Compare commits

..

2 Commits

Author SHA1 Message Date
David Zbarsky
33ecbc2e9a Try more things 2026-01-09 21:53:01 -05:00
Michael Bolin
55019a06e4 fix: support remote arm64 builds, as well 2026-01-09 17:30:10 -08:00
36 changed files with 789 additions and 2029 deletions

View File

@@ -39,7 +39,10 @@ common --grpc_keepalive_time=30s
# memory in exchange for higher download concurrency.
common --jobs=30
# These configs are split so linux CI can configure a custom exec platform.
common:remote --extra_execution_platforms=//:rbe
common:remote --remote_executor=grpcs://remote.buildbuddy.io
common:remote --jobs=800
common:remote --config=remote-base
common:remote-base --remote_executor=grpcs://remote.buildbuddy.io
common:remote-base --jobs=800

View File

@@ -97,14 +97,24 @@ jobs:
# Use a very short path to reduce argv/path length issues.
"BAZEL_STARTUP_ARGS=--output_user_root=C:\" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append
- name: Configure Bazel startup args (Linux)
if: runner.os == 'Linux'
shell: bash
run: |
echo "BAZEL_STARTUP_ARGS=--bazelrc=.github/workflows/linux.bazelrc" >> "$GITHUB_ENV"
- name: bazel test //...
env:
BUILDBUDDY_API_KEY: ${{ secrets.BUILDBUDDY_API_KEY }}
shell: bash
run: |
target="${{ matrix.target }}"
host_arch="${target%%-*}" # e.g. aarch64 / x86_64
bazel $BAZEL_STARTUP_ARGS --bazelrc=.github/workflows/ci.bazelrc test //... \
--config="$host_arch" \
--build_metadata=REPO_URL=https://github.com/openai/codex.git \
--build_metadata=COMMIT_SHA=$(git rev-parse HEAD) \
--build_metadata=ROLE=CI \
--build_metadata=VISIBILITY=PUBLIC \
"--remote_header=x-buildbuddy-api-key=$BUILDBUDDY_API_KEY"
"--remote_header=x-buildbuddy-api-key=$BUILDBUDDY_API_KEY"

View File

@@ -1,15 +1,18 @@
common --remote_download_minimal
common --nobuild_runfile_links
common --keep_going
# These config settings are used to route linux RBE actions, but the linux bazelrc is included conditionally.
# This ensures that the configs are defined on non-linux as well.
common:aarch64 --keep_going
common:x86_64 --keep_going
# We prefer to run the build actions entirely remotely so we can dial up the concurrency.
# We have platform-specific tests, so we want to execute the tests on all platforms using the strongest sandboxing available on each platform.
# On linux, we can do a full remote build/test, by targeting the right (x86/arm) runners, so we have coverage of both.
# Linux crossbuilds don't work until we untangle the libc constraint mess.
common:linux --config=remote
common:linux --config=remote-base
common:linux --strategy=remote
common:linux --platforms=//:rbe
# On mac, we can run all the build actions remotely but test actions locally.
common:macos --config=remote

4
.github/workflows/linux.bazelrc vendored Normal file
View File

@@ -0,0 +1,4 @@
common:aarch64 --extra_execution_platforms=//:rbe_arm64
common:aarch64 --platforms=//:rbe_arm64
common:x86_64 --extra_execution_platforms=//:rbe
common:x86_64 --platforms=//:rbe

View File

@@ -11,9 +11,40 @@ platform(
],
)
alias(
platform(
name = "rbe",
actual = "@rbe_platform",
constraint_values = [
"@platforms//cpu:x86_64",
"@platforms//os:linux",
"@bazel_tools//tools/cpp:clang",
"@toolchains_llvm_bootstrapped//constraints/libc:gnu.2.28",
],
exec_properties = {
# Ubuntu-based image that includes git, python3, dotslash, and other
# tools that various integration tests need.
# Verify at https://hub.docker.com/layers/mbolin491/codex-bazel/latest/images/sha256:8c9ff94187ea7c08a31e9a81f5fe8046ea3972a6768983c955c4079fa30567fb
"container-image": "docker://docker.io/mbolin491/codex-bazel@sha256:8c9ff94187ea7c08a31e9a81f5fe8046ea3972a6768983c955c4079fa30567fb",
"Arch": "amd64",
"OSFamily": "Linux",
},
)
platform(
name = "rbe_arm64",
constraint_values = [
"@platforms//cpu:aarch64",
"@platforms//os:linux",
"@bazel_tools//tools/cpp:clang",
"@toolchains_llvm_bootstrapped//constraints/libc:gnu.2.28",
],
exec_properties = {
# Ubuntu-based image that includes git, python3, dotslash, and other
# tools that various integration tests need.
# Verify at https://hub.docker.com/layers/mbolin491/codex-bazel/latest/images/sha256:ad9506086215fccfc66ed8d2be87847324be56790ae6a1964c241c28b77ef141
"container-image": "docker://docker.io/mbolin491/codex-bazel@sha256:ad9506086215fccfc66ed8d2be87847324be56790ae6a1964c241c28b77ef141",
"Arch": "arm64",
"OSFamily": "Linux",
},
)
exports_files(["AGENTS.md"])

View File

@@ -120,9 +120,3 @@ crate.annotation(
deps = [":windows_import_lib"],
)
use_repo(crate, "crates")
rbe_platform_repository = use_repo_rule("//:rbe.bzl", "rbe_platform_repository")
rbe_platform_repository(
name = "rbe_platform",
)

View File

@@ -1,17 +0,0 @@
# Example announcement tips for Codex TUI.
# Each [[announcements]] entry is evaluated in order; the last matching one is shown.
# Dates are UTC, formatted as YYYY-MM-DD. The from_date is inclusive and the to_date is exclusive.
# version_regex matches against the CLI version (env!("CARGO_PKG_VERSION")); omit to apply to all versions.
# target_app specify which app should display the announcement (cli, vsce, ...).
[[announcements]]
content = "Welcome to Codex! Check out the new onboarding flow."
from_date = "2024-10-01"
to_date = "2024-10-15"
target_app = "cli"
# Test announcement only for local build version until 2026-01-10 excluded (past)
[[announcements]]
content = "This is a test announcement"
version_regex = "^0\\.0\\.0$"
to_date = "2026-01-10"

