mirror of
https://github.com/openai/codex.git
synced 2026-11-21 22:50:37 +00:00
Use automatic reasoning summaries by default in the TUI (#62904)
## Why The TUI left the reasoning summary unset on turn submission, so OpenAI requests could use `concise` even when the user had chosen no summary setting. Embedded WebSocket prewarming and the first turn need to use the same summary setting to reuse the warmup response. ## What changed - Send the configured `model_reasoning_summary` with TUI turns, defaulting to `auto` for OpenAI providers. Use the same fallback when prewarming through the in-process app server. - Read reasoning summary settings and custom provider definitions from the app server's effective config while honoring explicit launch settings. When resuming a remote session, use its working directory for the config lookup unless a remote directory override is set. ## Testing Cover summary defaults and explicit settings in Responses requests, and verify that the first embedded WebSocket turn uses `auto` and reuses the warmup response. GitOrigin-RevId: 1a2edfa44abdd031691a0d421eba077b0796ebe1
This commit is contained in:
@@ -22,14 +22,40 @@ pub(crate) fn has_launch_setting(
|
||||
.is_some_and(|layer| {
|
||||
matches!(
|
||||
layer.name,
|
||||
ConfigLayerSource::User {
|
||||
profile: Some(_),
|
||||
..
|
||||
}
|
||||
ConfigLayerSource::SessionFlags
|
||||
| ConfigLayerSource::User {
|
||||
profile: Some(_),
|
||||
..
|
||||
}
|
||||
)
|
||||
})
|
||||
}
|
||||
|
||||
pub(super) fn overlay_reasoning_summary(
|
||||
config: &mut Config,
|
||||
defaults: &codex_app_server_protocol::Config,
|
||||
cli_kv_overrides: &[(String, TomlValue)],
|
||||
) {
|
||||
if !has_launch_setting(config, cli_kv_overrides, "model_reasoning_summary") {
|
||||
config.model_reasoning_summary = defaults.model_reasoning_summary;
|
||||
}
|
||||
if let Some(providers) = defaults
|
||||
.additional
|
||||
.get("model_providers")
|
||||
.and_then(serde_json::Value::as_object)
|
||||
{
|
||||
let built_in =
|
||||
codex_model_provider_info::built_in_model_providers(/*openai_base_url*/ None);
|
||||
for (id, definition) in providers {
|
||||
if !built_in.contains_key(id)
|
||||
&& let Ok(provider) = serde_json::from_value(definition.clone())
|
||||
{
|
||||
config.model_providers.insert(id.clone(), provider);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
pub(super) fn overlay_new_session_defaults(
|
||||
config: &mut Config,
|
||||
defaults: &codex_app_server_protocol::Config,
|
||||
@@ -42,6 +68,7 @@ pub(super) fn overlay_new_session_defaults(
|
||||
if !has_launch_setting(config, cli_kv_overrides, "model_reasoning_effort") {
|
||||
config.model_reasoning_effort = defaults.model_reasoning_effort.clone();
|
||||
}
|
||||
overlay_reasoning_summary(config, defaults, cli_kv_overrides);
|
||||
if harness_overrides.service_tier.is_none()
|
||||
&& !has_launch_setting(config, cli_kv_overrides, "service_tier")
|
||||
{
|
||||
|
||||
@@ -56,12 +56,15 @@ pub(super) fn startup_config_cwd<'a>(
|
||||
config: &'a Config,
|
||||
app_server: &'a AppServerSession,
|
||||
app_server_target: &AppServerTarget,
|
||||
resumed_cwd: Option<&'a Path>,
|
||||
) -> &'a Path {
|
||||
match app_server_target {
|
||||
AppServerTarget::Embedded | AppServerTarget::LocalDaemon { .. } => config.cwd.as_path(),
|
||||
AppServerTarget::Remote { .. } | AppServerTarget::Cloud => {
|
||||
app_server.remote_cwd_override().unwrap_or(Path::new("."))
