Add API endpoint for automated commit message generation from git dif…

…fs (#443) * Add API endpoint for automated commit message generation from git diffs * revert * Add optional text prompt for commit message generation Enhance the commit message generation logic to accept an optional user-provided text prompt. This allows customization of the commit message context by using a template provided by the user. Ensure that the default behavior is maintained when no text is provided. Also, implement a helper function to remove markdown code fences from messages. The example script is updated to demonstrate usage with a text prompt. * Refactor chat handler constants and improve error messaging
smallcloudai · Dec 3, 2024 · 8a23874 · 8a23874
1 parent 0eadcc4
commit 8a23874
Show file tree

Hide file tree

Showing 3 changed files with 291 additions and 0 deletions.
diff --git a/examples/http_commit_message.sh b/examples/http_commit_message.sh
@@ -0,0 +1,12 @@
+curl http://127.0.0.1:8001/v1/commit-message-from-diff -k \
+  -H 'Content-Type: application/json' \
+  -d '{
+        "diff": "diff --git a/calculator.py b/calculator.py\nindex 1234567..89abcde 100644\n--- a/calculator.py\n+++ b/calculator.py\n@@ -1,6 +1,9 @@\n class Calculator:\n     def add(self, a, b):\n         return a + b\n\n+    def subtract(self, a, b):\n+        return a - b\n+\n import unittest\n\n class TestCalculator(unittest.TestCase):\n@@ -9,6 +12,9 @@ class TestCalculator(unittest.TestCase):\n     def test_add(self):\n         self.assertEqual(self.calc.add(2, 3), 5)\n\n+    def test_subtract(self):\n+        self.assertEqual(self.calc.subtract(5, 3), 2)\n+"
+      }'
+
+curl http://127.0.0.1:8001/v1/commit-message-from-diff -k \
+  -H 'Content-Type: application/json' \
+  -d '{
+        "diff": "diff --git a/calculator.py b/calculator.py\nindex 1234567..89abcde 100644\n--- a/calculator.py\n+++ b/calculator.py\n@@ -1,6 +1,9 @@\n class Calculator:\n     def add(self, a, b):\n         return a + b\n\n+    def subtract(self, a, b):\n+        return a - b\n+\n import unittest\n\n class TestCalculator(unittest.TestCase):\n@@ -9,6 +12,9 @@ class TestCalculator(unittest.TestCase):\n     def test_add(self):\n         self.assertEqual(self.calc.add(2, 3), 5)\n\n+    def test_subtract(self):\n+        self.assertEqual(self.calc.subtract(5, 3), 2)\n+",
+        "text": "[CI/CD] calculator features"
+      }'
diff --git a/src/http/routers/v1.rs b/src/http/routers/v1.rs
@@ -19,6 +19,7 @@ use crate::http::routers::v1::at_tools::{handle_v1_tools, handle_v1_tools_check_
 use crate::http::routers::v1::caps::handle_v1_caps;
 use crate::http::routers::v1::caps::handle_v1_ping;
 use crate::http::routers::v1::chat::{handle_v1_chat, handle_v1_chat_completions};
+use crate::http::routers::v1::chat_based_handlers::handle_v1_commit_message_from_diff;
 use crate::http::routers::v1::dashboard::get_dashboard_plots;
 use crate::http::routers::v1::docker::{handle_v1_docker_container_action, handle_v1_docker_container_list};
 use crate::http::routers::v1::graceful_shutdown::handle_v1_graceful_shutdown;
@@ -61,6 +62,7 @@ pub mod system_prompt;
 pub mod sync_files;
 mod gui_help_handlers;
 mod patch;
+pub mod chat_based_handlers;
 
 #[cfg(feature="vecdb")]
 pub mod handlers_memdb;
@@ -117,6 +119,7 @@ pub fn make_v1_router() -> Router {
         .route("/docker-container-action", telemetry_post!(handle_v1_docker_container_action))
 
         .route("/patch-single-file-from-ticket", telemetry_post!(handle_v1_patch_single_file_from_ticket))
+        .route("/commit-message-from-diff", telemetry_post!(handle_v1_commit_message_from_diff))
 