View File

@@ -9,7 +9,6 @@ use codex_protocol::protocol::Op;
use codex_protocol::user_input::UserInput;
use std::sync::Arc;
use std::sync::Weak;
use tokio::sync::watch;
/// Control-plane handle for multi-agent operations.
/// `AgentControl` is held by each session (via `SessionServices`). It provides capability to
@@ -28,6 +27,7 @@ impl AgentControl {
Self { manager }
}
#[allow(dead_code)] // Used by upcoming multi-agent tooling.
/// Spawn a new agent thread and submit the initial prompt.
///
/// If `headless` is true, a background drain task is spawned to prevent unbounded event growth
@@ -50,6 +50,7 @@ impl AgentControl {
Ok(new_thread.thread_id)
}
#[allow(dead_code)] // Used by upcoming multi-agent tooling.
/// Send a `user` prompt to an existing agent thread.
pub(crate) async fn send_prompt(
&self,
@@ -68,13 +69,7 @@ impl AgentControl {
.await
}
/// Submit a shutdown request to an existing agent thread.
pub(crate) async fn shutdown_agent(&self, agent_id: ThreadId) -> CodexResult<String> {
let state = self.upgrade()?;
state.send_op(agent_id, Op::Shutdown {}).await
}
#[allow(dead_code)] // Will be used for collab tools.
#[allow(dead_code)] // Used by upcoming multi-agent tooling.
/// Fetch the last known status for `agent_id`, returning `NotFound` when unavailable.
pub(crate) async fn get_status(&self, agent_id: ThreadId) -> AgentStatus {
let Ok(state) = self.upgrade() else {
@@ -87,16 +82,6 @@ impl AgentControl {
thread.agent_status().await
}
/// Subscribe to status updates for `agent_id`, yielding the latest value and changes.
pub(crate) async fn subscribe_status(
&self,
agent_id: ThreadId,
) -> CodexResult<watch::Receiver<AgentStatus>> {
let state = self.upgrade()?;
let thread = state.get_thread(agent_id).await?;
Ok(thread.subscribe_status())
}
fn upgrade(&self) -> CodexResult<Arc<ThreadManagerState>> {
self.manager
.upgrade()
@@ -129,63 +114,13 @@ fn spawn_headless_drain(thread: Arc<CodexThread>) {
#[cfg(test)]
mod tests {
use super::*;
use crate::CodexAuth;
use crate::ThreadManager;
use crate::agent::agent_status_from_event;
use crate::config::Config;
use crate::config::ConfigBuilder;
use assert_matches::assert_matches;
use codex_protocol::protocol::ErrorEvent;
use codex_protocol::protocol::TurnAbortReason;
use codex_protocol::protocol::TurnAbortedEvent;
use codex_protocol::protocol::TurnCompleteEvent;
use codex_protocol::protocol::TurnStartedEvent;
use pretty_assertions::assert_eq;
use tempfile::TempDir;
async fn test_config() -> (TempDir, Config) {
let home = TempDir::new().expect("create temp dir");
let config = ConfigBuilder::default()
.codex_home(home.path().to_path_buf())
.build()
.await
.expect("load default test config");
(home, config)
}
struct AgentControlHarness {
_home: TempDir,
config: Config,
manager: ThreadManager,
control: AgentControl,
}
impl AgentControlHarness {
async fn new() -> Self {
let (home, config) = test_config().await;
let manager = ThreadManager::with_models_provider_and_home(
CodexAuth::from_api_key("dummy"),
config.model_provider.clone(),
config.codex_home.clone(),
);
let control = manager.agent_control();
Self {
_home: home,
config,
manager,
control,
}
}
async fn start_thread(&self) -> (ThreadId, Arc<CodexThread>) {
let new_thread = self
.manager
.start_thread(self.config.clone())
.await
.expect("start thread");
(new_thread.thread_id, new_thread.thread)
}
}
#[tokio::test]
async fn send_prompt_errors_when_manager_dropped() {
@@ -250,135 +185,4 @@ mod tests {
let status = agent_status_from_event(&EventMsg::ShutdownComplete);
assert_eq!(status, Some(AgentStatus::Shutdown));
}
#[tokio::test]
async fn spawn_agent_errors_when_manager_dropped() {
let control = AgentControl::default();
let (_home, config) = test_config().await;
let err = control
.spawn_agent(config, "hello".to_string(), false)
.await
.expect_err("spawn_agent should fail without a manager");
assert_eq!(
err.to_string(),
"unsupported operation: thread manager dropped"
);
}
#[tokio::test]
async fn send_prompt_errors_when_thread_missing() {
let harness = AgentControlHarness::new().await;
let thread_id = ThreadId::new();
let err = harness
.control
.send_prompt(thread_id, "hello".to_string())
.await
.expect_err("send_prompt should fail for missing thread");
assert_matches!(err, CodexErr::ThreadNotFound(id) if id == thread_id);
}
#[tokio::test]
async fn get_status_returns_not_found_for_missing_thread() {
let harness = AgentControlHarness::new().await;
let status = harness.control.get_status(ThreadId::new()).await;
assert_eq!(status, AgentStatus::NotFound);
}
#[tokio::test]
async fn get_status_returns_pending_init_for_new_thread() {
let harness = AgentControlHarness::new().await;
let (thread_id, _) = harness.start_thread().await;
let status = harness.control.get_status(thread_id).await;
assert_eq!(status, AgentStatus::PendingInit);
}
#[tokio::test]
async fn subscribe_status_errors_for_missing_thread() {
let harness = AgentControlHarness::new().await;
let thread_id = ThreadId::new();
let err = harness
.control
.subscribe_status(thread_id)
.await
.expect_err("subscribe_status should fail for missing thread");
assert_matches!(err, CodexErr::ThreadNotFound(id) if id == thread_id);
}
#[tokio::test]
async fn subscribe_status_updates_on_shutdown() {
let harness = AgentControlHarness::new().await;
let (thread_id, thread) = harness.start_thread().await;
let mut status_rx = harness
.control
.subscribe_status(thread_id)
.await
.expect("subscribe_status should succeed");
assert_eq!(status_rx.borrow().clone(), AgentStatus::PendingInit);
let _ = thread
.submit(Op::Shutdown {})
.await
.expect("shutdown should submit");
let _ = status_rx.changed().await;
assert_eq!(status_rx.borrow().clone(), AgentStatus::Shutdown);
}
#[tokio::test]
async fn send_prompt_submits_user_message() {
let harness = AgentControlHarness::new().await;
let (thread_id, _thread) = harness.start_thread().await;
let submission_id = harness
.control
.send_prompt(thread_id, "hello from tests".to_string())
.await
.expect("send_prompt should succeed");
assert!(!submission_id.is_empty());
let expected = (
thread_id,
Op::UserInput {
items: vec![UserInput::Text {
text: "hello from tests".to_string(),
}],
final_output_json_schema: None,
},
);
let captured = harness
.manager
.captured_ops()
.into_iter()
.find(|entry| *entry == expected);
assert_eq!(captured, Some(expected));
}
#[tokio::test]
async fn spawn_agent_creates_thread_and_sends_prompt() {
let harness = AgentControlHarness::new().await;
let thread_id = harness
.control
.spawn_agent(harness.config.clone(), "spawned".to_string(), false)
.await
.expect("spawn_agent should succeed");
let _thread = harness
.manager
.get_thread(thread_id)
.await
.expect("thread should be registered");
let expected = (
thread_id,
Op::UserInput {
items: vec![UserInput::Text {
text: "spawned".to_string(),
}],
final_output_json_schema: None,
},
);
let captured = harness
.manager
.captured_ops()
.into_iter()
.find(|entry| *entry == expected);
assert_eq!(captured, Some(expected));
}
}

View File

@@ -13,7 +13,3 @@ pub(crate) fn agent_status_from_event(msg: &EventMsg) -> Option<AgentStatus> {
_ => None,
}
}
pub(crate) fn is_final(status: &AgentStatus) -> bool {
!matches!(status, AgentStatus::PendingInit | AgentStatus::Running)
}