|
||||
}
|
||||
AppServerTarget::Remote { .. } => app_server
|
||||
.remote_cwd_override()
|
||||
.or(resumed_cwd)
|
||||
.unwrap_or(Path::new(".")),
|
||||
AppServerTarget::Cloud => app_server.remote_cwd_override().unwrap_or(Path::new(".")),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -395,7 +398,11 @@ impl App {
|
||||
config.model_reasoning_effort = None;
|
||||
}
|
||||
}
|
||||
let cwd = startup_config_cwd(&config, &app_server, &app_server_target)
|
||||
let resumed_cwd = match &session_selection {
|
||||
SessionSelection::Resume(target) => target.cwd.as_deref(),
|
||||
_ => None,
|
||||
};
|
||||
let cwd = startup_config_cwd(&config, &app_server, &app_server_target, resumed_cwd)
|
||||
.display()
|
||||
.to_string();
|
||||
let startup_config = match startup_draft
|
||||
@@ -463,6 +470,13 @@ impl App {
|
||||
}
|
||||
},
|
||||
};
|
||||
if let Some(defaults) = startup_config.as_ref() {
|
||||
super::new_session::overlay_reasoning_summary(
|
||||
&mut config,
|
||||
&defaults.config,
|
||||
&cli_kv_overrides,
|
||||
);
|
||||
}
|
||||
launch_choices.apply_projectless_approval_policy(
|
||||
startup_defaults.approval_policy,
|
||||
bootstrap.allowed_approval_policies.as_deref(),
|
||||
|
||||
@@ -465,7 +465,7 @@ requires_openai_auth = false
|
||||
}
|
||||
let defaults = crate::config_update::read_effective_config_with_layers(
|
||||
&server.request_handle(),
|
||||
startup_config_cwd(&config, &server, &target)
|
||||
startup_config_cwd(&config, &server, &target, /*resumed_cwd*/ None)
|
||||
.display()
|
||||
.to_string(),
|
||||
)
|
||||
@@ -636,7 +636,7 @@ async fn fresh_startup_reads_destination_and_cleared_model_uses_catalog() -> Res
|
||||
)?;
|
||||
std::fs::write(
|
||||
server_home.path().join("config.toml"),
|
||||
"model_reasoning_effort = \"high\"\nsandbox_mode = \"read-only\"\nmodel_reasoning_summary = \"concise\"\n",
|
||||
"model_reasoning_effort = \"high\"\nsandbox_mode = \"read-only\"\nmodel_reasoning_summary = \"concise\"\n[model_providers.server-only]\nname = \"OpenAI\"\nwire_api = \"responses\"\n",
|
||||
)?;
|
||||
let mut config = ConfigBuilder::default()
|
||||
.codex_home(client_home.path().to_path_buf())
|
||||
@@ -680,6 +680,16 @@ async fn fresh_startup_reads_destination_and_cleared_model_uses_catalog() -> Res
|
||||
} else {
|
||||
AppServerTarget::Embedded
|
||||
};
|
||||
if remote {
|
||||
assert_eq!(
|
||||
startup_config_cwd(&config, &server, &target, Some(destination.path())),
|
||||
if override_cwd {
|
||||
launch_cwd.path()
|
||||
} else {
|
||||
destination.path()
|
||||
},
|
||||
);
|
||||
}
|
||||
let bootstrap = server.bootstrap(&config).await?;
|
||||
assert_eq!(
|
||||
bootstrap.default_model.as_deref(),
|
||||
@@ -687,7 +697,7 @@ async fn fresh_startup_reads_destination_and_cleared_model_uses_catalog() -> Res
|
||||
);
|
||||
let defaults = crate::config_update::read_effective_config_with_layers(
|
||||
&server.request_handle(),
|
||||
startup_config_cwd(&config, &server, &target)
|
||||
startup_config_cwd(&config, &server, &target, /*resumed_cwd*/ None)
|
||||
.display()
|
||||
.to_string(),
|
||||
)
|
||||
@@ -702,6 +712,11 @@ async fn fresh_startup_reads_destination_and_cleared_model_uses_catalog() -> Res
