0, remainingLength)
       break
     }
   }
 
-  return combinedString
+  return { codeDiff, maxLengthExceeded }
 }

lib/replyIssueComment.ts

@@ -0,0 +1,55 @@
+import { Octokit } from "@octokit/rest"
+import { ChatCompletionRequestMessage } from "openai-streams"
+import { generateChatGpt } from "../utils/generateChatGpt"
+import { getCodeDiff } from "../utils/getCodeDiff"
+
+export const startDescription = "\n\n<!-- start pr-codex -->"
+export const endDescription = "<!-- end pr-codex -->"
+const systemPrompt =
+  "You are a Git diff assistant. Given a code diff, you answer any question related to it. Be concise. Always wrap file names, functions, objects and similar in backticks (`)."
+
+export async function replyIssueComment(payload: any, octokit: Octokit) {
+  // Get relevant PR information
+  const { repository, issue, sender, comment } = payload
+
+  const question = comment.body.split("/ask-codex")[1].trim()
+
+  if (question) {
+    const { owner, repo, issue_number } = {
+      owner: repository.owner.login,
+      repo: repository.name,
+      issue_number: issue.number
+    }
+
+    // Get the diff content using Octokit and GitHub API
+    const { codeDiff } = await getCodeDiff(owner, repo, issue_number, octokit)
+
+    // If there are changes, trigger workflow
+    if (codeDiff?.length != 0) {
+      const messages: ChatCompletionRequestMessage[] = [
+        {
+          role: "system",
+          content: `${systemPrompt}\n\nHere is the code diff:\n\n${codeDiff}`
+        },
+        {
+          role: "user",
+          content: `${question}`
+        }
+      ]
+
+      const codexResponse = await generateChatGpt(messages)
+
+      const description = `> ${question}\n\n@${sender.login} ${codexResponse}`
+
+      await octokit.issues.createComment({
+        owner,
+        repo,
+        issue_number,
+        body: description
+      })
+
+      return codexResponse
+    }
+    throw new Error("No changes in PR")
+  }
+}

lib/summarizePullRequest.ts

@@ -1,139 +1,81 @@
 import { Octokit } from "@octokit/rest"
-import { ChatCompletionRequestMessage, OpenAI } from "openai-streams"
-import { yieldStream } from "yield-stream"
-import { parseDiff } from "../utils/parseDiff"
-import { joinStringsUntilMaxLength } from "./joinStringsUntilMaxLength"
+import { ChatCompletionRequestMessage } from "openai-streams"
+import { generateChatGpt } from "../utils/generateChatGpt"
+import { getCodeDiff } from "../utils/getCodeDiff"
 
-export const startDescription = "<!-- start pr-codex -->"
+export const startDescription = "\n\n<!-- start pr-codex -->"
 export const endDescription = "<!-- end pr-codex -->"
+const systemPrompt =
+  "You are a Git diff assistant. Given a code diff, you provide a clear and concise description of its content. Always wrap file names, functions, objects and similar in backticks (`)."
 
 export async function summarizePullRequest(payload: any, octokit: Octokit) {
   // Get relevant PR information
   const pr = payload.pull_request
-  const { owner, repo, number } = {
+  const { owner, repo, pull_number } = {
     owner: pr.base.repo.owner.login,
     repo: pr.base.repo.name,
-    number: pr.number
+    pull_number: pr.number
   }
 
   // Get the diff content using Octokit and GitHub API
-  const compareResponse = await octokit.rest.repos.compareCommits({
+  const { codeDiff, skippedFiles, maxLengthExceeded } = await getCodeDiff(
     owner,
     repo,
-    base: pr.base.sha,
-    head: pr.head.sha,
-    mediaType: {
-      format: "diff"
-    }
-  })
-  const diffContent = String(compareResponse.data)
-
-  // Parses the diff content and returns the parsed files.
-  // If the number of changes in a file is greater than 1k changes, the file will be skipped.
-  // The codeDiff is the joined string of parsed files, up to a max length of 10k.
-  const maxChanges = 1000
-  const { parsedFiles, skippedFiles } = parseDiff(diffContent, maxChanges)
-  const codeDiff = joinStringsUntilMaxLength(parsedFiles, 10000)
+    pull_number,
+    octokit
+  )
 
   // If there are changes, trigger workflow
   if (codeDiff?.length != 0) {
-    const systemPrompt = `You are a Git diff assistant. Always begin with "This PR". Given a code diff, you provide a simple description in prose, in less than 300 chars, which sums up the changes. Continue with "\n\n### Detailed summary\n" and make a comprehensive list of all changes, excluding any eventual skipped files. Be concise. Always wrap file names, functions, objects and similar in backticks (\`).${
-      skippedFiles.length != 0
-        ? ` After the list, conclude with "\n\n> " and mention that the following files were skipped due to too many changes: ${skippedFiles.join(
-            ","
-          )}.`
-        : ""
-    }`
-
     const messages: ChatCompletionRequestMessage[] = [
       {
         role: "system",
-        content: systemPrompt
+        content: `${systemPrompt}\n\nHere is the code diff:\n\n${codeDiff}`
       },
       {
         role: "user",
-        content: `Here is the code diff:\n\n${codeDiff}`
+        content:
+          'Starting with "This PR", clearly explain the focus of this PR in prose, in less than 300 characters. Then follow up with "\n\n### Detailed summary\n" and make a comprehensive list of all changes.'
       }
     ]
 