         .route("/subchat", telemetry_post!(handle_v1_subchat))
         .route("/subchat-single", telemetry_post!(handle_v1_subchat_single))

diff --git a/src/http/routers/v1/chat_based_handlers.rs b/src/http/routers/v1/chat_based_handlers.rs
@@ -0,0 +1,276 @@
+use std::sync::Arc;
+use tokio::sync::Mutex as AMutex;
+use axum::Extension;
+use axum::http::{Response, StatusCode};
+use hyper::Body;
+use serde::Deserialize;
+use tokio::sync::RwLock as ARwLock;
+use crate::subchat::subchat_single;
+use crate::at_commands::at_commands::AtCommandsContext;
+use crate::call_validation::{ChatContent, ChatMessage};
+use crate::custom_error::ScratchError;
+use crate::global_context::{try_load_caps_quickly_if_not_present, GlobalContext};
+
+
+const DIFF_ONLY_PROMPT: &str = r#"Analyze the given diff and generate a clear and descriptive commit message that explains the purpose of the changes. Your commit message should convey *why* the changes were made, *how* they improve the code, or what features or fixes are implemented, rather than just restating *what* the changes are. Aim for an informative, concise summary that would be easy for others to understand when reviewing the commit history.
+
+# Steps
+1. Analyze the code diff to understand the changes made.
+2. Determine the functionality added or removed, and the reason for these adjustments.
+3. Summarize the details of the change in an accurate and informative, yet concise way.
+4. Structure the message in a way that starts with a short summary line, followed by optional details if the change is complex.
+
+# Output Format
+
+The output should be a single commit message in the following format:
+- A **first line summarizing** the purpose of the change. This line should be concise.
+- Optionally, include a **second paragraph** with *additional context* if the change is complex or otherwise needs further clarification.
+  (e.g., if there's a bug fix, mention what problem was fixed and why the change works.)
+
+# Examples
+
+**Input (diff)**:
+```diff
+- public class UserManager { 
+-     private final UserDAO userDAO;
+  
++ public class UserManager { 
++     private final UserService userService;
++     private final NotificationService notificationService;
+
+  public UserManager(UserDAO userDAO) {
+-     this.userDAO = userDAO;
++     this.userService = new UserService();
++     this.notificationService = new NotificationService();
+  }
+```
+
+**Output (commit message)**:
+```
+Refactor `UserManager` to use `UserService` and `NotificationService`
+
+Replaced `UserDAO` with `UserService` and introduced `NotificationService` to improve separation of concerns and make user management logic reusable and extendable.
+```
+
+**Input (diff)**:
+```diff
+- if (age > 17) {
+-     accessAllowed = true;
+- } else {
+-     accessAllowed = false;
+- }
++ accessAllowed = age > 17;
+```
+
+**Output (commit message)**:
+```
+Simplify age check logic for accessing permissions by using a single expression
+```
+
+# Notes
+- Make sure the commit messages are descriptive enough to convey why the change is being made without being too verbose.
+- If applicable, add `Fixes #<issue-number>` or other references to link the commit to specific tickets.
+- Avoid wording: "Updated", "Modified", or "Changed" without explicitly stating *why*—focus on *intent*."#;
+
+
+
+const DIFF_WITH_USERS_TEXT_PROMPT: &str = r#"Generate a commit message using the diff and the provided initial commit message as a template for context.
+
+[Additional details as needed.]
+
+# Steps
+
+1. Analyze the code diff to understand the changes made.
+2. Review the user's initial commit message to understand the intent and use it as a contextual starting point.
+3. Determine the functionality added or removed, and the reason for these adjustments.
+4. Combine insights from the diff and user's initial commit message to generate a more descriptive and complete commit message.
+5. Summarize the details of the change in an accurate and informative, yet concise way.
+6. Structure the message in a way that starts with a short summary line, followed by optional details if the change is complex.
+
+# Output Format
+
+The output should be a single commit message in the following format:
+- A **first line summarizing** the purpose of the change. This line should be concise.
+- Optionally, include a **second paragraph** with *additional context* if the change is complex or otherwise needs further clarification.
+  (e.g., if there's a bug fix, mention what problem was fixed and why the change works.)
+
+# Examples
+
+**Input (initial commit message)**:
+```
+Refactor UserManager to use services instead of DAOs
+```
+
+**Input (diff)**:
+```diff
+- public class UserManager { 
+-     private final UserDAO userDAO;
+  
++ public class UserManager { 
++     private final UserService userService;
++     private final NotificationService notificationService;
+
+  public UserManager(UserDAO userDAO) {
+-     this.userDAO = userDAO;
++     this.userService = new UserService();
++     this.notificationService = new NotificationService();
+  }
+```
+
+**Output (commit message)**:
+```
+Refactor `UserManager` to use `UserService` and `NotificationService`
+