View File

@@ -164,7 +164,6 @@ use codex_protocol::protocol::InitialHistory;
use codex_protocol::user_input::UserInput;
use codex_utils_readiness::Readiness;
use codex_utils_readiness::ReadinessFlag;
use tokio::sync::watch;
/// The high-level interface to the Codex system.
/// It operates as a queue pair where you send submissions and receive events.
@@ -173,7 +172,7 @@ pub struct Codex {
pub(crate) tx_sub: Sender<Submission>,
pub(crate) rx_event: Receiver<Event>,
// Last known status of the agent.
pub(crate) agent_status: watch::Receiver<AgentStatus>,
pub(crate) agent_status: Arc<RwLock<AgentStatus>>,
}
/// Wrapper returned by [`Codex::spawn`] containing the spawned [`Codex`],
@@ -276,7 +275,7 @@ impl Codex {
// Generate a unique ID for the lifetime of this Codex session.
let session_source_clone = session_configuration.session_source.clone();
let (agent_status_tx, agent_status_rx) = watch::channel(AgentStatus::PendingInit);
let agent_status = Arc::new(RwLock::new(AgentStatus::PendingInit));
let session = Session::new(
session_configuration,
@@ -285,7 +284,7 @@ impl Codex {
models_manager.clone(),
exec_policy,
tx_event.clone(),
agent_status_tx.clone(),
Arc::clone(&agent_status),
conversation_history,
session_source_clone,
skills_manager,
@@ -304,7 +303,7 @@ impl Codex {
next_id: AtomicU64::new(0),
tx_sub,
rx_event,
agent_status: agent_status_rx,
agent_status,
};
#[allow(deprecated)]
@@ -346,7 +345,8 @@ impl Codex {
}
pub(crate) async fn agent_status(&self) -> AgentStatus {
self.agent_status.borrow().clone()
let status = self.agent_status.read().await;
status.clone()
}
}
@@ -354,9 +354,9 @@ impl Codex {
///
/// A session has at most 1 running task at a time, and can be interrupted by user input.
pub(crate) struct Session {
pub(crate) conversation_id: ThreadId,
conversation_id: ThreadId,
tx_event: Sender<Event>,
agent_status: watch::Sender<AgentStatus>,
agent_status: Arc<RwLock<AgentStatus>>,
state: Mutex<SessionState>,
/// The set of enabled features should be invariant for the lifetime of the
/// session.
@@ -557,7 +557,7 @@ impl Session {
models_manager: Arc<ModelsManager>,
exec_policy: ExecPolicyManager,
tx_event: Sender<Event>,
agent_status: watch::Sender<AgentStatus>,
agent_status: Arc<RwLock<AgentStatus>>,
initial_history: InitialHistory,
session_source: SessionSource,
skills_manager: Arc<SkillsManager>,
@@ -703,7 +703,7 @@ impl Session {
let sess = Arc::new(Session {
conversation_id,
tx_event: tx_event.clone(),
agent_status,
agent_status: Arc::clone(&agent_status),
state: Mutex::new(state),
features: config.features.clone(),
active_turn: Mutex::new(None),
@@ -1026,7 +1026,8 @@ impl Session {
pub(crate) async fn send_event_raw(&self, event: Event) {
// Record the last known agent status.
if let Some(status) = agent_status_from_event(&event.msg) {
self.agent_status.send_replace(status);
let mut guard = self.agent_status.write().await;
*guard = status;
}
// Persist the event into rollout (recorder filters as needed)
let rollout_items = vec![RolloutItem::EventMsg(event.msg.clone())];
@@ -1044,7 +1045,8 @@ impl Session {
pub(crate) async fn send_event_raw_flushed(&self, event: Event) {
// Record the last known agent status.
if let Some(status) = agent_status_from_event(&event.msg) {
self.agent_status.send_replace(status);
let mut guard = self.agent_status.write().await;
*guard = status;
}
self.persist_rollout_items(&[RolloutItem::EventMsg(event.msg.clone())])
.await;
@@ -1538,24 +1540,6 @@ impl Session {
}
}
/// Returns the input if there was no task running to inject into
pub async fn inject_response_items(
&self,
input: Vec<ResponseInputItem>,
) -> Result<(), Vec<ResponseInputItem>> {
let mut active = self.active_turn.lock().await;
match active.as_mut() {
Some(at) => {
let mut ts = at.turn_state.lock().await;
for item in input {
ts.push_pending_input(item);
}
Ok(())
}
None => Err(input),
}
}
pub async fn get_pending_input(&self) -> Vec<ResponseInputItem> {
let mut active = self.active_turn.lock().await;
match active.as_mut() {
@@ -3492,7 +3476,7 @@ mod tests {
));
let agent_control = AgentControl::default();
let exec_policy = ExecPolicyManager::default();
let (agent_status_tx, _agent_status_rx) = watch::channel(AgentStatus::PendingInit);
let agent_status = Arc::new(RwLock::new(AgentStatus::PendingInit));
let model = ModelsManager::get_model_offline(config.model.as_deref());
let session_configuration = SessionConfiguration {
provider: config.model_provider.clone(),
@@ -3555,7 +3539,7 @@ mod tests {
let session = Session {
conversation_id,
tx_event,
agent_status: agent_status_tx,
agent_status: Arc::clone(&agent_status),
state: Mutex::new(state),
features: config.features.clone(),
active_turn: Mutex::new(None),
@@ -3586,7 +3570,7 @@ mod tests {
));
let agent_control = AgentControl::default();
let exec_policy = ExecPolicyManager::default();
let (agent_status_tx, _agent_status_rx) = watch::channel(AgentStatus::PendingInit);
let agent_status = Arc::new(RwLock::new(AgentStatus::PendingInit));
let model = ModelsManager::get_model_offline(config.model.as_deref());
let session_configuration = SessionConfiguration {
provider: config.model_provider.clone(),
@@ -3649,7 +3633,7 @@ mod tests {
let session = Arc::new(Session {
conversation_id,
tx_event,
agent_status: agent_status_tx,
agent_status: Arc::clone(&agent_status),
state: Mutex::new(state),
features: config.features.clone(),
active_turn: Mutex::new(None),

View File

@@ -87,7 +87,7 @@ pub(crate) async fn run_codex_thread_interactive(
next_id: AtomicU64::new(0),
tx_sub: tx_ops,
rx_event: rx_sub,
agent_status: codex.agent_status.clone(),
agent_status: Arc::clone(&codex.agent_status),
})
}
@@ -129,7 +129,7 @@ pub(crate) async fn run_codex_thread_one_shot(
// Bridge events so we can observe completion and shut down automatically.
let (tx_bridge, rx_bridge) = async_channel::bounded(SUBMISSION_CHANNEL_CAPACITY);
let ops_tx = io.tx_sub.clone();
let agent_status = io.agent_status.clone();
let agent_status = Arc::clone(&io.agent_status);
let io_for_bridge = io;
tokio::spawn(async move {
while let Ok(event) = io_for_bridge.next_event().await {
@@ -363,23 +363,20 @@ mod tests {
use super::*;
use async_channel::bounded;
use codex_protocol::models::ResponseItem;
use codex_protocol::protocol::AgentStatus;
use codex_protocol::protocol::RawResponseItemEvent;
use codex_protocol::protocol::TurnAbortReason;
use codex_protocol::protocol::TurnAbortedEvent;
use pretty_assertions::assert_eq;
use tokio::sync::watch;
#[tokio::test]
async fn forward_events_cancelled_while_send_blocked_shuts_down_delegate() {
let (tx_events, rx_events) = bounded(1);
let (tx_sub, rx_sub) = bounded(SUBMISSION_CHANNEL_CAPACITY);
let (_agent_status_tx, agent_status) = watch::channel(AgentStatus::PendingInit);
let codex = Arc::new(Codex {
next_id: AtomicU64::new(0),
tx_sub,
rx_event: rx_events,
agent_status,
agent_status: Default::default(),
});
let (session, ctx, _rx_evt) = crate::codex::make_session_and_context_with_rx().await;

View File

@@ -5,7 +5,6 @@ use crate::protocol::Event;
use crate::protocol::Op;
use crate::protocol::Submission;
use std::path::PathBuf;
use tokio::sync::watch;
pub struct CodexThread {
codex: Codex,
@@ -39,10 +38,6 @@ impl CodexThread {
self.codex.agent_status().await
}
pub(crate) fn subscribe_status(&self) -> watch::Receiver<AgentStatus> {
self.codex.agent_status.clone()
}
pub fn rollout_path(&self) -> PathBuf {
self.rollout_path.clone()
}

View File

@@ -9,10 +9,6 @@ use codex_protocol::models::ReasoningItemContent;
use codex_protocol::models::ReasoningItemReasoningSummary;
use codex_protocol::models::ResponseItem;
use codex_protocol::models::WebSearchAction;
use codex_protocol::models::is_image_close_tag_text;
use codex_protocol::models::is_image_open_tag_text;
use codex_protocol::models::is_local_image_close_tag_text;
use codex_protocol::models::is_local_image_open_tag_text;
use codex_protocol::user_input::UserInput;
use tracing::warn;
use uuid::Uuid;
@@ -36,17 +32,9 @@ fn parse_user_message(message: &[ContentItem]) -> Option<UserMessageItem> {
let mut content: Vec<UserInput> = Vec::new();
for (idx, content_item) in message.iter().enumerate() {
for content_item in message.iter() {
match content_item {
ContentItem::InputText { text } => {
if (is_local_image_open_tag_text(text) || is_image_open_tag_text(text))
&& (matches!(message.get(idx + 1), Some(ContentItem::InputImage { .. })))
|| (idx > 0
&& (is_local_image_close_tag_text(text) || is_image_close_tag_text(text))
&& matches!(message.get(idx - 1), Some(ContentItem::InputImage { .. })))
{
continue;
}
if is_session_prefix(text) || is_user_shell_command_text(text) {
return None;
}
@@ -189,80 +177,6 @@ mod tests {
}
}
#[test]
fn skips_local_image_label_text() {
let image_url = "data:image/png;base64,abc".to_string();
let label = codex_protocol::models::local_image_open_tag_text(1);
let user_text = "Please review this image.".to_string();
let item = ResponseItem::Message {
id: None,
role: "user".to_string(),
content: vec![
ContentItem::InputText { text: label },
ContentItem::InputImage {
image_url: image_url.clone(),
},
ContentItem::InputText {
text: "</image>".to_string(),
},
ContentItem::InputText {
text: user_text.clone(),
},
],
};
let turn_item = parse_turn_item(&item).expect("expected user message turn item");
match turn_item {
TurnItem::UserMessage(user) => {
let expected_content = vec![
UserInput::Image { image_url },
UserInput::Text { text: user_text },
];
assert_eq!(user.content, expected_content);
}
other => panic!("expected TurnItem::UserMessage, got {other:?}"),
}
}
#[test]
fn skips_unnamed_image_label_text() {
let image_url = "data:image/png;base64,abc".to_string();
let label = codex_protocol::models::image_open_tag_text();
let user_text = "Please review this image.".to_string();
let item = ResponseItem::Message {
id: None,
role: "user".to_string(),
content: vec![
ContentItem::InputText { text: label },
ContentItem::InputImage {
image_url: image_url.clone(),
},
ContentItem::InputText {
text: codex_protocol::models::image_close_tag_text(),
},
ContentItem::InputText {
text: user_text.clone(),
},
],
};
let turn_item = parse_turn_item(&item).expect("expected user message turn item");
match turn_item {
TurnItem::UserMessage(user) => {
let expected_content = vec![
UserInput::Image { image_url },
UserInput::Text { text: user_text },
];
assert_eq!(user.content, expected_content);
}
other => panic!("expected TurnItem::UserMessage, got {other:?}"),
}
}
#[test]
fn skips_user_instructions_and_env() {
let items = vec![