|
||||
&EnvironmentManager::default_for_tests(),
|
||||
);
|
||||
assert!(!defaults_read.prompt_windows_sandbox);
|
||||
assert_eq!(
|
||||
config.model_reasoning_summary,
|
||||
Some(codex_protocol::config_types::ReasoningSummary::Concise)
|
||||
);
|
||||
assert!(config.model_providers["server-only"].is_openai());
|
||||
assert_eq!(config.model, None);
|
||||
let selected_model =
|
||||
startup_model(&config, &bootstrap, /*server_defaults_read*/ true)?;
|
||||
|
||||
@@ -321,6 +321,11 @@ impl App {
|
||||
.await
|
||||
{
|
||||
Ok(Some(defaults)) => {
|
||||
super::new_session::overlay_reasoning_summary(
|
||||
&mut config,
|
||||
&defaults.config,
|
||||
&self.cli_kv_overrides,
|
||||
);
|
||||
crate::projectless::apply_defaults(
|
||||
&mut config,
|
||||
&self.harness_overrides,
|
||||
|
||||
@@ -1199,6 +1199,7 @@ impl AppServerSession {
|
||||
remote_cwd_override.or(self.remote_cwd_override.as_deref()),
|
||||
session_start_source,
|
||||
);
|
||||
apply_prewarm_reasoning_summary(&mut params, &session_config, &self.request_handle);
|
||||
if self.thread_params_mode.settings_owner() == ThreadSettingsOwner::Client {
|
||||
params.model_provider = self
|
||||
.model_provider_override
|
||||
@@ -2434,6 +2435,7 @@ pub(crate) async fn start_thread_with_request_handle(
|
||||
launch_choices.approvals_reviewer,
|
||||
&launch_choices.approval_policy,
|
||||
);
|
||||
apply_prewarm_reasoning_summary(&mut params, &config, &request_handle);
|
||||
if thread_params_mode.settings_owner() == ThreadSettingsOwner::Client {
|
||||
thread_tool_transport
|
||||
.check_namespace(&request_handle, &config.cwd)
|
||||
@@ -2850,6 +2852,34 @@ pub(crate) fn thread_start_params_from_config(
|
||||
)
|
||||
}
|
||||
|
||||
pub(crate) fn reasoning_summary_from_config(
|
||||
config: &Config,
|
||||
) -> Option<codex_protocol::config_types::ReasoningSummary> {
|
||||
config.model_reasoning_summary.or_else(|| {
|
||||
config
|
||||
.model_providers
|
||||
.get(&config.model_provider_id)
|
||||
.is_some_and(codex_model_provider_info::ModelProviderInfo::is_openai)
|
||||
.then_some(codex_protocol::config_types::ReasoningSummary::Auto)
|
||||
})
|
||||
}
|
||||
|
||||
fn apply_prewarm_reasoning_summary(
|
||||
params: &mut ThreadStartParams,
|
||||
config: &Config,
|
||||
request_handle: &AppServerRequestHandle,
|
||||
) {
|
||||
if matches!(request_handle, AppServerRequestHandle::InProcess(_))
|
||||
&& config.model_reasoning_summary.is_none()
|
||||
&& let Some(summary) = reasoning_summary_from_config(config)
|
||||
{
|
||||
params.config.get_or_insert_default().insert(
|
||||
"model_reasoning_summary".to_string(),
|
||||
serde_json::json!(summary),
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
fn thread_start_params_with_reviewer(
|
||||
config: &Config,
|
||||
thread_params_mode: ThreadParamsMode,
|
||||
@@ -2871,6 +2901,7 @@ fn thread_start_params_with_reviewer(
|
||||
)
|
||||
})
|
||||
.flatten();
|
||||
let overrides = config_request_overrides_from_config(config, thread_params_mode);
|
||||
ThreadStartParams {
|
||||
model: config.model.clone(),
|
||||
model_provider: provider_selection::explicit_provider(config),