+Replaced `UserDAO` with `UserService` and introduced `NotificationService` to improve separation of concerns and make user management logic reusable and extendable.
+```
+
+**Input (initial commit message)**:
+```
+Simplify age check logic
+```
+
+**Input (diff)**:
+```diff
+- if (age > 17) {
+-     accessAllowed = true;
+- } else {
+-     accessAllowed = false;
+- }
++ accessAllowed = age > 17;
+```
+
+**Output (commit message)**:
+```
+Simplify age check logic for accessing permissions by using a single expression
+```
+
+# Notes
+- Make sure the commit messages are descriptive enough to convey why the change is being made without being too verbose.
+- If applicable, add `Fixes #<issue-number>` or other references to link the commit to specific tickets.
+- Avoid wording: "Updated", "Modified", or "Changed" without explicitly stating *why*—focus on *intent*."#;
+const N_CTX: usize = 32000;
+const TEMPERATURE: f32 = 0.5;
+
+
+fn remove_fencing(message: &String) -> String {
+    let trimmed_message = message.trim();
+    let without_leading_fence = if trimmed_message.starts_with("```") {
+        let mut lines = trimmed_message.lines();
+        lines.next();
+        lines.collect::<Vec<&str>>().join("\n")
+    } else {
+        trimmed_message.to_string()
+    };
+    let without_trailing_fence = if without_leading_fence.ends_with("```") {
+        let mut lines = without_leading_fence.lines().collect::<Vec<&str>>();
+        lines.pop();
+        lines.join("\n")
+    } else {
+        without_leading_fence
+    };
+    without_trailing_fence.trim().to_string()
+}
+
+
+#[derive(Deserialize)]
+struct CommitMessageFromDiffPost {
+    diff: String,
+    #[serde(default)]
+    text: Option<String>  // a prompt for the commit message 
+}
+
+pub async fn handle_v1_commit_message_from_diff(
+    Extension(global_context): Extension<Arc<ARwLock<GlobalContext>>>,
+    body_bytes: hyper::body::Bytes,
+) -> axum::response::Result<Response<Body>, ScratchError> {
+    let post = serde_json::from_slice::<CommitMessageFromDiffPost>(&body_bytes)
+        .map_err(|e| ScratchError::new(StatusCode::UNPROCESSABLE_ENTITY, format!("JSON problem: {}", e)))?;
+    if post.diff.is_empty() {
+        return Err(ScratchError::new(StatusCode::UNPROCESSABLE_ENTITY, "The provided diff is empty".to_string()))
+    }
+
+    let messages = if let Some(text) = &post.text {
+        vec![
+            ChatMessage {
+                role: "system".to_string(),
+                content: ChatContent::SimpleText(DIFF_WITH_USERS_TEXT_PROMPT.to_string()),
+                ..Default::default()
+            },
+            ChatMessage {
+                role: "user".to_string(),
+                content: ChatContent::SimpleText(format!("Commit message:\n```\n{}\n```\nDiff:\n```\n{}\n```\n", text, post.diff)),
+                ..Default::default()
+            },
+        ]
+    } else {
+        vec![
+            ChatMessage {
+                role: "system".to_string(),
+                content: ChatContent::SimpleText(DIFF_ONLY_PROMPT.to_string()),
+                ..Default::default()
+            },
+            ChatMessage {
+                role: "user".to_string(),
+                content: ChatContent::SimpleText(format!("Diff:\n```\n{}\n```\n", post.diff)),
+                ..Default::default()
+            },
+        ]
+    };
+    let model_name = match try_load_caps_quickly_if_not_present(global_context.clone(), 0).await {
+        Ok(caps) => {
+            caps.read()
+                .map(|x| Ok(x.code_chat_default_model.clone()))
+                .map_err(|_|
+                    ScratchError::new(StatusCode::INTERNAL_SERVER_ERROR, "Caps are not available".to_string())
+                )?
+        },
+        Err(_) => Err(ScratchError::new(StatusCode::INTERNAL_SERVER_ERROR, "No caps available".to_string()))
+    }?;
+
+    let ccx: Arc<AMutex<AtCommandsContext>> = Arc::new(AMutex::new(
+        AtCommandsContext::new(
+            global_context.clone(),
+            N_CTX, 
+            1, 
+            false, 
+            messages.clone(), 
+            "".to_string(), 
+            false
+        ).await)
+    );
+
+    let new_messages = subchat_single(
+        ccx.clone(),
+        model_name.as_str(),
+        messages,
+        vec![],
+        None,
+        false,
+        Some(TEMPERATURE),
+        None,
+        1,
+        None,
+        None,
+        None,
+    ).await.map_err(|e| ScratchError::new(StatusCode::INTERNAL_SERVER_ERROR, format!("Error: {}", e)))?;
+
+    let commit_message = new_messages
+        .into_iter()
+        .next()
+        .map(|x| x.into_iter().last().map(|last_m| {
+            match last_m.content {
+                ChatContent::SimpleText(text) => Some(text),
+                ChatContent::Multimodal(_) => { None }
+            }
+        }))
+        .flatten()
+        .flatten()
+        .ok_or(ScratchError::new(StatusCode::INTERNAL_SERVER_ERROR, "No commit message was generated".to_string()))?;
+    Ok(
+        Response::builder()
+            .status(StatusCode::OK)
+            .header("Content-Type", "application/json")
+            .body(Body::from(remove_fencing(&commit_message)))
+            .unwrap()
+    )
+}