View File

@@ -90,7 +90,6 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
| EventMsg::AgentMessageContentDelta(_)
| EventMsg::ReasoningContentDelta(_)
| EventMsg::ReasoningRawContentDelta(_)
| EventMsg::SkillsUpdateAvailable
| EventMsg::CollabInteraction(_) => false,
| EventMsg::SkillsUpdateAvailable => false,
}
}

View File

@@ -56,10 +56,6 @@ pub(crate) struct ThreadManagerState {
models_manager: Arc<ModelsManager>,
skills_manager: Arc<SkillsManager>,
session_source: SessionSource,
#[cfg(any(test, feature = "test-support"))]
#[allow(dead_code)]
// Captures submitted ops for testing purpose.
ops_log: Arc<std::sync::Mutex<Vec<(ThreadId, Op)>>>,
}
impl ThreadManager {
@@ -78,8 +74,6 @@ impl ThreadManager {
skills_manager: Arc::new(SkillsManager::new(codex_home)),
auth_manager,
session_source,
#[cfg(any(test, feature = "test-support"))]
ops_log: Arc::new(std::sync::Mutex::new(Vec::new())),
}),
#[cfg(any(test, feature = "test-support"))]
_test_codex_home_guard: None,
@@ -117,8 +111,6 @@ impl ThreadManager {
skills_manager: Arc::new(SkillsManager::new(codex_home)),
auth_manager,
session_source: SessionSource::Exec,
#[cfg(any(test, feature = "test-support"))]
ops_log: Arc::new(std::sync::Mutex::new(Vec::new())),
}),
_test_codex_home_guard: None,
}
@@ -210,19 +202,9 @@ impl ThreadManager {
.await
}
pub(crate) fn agent_control(&self) -> AgentControl {
fn agent_control(&self) -> AgentControl {
AgentControl::new(Arc::downgrade(&self.state))
}
#[cfg(any(test, feature = "test-support"))]
#[allow(dead_code)]
pub(crate) fn captured_ops(&self) -> Vec<(ThreadId, Op)> {
self.state
.ops_log
.lock()
.map(|log| log.clone())
.unwrap_or_default()
}
}
impl ThreadManagerState {
@@ -235,14 +217,7 @@ impl ThreadManagerState {
}
pub(crate) async fn send_op(&self, thread_id: ThreadId, op: Op) -> CodexResult<String> {
let thread = self.get_thread(thread_id).await?;
#[cfg(any(test, feature = "test-support"))]
{
if let Ok(mut log) = self.ops_log.lock() {
log.push((thread_id, op.clone()));
}
}
thread.submit(op).await
self.get_thread(thread_id).await?.submit(op).await
}
#[allow(dead_code)] // Used by upcoming multi-agent tooling.