|
||||
@@ -2892,7 +2923,7 @@ fn thread_start_params_with_reviewer(
|
||||
},
|
||||
sandbox,
|
||||
permissions,
|
||||
config: config_request_overrides_from_config(config, thread_params_mode),
|
||||
config: overrides,
|
||||
ephemeral: Some(config.ephemeral),
|
||||
history_mode: (!config.ephemeral).then_some(ThreadHistoryMode::Paginated),
|
||||
session_start_source,
|
||||
|
||||
@@ -3,46 +3,40 @@
|
||||
use super::*;
|
||||
use crate::legacy_core::config::ConfigBuilder;
|
||||
use codex_app_server_client::AppServerEvent;
|
||||
use codex_app_server_protocol::JSONRPCMessage;
|
||||
use codex_app_server_protocol::ServerNotification;
|
||||
use codex_protocol::config_types::ReasoningSummary;
|
||||
use core_test_support::responses;
|
||||
use futures::SinkExt;
|
||||
use futures::StreamExt;
|
||||
use pretty_assertions::assert_eq;
|
||||
use serde_json::json;
|
||||
use tokio_tungstenite::tungstenite::Message;
|
||||
|
||||
#[tokio::test]
|
||||
async fn reasoning_defaults_reach_responses() -> Result<()> {
|
||||
let concurrent = json!({"reasoning_summary_delivery": "sequential_cutoff"});
|
||||
for (settings, summary, stream_options, mode) in [
|
||||
(
|
||||
"",
|
||||
json!("concise"),
|
||||
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
|
||||
json!("auto"),
|
||||
concurrent.clone(),
|
||||
ThreadParamsMode::Embedded,
|
||||
),
|
||||
(
|
||||
"[features]\nconcurrent_reasoning_summaries = false",
|
||||
json!("concise"),
|
||||
json!("auto"),
|
||||
json!(null),
|
||||
ThreadParamsMode::Embedded,
|
||||
),
|
||||
(
|
||||
"[features]\nconcurrent_reasoning_summaries = true",
|
||||
json!("concise"),
|
||||
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
|
||||
ThreadParamsMode::Embedded,
|
||||
),
|
||||
(
|
||||
"model_reasoning_summary = 'detailed'",
|
||||
json!("detailed"),
|
||||
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
|
||||
concurrent.clone(),
|
||||
ThreadParamsMode::Embedded,
|
||||
),
|
||||
(
|
||||
"model_reasoning_summary = 'detailed'\n[features]\nconcurrent_reasoning_summaries = true",
|
||||
json!("detailed"),
|
||||
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
|
||||
ThreadParamsMode::Embedded,
|
||||
),
|
||||
(
|
||||
"model_reasoning_summary = 'none'\n[features]\nconcurrent_reasoning_summaries = true",
|
||||
"model_reasoning_summary = 'none'",
|
||||
json!(null),
|
||||
json!(null),
|
||||
ThreadParamsMode::Embedded,
|
||||
@@ -50,7 +44,7 @@ async fn reasoning_defaults_reach_responses() -> Result<()> {
|
||||
(
|
||||
"model_reasoning_summary = 'detailed'\nmodel_verbosity = 'high'\nweb_search = 'live'",
|
||||
json!("detailed"),
|
||||
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
|
||||
concurrent,
|
||||
ThreadParamsMode::Remote,
|
||||
),
|
||||
] {
|
||||
@@ -148,7 +142,11 @@ stream_max_retries = 0
|
||||
&config.workspace_roots,
|
||||
started.session.model.clone(),
|
||||
/*effort*/ None,
|
||||
/*summary*/ None,
|
||||
Some(
|
||||
server_config
|
||||
.model_reasoning_summary
|
||||
.unwrap_or(ReasoningSummary::Auto),
|
||||
),
|
||||
/*service_tier*/ None,
|
||||
/*collaboration_mode*/ None,
|
||||
/*output_schema*/ None,