-    const summary = await generateChatGpt(messages)
+    const codexResponse = await generateChatGpt(messages)
 
     // Check if the PR already has a comment from the bot
     const hasCodexCommented =
       payload.action == "synchronize" &&
-      pr.body?.split("\n\n" + startDescription).length > 1
-
-    // if (firstComment) {
-    //   // Edit pinned bot comment to the PR
-    //   await octokit.issues.updateComment({
-    //     owner,
-    //     repo,
-    //     comment_id: firstComment.id,
-    //     body: summary
-    //   })
-    // } else {
-    //   // Add a comment to the PR
-    //   await octokit.issues.createComment({
-    //     owner,
-    //     repo,
-    //     issue_number: number,
-    //     body: summary
-    //   })
-    // }
+      pr.body?.split(startDescription).length > 1
 
-    const prCodexText = `\n\n${startDescription}\n\n---\n\n## PR-Codex overview\n${summary}\n\n${endDescription}`
+    const prCodexText = `${startDescription}\n\n${
+      (hasCodexCommented ? pr.body.split(startDescription)[0].trim() : pr.body)
+        ? "---\n\n"
+        : ""
+    }## PR-Codex overview\n${codexResponse}${
+      skippedFiles.length != 0
+        ? `\n\n> The following files were skipped due to too many changes: ${skippedFiles.join(
+            ", "
+          )}`
+        : ""
+    }${
+      maxLengthExceeded
+        ? "\n\n> The code diff exceeds the max number of characters, so this overview may be incomplete. Keep PRs small to avoid this issue."
+        : ""
+    }\n\n✨ Ask PR-Codex anything about this PR by commenting below with \`/ask-codex {your question}\`\n\n${endDescription}`
 
     const description = hasCodexCommented
-      ? pr.body.split("\n\n" + startDescription)[0] +
+      ? pr.body.split(startDescription)[0] +
         prCodexText +
         pr.body.split(endDescription)[1]
-      : pr.body + prCodexText
+      : (pr.body ?? "") + prCodexText
 
     await octokit.issues.update({
       owner,
       repo,
-      issue_number: number,
+      issue_number: pull_number,
       body: description
     })
 
-    return summary
+    return codexResponse
   }
   throw new Error("No changes in PR")
 }
-
-const generateChatGpt = async (messages: ChatCompletionRequestMessage[]) => {
-  const DECODER = new TextDecoder()
-  let text = ""
-
-  try {
-    const stream = await OpenAI(
-      "chat",
-      {
-        model: "gpt-3.5-turbo",
-        temperature: 0.7,
-        messages
-      },
-      { apiKey: process.env.OPENAI_API_KEY }
-    )
-
-    for await (const chunk of yieldStream(stream)) {
-      try {
-        const decoded: string = DECODER.decode(chunk)
-
-        if (decoded === undefined)
-          throw new Error(
-            "No choices in response. Decoded response: " +
-              JSON.stringify(decoded)
-          )
-
-        text += decoded
-      } catch (err) {
-        console.error(err)
-      }
-    }
-  } catch (err) {
-    console.error(err)
-  }
-
-  return text
-}

package.json

@@ -9,7 +9,8 @@
     "start": "next start",
     "lint": "next lint",
     "prettier": "prettier --write . --ignore-path .gitignore",
-    "summarize": "npx ts-node scripts/summarize"
+    "summarize": "npx ts-node scripts/summarize",
+    "reply": "npx ts-node scripts/reply"
   },
   "dependencies": {
     "@headlessui/react": "^1.7.14",

scripts/reply.ts

@@ -0,0 +1,33 @@
+import dotenv from "dotenv"
+import { handleGithubAuth } from "../lib/handleGithubAuth"
+import { replyIssueComment } from "../lib/replyIssueComment"
+import { testPayloadComment } from "../utils/github/testPayloadComment"
+
+dotenv.config()
+
+// Customize payload in `utils/testPayloadComment`
+
+async function main() {
+  try {
+    const octokit = await handleGithubAuth(testPayloadComment)
+
+    console.log("Generating comment...")
+
+    const comment = await replyIssueComment(testPayloadComment, octokit)
+
+    console.log(
+      "PR-Codex commented:\n\n",
+      comment,
+      "\n\nView on Github: https://github.com/decentralizedlabs/pr-codex/pull/4"
+    )
+  } catch (error) {
+    console.log(error)
+  }
+}
+
+main()
+  .then(() => process.exit(0))
+  .catch((error) => {
+    console.error(error)
+    process.exit(1)
+  })

scripts/summarize.ts

@@ -1,19 +1,19 @@
 import dotenv from "dotenv"
 import { handleGithubAuth } from "../lib/handleGithubAuth"
 import { summarizePullRequest } from "../lib/summarizePullRequest"
-import { testPayload } from "../utils/github/testPayload"
+import { testPayloadSyncPr } from "../utils/github/testPayloadSyncPr"
 
 dotenv.config()
 
-// Customize payload in `utils/testPayload`
+// Customize payload in `utils/testPayloadSyncPr`
 
 async function main() {
-  const octokit = await handleGithubAuth(testPayload)
-
   try {
+    const octokit = await handleGithubAuth(testPayloadSyncPr)
+
     console.log("Generating summary...")
 
-    const summary = await summarizePullRequest(testPayload, octokit)
+    const summary = await summarizePullRequest(testPayloadSyncPr, octokit)
 
     console.log(
       "PR-Codex wrote:\n\n",