View File

@@ -1,4 +1,3 @@
use crate::agent::AgentStatus;
use crate::codex::TurnContext;
use crate::config::Config;
use crate::error::CodexErr;
@@ -11,16 +10,30 @@ use crate::tools::registry::ToolHandler;
use crate::tools::registry::ToolKind;
use async_trait::async_trait;
use codex_protocol::ThreadId;
use codex_protocol::protocol::CollabInteractionEvent;
use codex_protocol::protocol::EventMsg;
use serde::Deserialize;
use serde::Serialize;
pub struct CollabHandler;
pub(crate) const DEFAULT_WAIT_TIMEOUT_MS: i64 = 30_000;
pub(crate) const MAX_WAIT_TIMEOUT_MS: i64 = 300_000;
#[derive(Debug, Deserialize)]
struct SpawnAgentArgs {
message: String,
}
#[derive(Debug, Deserialize)]
struct SendInputArgs {
id: String,
message: String,
}
#[derive(Debug, Deserialize)]
struct WaitArgs {
id: String,
timeout_ms: Option<i64>,
}
#[derive(Debug, Deserialize)]
struct CloseAgentArgs {
id: String,
@@ -55,10 +68,10 @@ impl ToolHandler for CollabHandler {
};
match tool_name.as_str() {
"spawn_agent" => spawn::handle(session, turn, arguments).await,
"send_input" => send_input::handle(session, turn, arguments).await,
"wait" => wait::handle(session, turn, arguments).await,
"close_agent" => close_agent::handle(session, turn, arguments).await,
"spawn_agent" => handle_spawn_agent(session, turn, arguments).await,
"send_input" => handle_send_input(session, arguments).await,
"wait" => handle_wait(arguments).await,
"close_agent" => handle_close_agent(arguments).await,
other => Err(FunctionCallError::RespondToModel(format!(
"unsupported collab tool {other}"
))),
@@ -66,309 +79,84 @@ impl ToolHandler for CollabHandler {
}
}
mod spawn {
use super::*;
use crate::codex::Session;
use std::sync::Arc;
#[derive(Debug, Deserialize)]
struct SpawnAgentArgs {
message: String,
async fn handle_spawn_agent(
session: std::sync::Arc<crate::codex::Session>,
turn: std::sync::Arc<TurnContext>,
arguments: String,
) -> Result<ToolOutput, FunctionCallError> {
let args: SpawnAgentArgs = parse_arguments(&arguments)?;
if args.message.trim().is_empty() {
return Err(FunctionCallError::RespondToModel(
"Empty message can't be send to an agent".to_string(),
));
}
let config = build_agent_spawn_config(turn.as_ref())?;
let result = session
.services
.agent_control
.spawn_agent(config, args.message, true)
.await
.map_err(|err| FunctionCallError::Fatal(err.to_string()))?;
pub async fn handle(
session: Arc<Session>,
turn: Arc<TurnContext>,
arguments: String,
) -> Result<ToolOutput, FunctionCallError> {
let args: SpawnAgentArgs = parse_arguments(&arguments)?;
if args.message.trim().is_empty() {
return Err(FunctionCallError::RespondToModel(
"Empty message can't be send to an agent".to_string(),
));
}
let config = build_agent_spawn_config(turn.as_ref())?;
let result = session
.services
.agent_control
.spawn_agent(config, args.message.clone(), true)
.await
.map_err(|err| FunctionCallError::Fatal(err.to_string()))?;
emit_event(session, turn, args.message, result).await;
Ok(ToolOutput::Function {
content: format!("agent_id: {result}"),
success: Some(true),
content_items: None,
})
}
async fn emit_event(
session: Arc<Session>,
turn: Arc<TurnContext>,
prompt: String,
new_id: ThreadId,
) {
session
.send_event(
&turn,
EventMsg::CollabInteraction(CollabInteractionEvent::AgentSpawned {
sender_id: session.conversation_id,
new_id,
prompt,
}),
)
.await
}
Ok(ToolOutput::Function {
content: format!("agent_id: {result}"),
success: Some(true),
content_items: None,
})
}
mod send_input {
use super::*;
use crate::codex::Session;
use std::sync::Arc;
#[derive(Debug, Deserialize)]
struct SendInputArgs {
id: String,
message: String,
async fn handle_send_input(
session: std::sync::Arc<crate::codex::Session>,
arguments: String,
) -> Result<ToolOutput, FunctionCallError> {
let args: SendInputArgs = parse_arguments(&arguments)?;
let agent_id = agent_id(&args.id)?;
if args.message.trim().is_empty() {
return Err(FunctionCallError::RespondToModel(
"Empty message can't be send to an agent".to_string(),
));
}
pub async fn handle(
session: Arc<Session>,
turn: Arc<TurnContext>,
arguments: String,
) -> Result<ToolOutput, FunctionCallError> {
let args: SendInputArgs = parse_arguments(&arguments)?;
let agent_id = agent_id(&args.id)?;
if args.message.trim().is_empty() {
return Err(FunctionCallError::RespondToModel(
"Empty message can't be send to an agent".to_string(),
));
}
let content = session
.services
.agent_control
.send_prompt(agent_id, args.message.clone())
.await
.map_err(|err| match err {
CodexErr::ThreadNotFound(id) => {
FunctionCallError::RespondToModel(format!("agent with id {id} not found"))
}
err => FunctionCallError::Fatal(err.to_string()),
})?;
emit_event(session, turn, agent_id, args.message).await;
Ok(ToolOutput::Function {
content,
success: Some(true),
content_items: None,
})
}
async fn emit_event(
session: Arc<Session>,
turn: Arc<TurnContext>,
receiver_id: ThreadId,
prompt: String,
) {
session
.send_event(
&turn,
EventMsg::CollabInteraction(CollabInteractionEvent::AgentInteraction {
sender_id: session.conversation_id,
receiver_id,
prompt,
}),
)
.await
}
}
mod wait {
use super::*;
use crate::agent::status::is_final;
use crate::codex::Session;
use std::sync::Arc;
use std::time::Duration;
use tokio::time::Instant;
use tokio::time::timeout_at;
#[derive(Debug, Deserialize)]
struct WaitArgs {
id: String,
timeout_ms: Option<i64>,
}
#[derive(Debug, Serialize)]
struct WaitResult {
status: AgentStatus,
timed_out: bool,
}
pub async fn handle(
session: Arc<Session>,
turn: Arc<TurnContext>,
arguments: String,
) -> Result<ToolOutput, FunctionCallError> {
let args: WaitArgs = parse_arguments(&arguments)?;
let agent_id = agent_id(&args.id)?;
// Validate timeout.
let timeout_ms = args.timeout_ms.unwrap_or(DEFAULT_WAIT_TIMEOUT_MS);
let timeout_ms = match timeout_ms {
ms if ms <= 0 => {
return Err(FunctionCallError::RespondToModel(
"timeout_ms must be greater than zero".to_owned(),
));
let content = session
.services
.agent_control
.send_prompt(agent_id, args.message)
.await
.map_err(|err| match err {
CodexErr::ThreadNotFound(id) => {
FunctionCallError::RespondToModel(format!("agent with id {id} not found"))
}
ms => ms.min(MAX_WAIT_TIMEOUT_MS),
};
let mut status_rx = session
.services
.agent_control
.subscribe_status(agent_id)
.await
.map_err(|err| match err {
CodexErr::ThreadNotFound(id) => {
FunctionCallError::RespondToModel(format!("agent with id {id} not found"))
}
err => FunctionCallError::Fatal(err.to_string()),
})?;
let waiting_id = format!("collab-waiting-{}", uuid::Uuid::new_v4());
session
.send_event(
&turn,
EventMsg::CollabInteraction(CollabInteractionEvent::WaitingBegin {
sender_id: session.conversation_id,
receiver_id: agent_id,
waiting_id: waiting_id.clone(),
}),
)
.await;
// Get last known status.
let mut status = status_rx.borrow_and_update().clone();
let deadline = Instant::now() + Duration::from_millis(timeout_ms as u64);
let timed_out = loop {
if is_final(&status) {
break false;
}
match timeout_at(deadline, status_rx.changed()).await {
Ok(Ok(())) => status = status_rx.borrow().clone(),
Ok(Err(_)) => {
let last_status = session.services.agent_control.get_status(agent_id).await;
if last_status != AgentStatus::NotFound {
// On-purpose we keep the last known status if the agent gets dropped. This
// event is not supposed to happen.
status = last_status;
}
break false;
}
Err(_) => break true,
}
};
session
.send_event(
&turn,
EventMsg::CollabInteraction(CollabInteractionEvent::WaitingEnd {
sender_id: session.conversation_id,
receiver_id: agent_id,
waiting_id,
status: status.clone(),
}),
)
.await;
if matches!(status, AgentStatus::NotFound) {
return Err(FunctionCallError::RespondToModel(format!(
"agent with id {agent_id} not found"
)));
}
let result = WaitResult { status, timed_out };
let content = serde_json::to_string(&result).map_err(|err| {
FunctionCallError::Fatal(format!("failed to serialize wait result: {err}"))
err => FunctionCallError::Fatal(err.to_string()),
})?;
Ok(ToolOutput::Function {
content,
success: Some(!result.timed_out),
content_items: None,
})
}
Ok(ToolOutput::Function {
content,
success: Some(true),
content_items: None,
})
}
pub mod close_agent {
use super::*;
use crate::codex::Session;
use std::sync::Arc;
async fn handle_wait(arguments: String) -> Result<ToolOutput, FunctionCallError> {
let args: WaitArgs = parse_arguments(&arguments)?;
let _agent_id = agent_id(&args.id)?;
#[derive(Debug, Deserialize, Serialize)]
pub(super) struct CloseAgentResult {
pub(super) status: AgentStatus,
let timeout_ms = args.timeout_ms.unwrap_or(DEFAULT_WAIT_TIMEOUT_MS);
if timeout_ms <= 0 {
return Err(FunctionCallError::RespondToModel(
"timeout_ms must be greater than zero".to_string(),
));
}
let _timeout_ms = timeout_ms.min(MAX_WAIT_TIMEOUT_MS);
// TODO(jif): implement agent wait once lifecycle tracking is wired up.
Err(FunctionCallError::Fatal("wait not implemented".to_string()))
}
pub async fn handle(
session: Arc<Session>,
turn: Arc<TurnContext>,
arguments: String,
) -> Result<ToolOutput, FunctionCallError> {
let args: CloseAgentArgs = parse_arguments(&arguments)?;
let agent_id = agent_id(&args.id)?;
let status = session.services.agent_control.get_status(agent_id).await;
if !matches!(status, AgentStatus::Shutdown) {
let _ = session
.services
.agent_control
.shutdown_agent(agent_id)
.await
.map_err(|err| match err {
CodexErr::ThreadNotFound(id) => {
FunctionCallError::RespondToModel(format!("agent with id {id} not found"))
}
err => FunctionCallError::Fatal(err.to_string()),
})?;
}
emit_event(session, turn, agent_id, status.clone()).await;
let content = serde_json::to_string(&CloseAgentResult { status }).map_err(|err| {
FunctionCallError::Fatal(format!("failed to serialize close_agent result: {err}"))
})?;
Ok(ToolOutput::Function {
content,
success: Some(true),
content_items: None,
})
}
async fn emit_event(
session: Arc<Session>,
turn: Arc<TurnContext>,
receiver_id: ThreadId,
status: AgentStatus,
) {
session
.send_event(
&turn,
EventMsg::CollabInteraction(CollabInteractionEvent::Close {
sender_id: session.conversation_id,
receiver_id,
status,
}),
)
.await
}
async fn handle_close_agent(arguments: String) -> Result<ToolOutput, FunctionCallError> {
let args: CloseAgentArgs = parse_arguments(&arguments)?;
let _agent_id = agent_id(&args.id)?;