|
||||
@@ -209,3 +207,133 @@ stream_max_retries = 0
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn reasoning_summary_default_preserves_websocket_prewarm() -> Result<()> {
|
||||
let websocket = responses::start_websocket_server(vec![vec![
|
||||
vec![
|
||||
responses::ev_response_created("warmup"),
|
||||
responses::ev_completed("warmup"),
|
||||
],
|
||||
vec![
|
||||
responses::ev_response_created("first"),
|
||||
responses::ev_completed("first"),
|
||||
],
|
||||
]])
|
||||
.await;
|
||||
let home = tempfile::tempdir()?;
|
||||
let base_url = websocket.uri();
|
||||
std::fs::write(
|
||||
home.path().join("config.toml"),
|
||||
format!(
|
||||
r#"
|
||||
model = "gpt-5.5"
|
||||
model_provider = "reasoning-test"
|
||||
[model_providers.reasoning-test]
|
||||
name = "OpenAI"
|
||||
base_url = "{base_url}/v1"
|
||||
wire_api = "responses"
|
||||
supports_websockets = true
|
||||
"#
|
||||
),
|
||||
)?;
|
||||
let config = ConfigBuilder::default()
|
||||
.codex_home(home.path().to_path_buf())
|
||||
.build()
|
||||
.await?;
|
||||
let template = thread_start_params_from_config(
|
||||
&config,
|
||||
ThreadParamsMode::Embedded,
|
||||
/*remote_cwd_override*/ None,
|
||||
/*session_start_source*/ None,
|
||||
);
|
||||
assert_eq!(
|
||||
template
|
||||
.config
|
||||
.as_ref()
|
||||
.and_then(|config| config.get("model_reasoning_summary")),
|
||||
None,
|
||||
);
|
||||
let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await?;
|
||||
let endpoint = crate::resolve_remote_addr(&format!("ws://{}", listener.local_addr()?))?;
|
||||
let daemon = tokio::spawn(async move {
|
||||
let (stream, _) = listener.accept().await?;
|
||||
let mut socket = tokio_tungstenite::accept_async(stream).await?;
|
||||
while let Some(Ok(Message::Text(text))) = socket.next().await {
|
||||
let JSONRPCMessage::Request(request) = serde_json::from_str(&text)? else {
|
||||
continue;
|
||||
};
|
||||
let mut reply = match request.method.as_str() {
|
||||
"initialize" => json!({"result": {"userAgent": "daemon-test/1.0"}}),
|
||||
"thread/start" => json!({"error": {"code": -32603, "message": "recorded"}}),
|
||||
method => panic!("unexpected request: {method}"),
|
||||
};
|
||||
reply["id"] = json!(request.id);
|
||||
socket.send(Message::Text(reply.to_string().into())).await?;
|
||||
if request.method == "thread/start" {
|
||||
return Ok::<_, color_eyre::Report>(request.params);
|
||||
}
|
||||
}
|
||||
color_eyre::eyre::bail!("missing thread/start");
|
||||
});
|
||||
let mut daemon_session = AppServerSession::new(
|
||||
crate::backend_id::BackendId::new(),
|
||||
crate::connect_remote_app_server(endpoint).await?,
|
||||
ThreadParamsMode::Embedded,
|
||||
);
|
||||
assert!(daemon_session.start_thread(&config).await.is_err());
|
||||
let daemon_params = daemon.await??.expect("thread/start params");
|
||||
assert_eq!(
|
||||
daemon_params["config"]["model_reasoning_summary"],
|
||||
json!(null)
|
||||
);
|
||||
daemon_session.shutdown().await?;
|
||||
let mut app_server = crate::start_embedded_app_server_for_picker(&config).await?;
|
||||
let started = app_server.start_thread(&config).await?;
|
||||
let warmup = tokio::time::timeout(