// TODO(jif): implement agent shutdown and return the final status.
Err(FunctionCallError::Fatal(
"close_agent not implemented".to_string(),
))
}
fn agent_id(id: &str) -> Result<ThreadId, FunctionCallError> {
@@ -404,384 +192,3 @@ fn build_agent_spawn_config(turn: &TurnContext) -> Result<Config, FunctionCallEr
})?;
Ok(config)
}
#[cfg(test)]
mod tests {
use super::*;
use crate::CodexAuth;
use crate::ThreadManager;
use crate::built_in_model_providers;
use crate::codex::make_session_and_context;
use crate::config::types::ShellEnvironmentPolicy;
use crate::function_tool::FunctionCallError;
use crate::protocol::AskForApproval;
use crate::protocol::Op;
use crate::protocol::SandboxPolicy;
use crate::turn_diff_tracker::TurnDiffTracker;
use codex_protocol::ThreadId;
use pretty_assertions::assert_eq;
use serde_json::json;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::Duration;
use tokio::sync::Mutex;
use tokio::time::timeout;
fn invocation(
session: Arc<crate::codex::Session>,
turn: Arc<TurnContext>,
tool_name: &str,
payload: ToolPayload,
) -> ToolInvocation {
ToolInvocation {
session,
turn,
tracker: Arc::new(Mutex::new(TurnDiffTracker::default())),
call_id: "call-1".to_string(),
tool_name: tool_name.to_string(),
payload,
}
}
fn function_payload(args: serde_json::Value) -> ToolPayload {
ToolPayload::Function {
arguments: args.to_string(),
}
}
fn thread_manager() -> ThreadManager {
ThreadManager::with_models_provider(
CodexAuth::from_api_key("dummy"),
built_in_model_providers()["openai"].clone(),
)
}
#[tokio::test]
async fn handler_rejects_non_function_payloads() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"spawn_agent",
ToolPayload::Custom {
input: "hello".to_string(),
},
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("payload should be rejected");
};
assert_eq!(
err,
FunctionCallError::RespondToModel(
"collab handler received unsupported payload".to_string()
)
);
}
#[tokio::test]
async fn handler_rejects_unknown_tool() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"unknown_tool",
function_payload(json!({})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("tool should be rejected");
};
assert_eq!(
err,
FunctionCallError::RespondToModel("unsupported collab tool unknown_tool".to_string())
);
}
#[tokio::test]
async fn spawn_agent_rejects_empty_message() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"spawn_agent",
function_payload(json!({"message": " "})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("empty message should be rejected");
};
assert_eq!(
err,
FunctionCallError::RespondToModel(
"Empty message can't be send to an agent".to_string()
)
);
}
#[tokio::test]
async fn spawn_agent_errors_when_manager_dropped() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"spawn_agent",
function_payload(json!({"message": "hello"})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("spawn should fail without a manager");
};
assert_eq!(
err,
FunctionCallError::Fatal("unsupported operation: thread manager dropped".to_string())
);
}
#[tokio::test]
async fn send_input_rejects_empty_message() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"send_input",
function_payload(json!({"id": ThreadId::new().to_string(), "message": ""})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("empty message should be rejected");
};
assert_eq!(
err,
FunctionCallError::RespondToModel(
"Empty message can't be send to an agent".to_string()
)
);
}
#[tokio::test]
async fn send_input_rejects_invalid_id() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"send_input",
function_payload(json!({"id": "not-a-uuid", "message": "hi"})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("invalid id should be rejected");
};
let FunctionCallError::RespondToModel(msg) = err else {
panic!("expected respond-to-model error");
};
assert!(msg.starts_with("invalid agent id not-a-uuid:"));
}
#[tokio::test]
async fn send_input_reports_missing_agent() {
let (mut session, turn) = make_session_and_context().await;
let manager = thread_manager();
session.services.agent_control = manager.agent_control();
let agent_id = ThreadId::new();
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"send_input",
function_payload(json!({"id": agent_id.to_string(), "message": "hi"})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("missing agent should be reported");
};
assert_eq!(
err,
FunctionCallError::RespondToModel(format!("agent with id {agent_id} not found"))
);
}
#[tokio::test]
async fn wait_rejects_non_positive_timeout() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"wait",
function_payload(json!({"id": ThreadId::new().to_string(), "timeout_ms": 0})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("non-positive timeout should be rejected");
};
assert_eq!(
err,
FunctionCallError::RespondToModel("timeout_ms must be greater than zero".to_string())
);
}
#[tokio::test]
async fn wait_rejects_invalid_id() {
let (session, turn) = make_session_and_context().await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"wait",
function_payload(json!({"id": "invalid"})),
);
let Err(err) = CollabHandler.handle(invocation).await else {
panic!("invalid id should be rejected");
};
let FunctionCallError::RespondToModel(msg) = err else {
panic!("expected respond-to-model error");
};
assert!(msg.starts_with("invalid agent id invalid:"));
}
#[tokio::test]
async fn wait_times_out_when_status_is_not_final() {
let (mut session, turn) = make_session_and_context().await;
let manager = thread_manager();
session.services.agent_control = manager.agent_control();
let config = turn.client.config().as_ref().clone();
let thread = manager.start_thread(config).await.expect("start thread");
let agent_id = thread.thread_id;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"wait",
function_payload(json!({"id": agent_id.to_string(), "timeout_ms": 10})),
);
let output = CollabHandler
.handle(invocation)
.await
.expect("wait should succeed");
let ToolOutput::Function {
content, success, ..
} = output
else {
panic!("expected function output");
};
assert_eq!(content, r#"{"status":"pending_init","timed_out":true}"#);
assert_eq!(success, Some(false));
let _ = thread
.thread
.submit(Op::Shutdown {})
.await
.expect("shutdown should submit");
}
#[tokio::test]
async fn wait_returns_final_status_without_timeout() {
let (mut session, turn) = make_session_and_context().await;
let manager = thread_manager();
session.services.agent_control = manager.agent_control();
let config = turn.client.config().as_ref().clone();
let thread = manager.start_thread(config).await.expect("start thread");
let agent_id = thread.thread_id;
let mut status_rx = manager
.agent_control()
.subscribe_status(agent_id)
.await
.expect("subscribe should succeed");
let _ = thread
.thread
.submit(Op::Shutdown {})
.await
.expect("shutdown should submit");
let _ = timeout(Duration::from_secs(1), status_rx.changed())
.await
.expect("shutdown status should arrive");
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"wait",
function_payload(json!({"id": agent_id.to_string(), "timeout_ms": 1000})),
);
let output = CollabHandler
.handle(invocation)
.await
.expect("wait should succeed");
let ToolOutput::Function {
content, success, ..
} = output
else {
panic!("expected function output");
};
assert_eq!(content, r#"{"status":"shutdown","timed_out":false}"#);
assert_eq!(success, Some(true));
}
#[tokio::test]
async fn close_agent_submits_shutdown_and_returns_status() {
let (mut session, turn) = make_session_and_context().await;
let manager = thread_manager();
session.services.agent_control = manager.agent_control();
let config = turn.client.config().as_ref().clone();
let thread = manager.start_thread(config).await.expect("start thread");
let agent_id = thread.thread_id;
let status_before = manager.agent_control().get_status(agent_id).await;
let invocation = invocation(
Arc::new(session),
Arc::new(turn),
"close_agent",
function_payload(json!({"id": agent_id.to_string()})),
);
let output = CollabHandler
.handle(invocation)
.await
.expect("close_agent should succeed");
let ToolOutput::Function {
content, success, ..
} = output
else {
panic!("expected function output");
};
let result: close_agent::CloseAgentResult =
serde_json::from_str(&content).expect("close_agent result should be json");
assert_eq!(result.status, status_before);
assert_eq!(success, Some(true));
let ops = manager.captured_ops();
let submitted_shutdown = ops
.iter()
.any(|(id, op)| *id == agent_id && matches!(op, Op::Shutdown));
assert_eq!(submitted_shutdown, true);
}
#[tokio::test]
async fn build_agent_spawn_config_uses_turn_context_values() {
let (_session, mut turn) = make_session_and_context().await;
turn.developer_instructions = Some("dev".to_string());
turn.base_instructions = Some("base".to_string());
turn.compact_prompt = Some("compact".to_string());
turn.user_instructions = Some("user".to_string());
turn.shell_environment_policy = ShellEnvironmentPolicy {
use_profile: true,
..ShellEnvironmentPolicy::default()
};
let temp_dir = tempfile::tempdir().expect("temp dir");
turn.cwd = temp_dir.path().to_path_buf();
turn.codex_linux_sandbox_exe = Some(PathBuf::from("/bin/echo"));
turn.approval_policy = AskForApproval::Never;
turn.sandbox_policy = SandboxPolicy::DangerFullAccess;
let config = build_agent_spawn_config(&turn).expect("spawn config");
let mut expected = (*turn.client.config()).clone();
expected.model = Some(turn.client.get_model());
expected.model_provider = turn.client.get_provider();
expected.model_reasoning_effort = turn.client.get_reasoning_effort();
expected.model_reasoning_summary = turn.client.get_reasoning_summary();
expected.developer_instructions = turn.developer_instructions.clone();
expected.base_instructions = turn.base_instructions.clone();
expected.compact_prompt = turn.compact_prompt.clone();
expected.user_instructions = turn.user_instructions.clone();
expected.shell_environment_policy = turn.shell_environment_policy.clone();
expected.codex_linux_sandbox_exe = turn.codex_linux_sandbox_exe.clone();
expected.cwd = turn.cwd.clone();
expected
.approval_policy
.set(turn.approval_policy)
.expect("approval policy set");
expected
.sandbox_policy
.set(turn.sandbox_policy)
.expect("sandbox policy set");
assert_eq!(config, expected);
}
}

View File

@@ -11,9 +11,7 @@ use crate::tools::context::ToolPayload;
use crate::tools::handlers::parse_arguments;
use crate::tools::registry::ToolHandler;
use crate::tools::registry::ToolKind;
use codex_protocol::models::ContentItem;
use codex_protocol::models::ResponseInputItem;
use codex_protocol::models::local_image_content_items_with_label_number;
use codex_protocol::user_input::UserInput;
pub struct ViewImageHandler;
@@ -65,15 +63,8 @@ impl ToolHandler for ViewImageHandler {
}
let event_path = abs_path.clone();
let content: Vec<ContentItem> =
local_image_content_items_with_label_number(&abs_path, None);
let input = ResponseInputItem::Message {
role: "user".to_string(),
content,
};
session
.inject_response_items(vec![input])
.inject_input(vec![UserInput::LocalImage { path: abs_path }])
.await
.map_err(|_| {
FunctionCallError::RespondToModel(

View File

@@ -8,7 +8,6 @@ use crate::tools::handlers::apply_patch::create_apply_patch_json_tool;
use crate::tools::handlers::collab::DEFAULT_WAIT_TIMEOUT_MS;
use crate::tools::handlers::collab::MAX_WAIT_TIMEOUT_MS;
use crate::tools::registry::ToolRegistryBuilder;
use codex_protocol::models::VIEW_IMAGE_TOOL_NAME;
use codex_protocol::openai_models::ApplyPatchToolType;
use codex_protocol::openai_models::ConfigShellToolType;
use codex_protocol::openai_models::ModelInfo;
@@ -413,9 +412,10 @@ fn create_view_image_tool() -> ToolSpec {
)]);
ToolSpec::Function(ResponsesApiTool {
name: VIEW_IMAGE_TOOL_NAME.to_string(),
description: "View a local image from the filesystem (only use if given a full filepath by the user, and the image isn't already attached to the thread context within <image ...> tags)."
.to_string(),
name: "view_image".to_string(),
description:
"Attach a local image (by filesystem path) to the thread context for this turn."
.to_string(),
strict: false,
parameters: JsonSchema::Object {
properties,

View File

@@ -216,20 +216,6 @@ async fn view_image_tool_attaches_local_image() -> anyhow::Result<()> {
let image_message =
find_image_message(&body).expect("pending input image message not included in request");
let content_items = image_message
.get("content")
.and_then(Value::as_array)
.expect("image message has content array");
assert_eq!(
content_items.len(),
1,
"view_image should inject only the image content item (no tag/label text)"
);
assert_eq!(
content_items[0].get("type").and_then(Value::as_str),
Some("input_image"),
"view_image should inject only an input_image content item"
);
let image_url = image_message
.get("content")
.and_then(Value::as_array)

View File

@@ -571,9 +571,6 @@ impl EventProcessor for EventProcessorWithHumanOutput {
EventMsg::ContextCompacted(_) => {
ts_msg!(self, "context compacted");
}
EventMsg::CollabInteraction(_) => {
// TODO(jif) handle collab tools.
}
EventMsg::ShutdownComplete => return CodexStatus::Shutdown,
EventMsg::WebSearchBegin(_)
| EventMsg::ExecApprovalRequest(_)

View File

@@ -306,7 +306,6 @@ async fn run_codex_tool_session_inner(
| EventMsg::ExitedReviewMode(_)
| EventMsg::ContextCompacted(_)
| EventMsg::ThreadRolledBack(_)
| EventMsg::CollabInteraction(_)
| EventMsg::DeprecationNotice(_) => {
// For now, we do not do anything extra for these
// events. Note that

View File

@@ -180,48 +180,6 @@ fn local_image_error_placeholder(
}
}
pub const VIEW_IMAGE_TOOL_NAME: &str = "view_image";
const IMAGE_OPEN_TAG: &str = "<image>";
const IMAGE_CLOSE_TAG: &str = "</image>";
const LOCAL_IMAGE_OPEN_TAG_PREFIX: &str = "<image name=";
const LOCAL_IMAGE_OPEN_TAG_SUFFIX: &str = ">";
const LOCAL_IMAGE_CLOSE_TAG: &str = IMAGE_CLOSE_TAG;
pub fn image_open_tag_text() -> String {
IMAGE_OPEN_TAG.to_string()
}
pub fn image_close_tag_text() -> String {
IMAGE_CLOSE_TAG.to_string()
}
pub fn local_image_label_text(label_number: usize) -> String {
format!("[Image #{label_number}]")
}
pub fn local_image_open_tag_text(label_number: usize) -> String {
let label = local_image_label_text(label_number);
format!("{LOCAL_IMAGE_OPEN_TAG_PREFIX}{label}{LOCAL_IMAGE_OPEN_TAG_SUFFIX}")
}
pub fn is_local_image_open_tag_text(text: &str) -> bool {
text.strip_prefix(LOCAL_IMAGE_OPEN_TAG_PREFIX)
.is_some_and(|rest| rest.ends_with(LOCAL_IMAGE_OPEN_TAG_SUFFIX))
}
pub fn is_local_image_close_tag_text(text: &str) -> bool {
is_image_close_tag_text(text)
}
pub fn is_image_open_tag_text(text: &str) -> bool {
text == IMAGE_OPEN_TAG
}
pub fn is_image_close_tag_text(text: &str) -> bool {
text == IMAGE_CLOSE_TAG
}
fn invalid_image_error_placeholder(
path: &std::path::Path,
error: impl std::fmt::Display,
@@ -245,53 +203,6 @@ fn unsupported_image_error_placeholder(path: &std::path::Path, mime: &str) -> Co
}
}
pub fn local_image_content_items_with_label_number(
path: &std::path::Path,
label_number: Option<usize>,
) -> Vec<ContentItem> {
match load_and_resize_to_fit(path) {
Ok(image) => {
let mut items = Vec::with_capacity(3);
if let Some(label_number) = label_number {
items.push(ContentItem::InputText {
text: local_image_open_tag_text(label_number),
});
}
items.push(ContentItem::InputImage {
image_url: image.into_data_url(),
});
if label_number.is_some() {
items.push(ContentItem::InputText {
text: LOCAL_IMAGE_CLOSE_TAG.to_string(),
});
}
items
}
Err(err) => {
if matches!(&err, ImageProcessingError::Read { .. }) {
vec![local_image_error_placeholder(path, &err)]
} else if err.is_invalid_image() {
vec![invalid_image_error_placeholder(path, &err)]
} else {
let Some(mime_guess) = mime_guess::from_path(path).first() else {
return vec![local_image_error_placeholder(
path,
"unsupported MIME type (unknown)",
)];
};
let mime = mime_guess.essence_str().to_owned();
if !mime.starts_with("image/") {
return vec![local_image_error_placeholder(
path,
format!("unsupported MIME type `{mime}`"),
)];
}
vec![unsupported_image_error_placeholder(path, &mime)]
}
}
}
}
impl From<ResponseInputItem> for ResponseItem {
fn from(item: ResponseInputItem) -> Self {
match item {
@@ -385,27 +296,41 @@ pub enum ReasoningItemContent {
impl From<Vec<UserInput>> for ResponseInputItem {
fn from(items: Vec<UserInput>) -> Self {
let mut image_index = 0;
Self::Message {
role: "user".to_string(),
content: items
.into_iter()
.flat_map(|c| match c {
UserInput::Text { text } => vec![ContentItem::InputText { text }],
UserInput::Image { image_url } => vec![
ContentItem::InputText {
text: image_open_tag_text(),
},
ContentItem::InputImage { image_url },
ContentItem::InputText {
text: image_close_tag_text(),
},
],
UserInput::LocalImage { path } => {
image_index += 1;
local_image_content_items_with_label_number(&path, Some(image_index))
}
UserInput::Skill { .. } => Vec::new(), // Skill bodies are injected later in core
.filter_map(|c| match c {
UserInput::Text { text } => Some(ContentItem::InputText { text }),
UserInput::Image { image_url } => Some(ContentItem::InputImage { image_url }),
UserInput::LocalImage { path } => match load_and_resize_to_fit(&path) {
Ok(image) => Some(ContentItem::InputImage {
image_url: image.into_data_url(),
}),
Err(err) => {
if matches!(&err, ImageProcessingError::Read { .. }) {
Some(local_image_error_placeholder(&path, &err))
} else if err.is_invalid_image() {
Some(invalid_image_error_placeholder(&path, &err))
} else {
let Some(mime_guess) = mime_guess::from_path(&path).first() else {
return Some(local_image_error_placeholder(
&path,
"unsupported MIME type (unknown)",
));
};
let mime = mime_guess.essence_str().to_owned();
if !mime.starts_with("image/") {
return Some(local_image_error_placeholder(
&path,
format!("unsupported MIME type `{mime}`"),
));
}
Some(unsupported_image_error_placeholder(&path, &mime))
}
}
},
UserInput::Skill { .. } => None, // Skill bodies are injected later in core
})
.collect::<Vec<ContentItem>>(),
}
@@ -845,33 +770,6 @@ mod tests {
Ok(())
}
#[test]
fn wraps_image_user_input_with_tags() -> Result<()> {
let image_url = "data:image/png;base64,abc".to_string();
let item = ResponseInputItem::from(vec![UserInput::Image {
image_url: image_url.clone(),
}]);
match item {
ResponseInputItem::Message { content, .. } => {
let expected = vec![
ContentItem::InputText {
text: image_open_tag_text(),
},
ContentItem::InputImage { image_url },
ContentItem::InputText {
text: image_close_tag_text(),
},
];
assert_eq!(content, expected);
}
other => panic!("expected message response but got {other:?}"),
}
Ok(())
}
#[test]
fn local_image_read_error_adds_placeholder() -> Result<()> {
let dir = tempdir()?;