|
||||
std::time::Duration::from_secs(/*secs*/ 30),
|
||||
websocket.wait_for_request(/*connection_index*/ 0, /*request_index*/ 0),
|
||||
)
|
||||
.await?;
|
||||
assert_eq!(warmup.body_json()["reasoning"]["summary"], "auto");
|
||||
app_server
|
||||
.turn_start(
|
||||
started.session.thread_id,
|
||||
"user",
|
||||
"tui-user-message".to_string(),
|
||||
vec![UserInput::Text {
|
||||
text: "hello".to_string(),
|
||||
text_elements: Vec::new(),
|
||||
}],
|
||||
config.cwd.to_path_buf(),
|
||||
/*approval_policy*/ None,
|
||||
/*approvals_reviewer*/ None,
|
||||
TurnPermissionsOverride::Preserve,
|
||||
&config.workspace_roots,
|
||||
started.session.model,
|
||||
/*effort*/ None,
|
||||
reasoning_summary_from_config(&config),
|
||||
/*service_tier*/ None,
|
||||
/*collaboration_mode*/ None,
|
||||
/*output_schema*/ None,
|
||||
/*cyber_access_program*/ None,
|
||||
/*cloud_selection*/ None,
|
||||
)
|
||||
.await?;
|
||||
let first = tokio::time::timeout(
|
||||
std::time::Duration::from_secs(/*secs*/ 30),
|
||||
websocket.wait_for_request(/*connection_index*/ 0, /*request_index*/ 1),
|
||||
)
|
||||
.await?;
|
||||
let first = first.body_json();
|
||||
assert_eq!(
|
||||
(
|
||||
first["reasoning"]["summary"].clone(),
|
||||
first["previous_response_id"].clone()
|
||||
),
|
||||
(json!("auto"), json!("warmup")),
|
||||
);
|
||||
app_server.shutdown().await?;
|
||||
websocket.shutdown().await;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
@@ -517,7 +517,7 @@ impl ChatWidget {
|
||||
active_permission_profile,
|
||||
effective_mode.model().to_string(),
|
||||
effective_mode.reasoning_effort(),
|
||||
/*summary*/ None,
|
||||
crate::app_server_session::reasoning_summary_from_config(&self.config),
|
||||
service_tier,
|
||||
/*final_output_json_schema*/ None,
|
||||
collaboration_mode,
|
||||
|
||||
@@ -670,7 +670,10 @@ async fn submission_preserves_text_elements_and_local_images() {
|
||||
chat.handle_key_event(KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE));
|
||||
|
||||
let items = match next_submit_op(&mut op_rx) {
|
||||
Op::UserTurn { items, .. } => items,
|
||||
Op::UserTurn { items, summary, .. } => {
|
||||
assert_eq!(summary, None);
|
||||
items
|
||||
}
|
||||
other => panic!("expected Op::UserTurn, got {other:?}"),
|
||||
};
|
||||
assert_eq!(items.len(), 2);
|
||||
@@ -964,7 +967,7 @@ async fn enter_with_only_remote_images_submits_user_turn() {
|
||||
fork_parent_title: None,
|
||||
thread_name: None,
|
||||
model: "test-model".to_string(),
|
||||
model_provider_id: "test-provider".to_string(),
|
||||
model_provider_id: "openai".to_string(),
|
||||
service_tier: None,
|
||||
approval_policy: AskForApproval::Never,
|
||||
approvals_reviewer: ApprovalsReviewer::User,
|
||||
@@ -1001,7 +1004,10 @@ async fn enter_with_only_remote_images_submits_user_turn() {
|
||||
detail: None,
|
||||
}]
|
||||
);
|
||||
assert_eq!(summary, None);
|
||||
assert_eq!(
|
||||
summary,
|
||||
Some(codex_protocol::config_types::ReasoningSummary::Auto)
|
||||
);
|
||||
assert!(chat.remote_image_urls().is_empty());
|
||||
|
||||
let mut user_cell = None;
|
||||
|
||||
Reference in New Issue
Block a user