View File

@@ -683,9 +683,6 @@ pub enum EventMsg {
AgentMessageContentDelta(AgentMessageContentDeltaEvent),
ReasoningContentDelta(ReasoningContentDeltaEvent),
ReasoningRawContentDelta(ReasoningRawContentDeltaEvent),
/// Collab interaction.
CollabInteraction(CollabInteractionEvent),
}
/// Agent lifecycle status, derived from emitted events.
@@ -702,7 +699,7 @@ pub enum AgentStatus {
Completed(Option<String>),
/// Agent encountered an error.
Errored(String),
/// Agent has been shutdown.
/// Agent has been shutdowned.
Shutdown,
/// Agent is not found.
NotFound,
@@ -1936,56 +1933,6 @@ pub enum TurnAbortReason {
ReviewEnded,
}
#[derive(Debug, Clone, Deserialize, Serialize, PartialEq, JsonSchema, TS)]
#[serde(rename_all = "snake_case")]
pub enum CollabInteractionEvent {
AgentSpawned {
/// Thread ID of the sender.
sender_id: ThreadId,
/// Thread ID of the newly spawned agent.
new_id: ThreadId,
/// Initial prompt sent to the agent. Can be empty to prevent CoT leaking at the
/// beginning.
prompt: String,
},
AgentInteraction {
/// Thread ID of the sender.
sender_id: ThreadId,
/// Thread ID of the receiver.
receiver_id: ThreadId,
/// Prompt sent from the sender to the receiver. Can be empty to prevent CoT
/// leaking at the beginning.
prompt: String,
},
WaitingBegin {
/// Thread ID of the sender.
sender_id: ThreadId,
/// Thread ID of the receiver.
receiver_id: ThreadId,
/// ID of the waiting call.
waiting_id: String,
},
WaitingEnd {
/// Thread ID of the sender.
sender_id: ThreadId,
/// Thread ID of the receiver.
receiver_id: ThreadId,
/// ID of the waiting call.
waiting_id: String,
/// Final status of the receiver agent reported to the sender agent.
status: AgentStatus,
},
Close {
/// Thread ID of the sender.
sender_id: ThreadId,
/// Thread ID of the receiver.
receiver_id: ThreadId,
/// Last known status of the receiver agent reported to the sender agent before
/// the close.
status: AgentStatus,
},
}
#[cfg(test)]
mod tests {
use super::*;

View File

@@ -46,7 +46,6 @@ use crate::style::user_message_style;
use codex_common::fuzzy_match::fuzzy_match;
use codex_protocol::custom_prompts::CustomPrompt;
use codex_protocol::custom_prompts::PROMPTS_CMD_PREFIX;
use codex_protocol::models::local_image_label_text;
use crate::app_event::AppEvent;
use crate::app_event_sender::AppEventSender;
@@ -60,7 +59,7 @@ use codex_core::skills::model::SkillMetadata;
use codex_file_search::FileMatch;
use std::cell::RefCell;
use std::collections::HashMap;
use std::collections::HashSet;
use std::path::Path;
use std::path::PathBuf;
use std::time::Duration;
use std::time::Instant;
@@ -275,12 +274,10 @@ impl ChatComposer {
// normalize_pasted_path already handles Windows → WSL path conversion,
// so we can directly try to read the image dimensions.
match image::image_dimensions(&path_buf) {
Ok((width, height)) => {
Ok((w, h)) => {
tracing::info!("OK: {pasted}");
tracing::debug!("image dimensions={}x{}", width, height);
let format = pasted_image_format(&path_buf);
tracing::debug!("attached image format={}", format.label());
self.attach_image(path_buf);
let format_label = pasted_image_format(&path_buf).label();
self.attach_image(path_buf, w, h, format_label);
true
}
Err(err) => {
@@ -411,9 +408,13 @@ impl ChatComposer {
}
/// Attempt to start a burst by retro-capturing recent chars before the cursor.
pub fn attach_image(&mut self, path: PathBuf) {
let image_number = self.attached_images.len() + 1;
let placeholder = local_image_label_text(image_number);
pub fn attach_image(&mut self, path: PathBuf, width: u32, height: u32, _format_label: &str) {
let file_label = path
.file_name()
.map(|name| name.to_string_lossy().into_owned())
.unwrap_or_else(|| "image".to_string());
let base_placeholder = format!("{file_label} {width}x{height}");
let placeholder = self.next_image_placeholder(&base_placeholder);
// Insert as an element to match large paste placeholder behavior:
// styled distinctly and treated atomically for cursor/mutations.
self.textarea.insert_element(&placeholder);
@@ -476,6 +477,22 @@ impl ChatComposer {
}
}
fn next_image_placeholder(&mut self, base: &str) -> String {
let text = self.textarea.text();
let mut suffix = 1;
loop {
let placeholder = if suffix == 1 {
format!("[{base}]")
} else {
format!("[{base} #{suffix}]")
};
if !text.contains(&placeholder) {
return placeholder;
}
suffix += 1;
}
}
pub(crate) fn insert_str(&mut self, text: &str) {
self.textarea.insert_str(text);
self.sync_popups();
@@ -801,43 +818,47 @@ impl ChatComposer {
if is_image {
// Determine dimensions; if that fails fall back to normal path insertion.
let path_buf = PathBuf::from(&sel_path);
match image::image_dimensions(&path_buf) {
Ok((width, height)) => {
tracing::debug!("selected image dimensions={}x{}", width, height);
// Remove the current @token (mirror logic from insert_selected_path without inserting text)
// using the flat text and byte-offset cursor API.
let cursor_offset = self.textarea.cursor();
let text = self.textarea.text();
// Clamp to a valid char boundary to avoid panics when slicing.
let safe_cursor = Self::clamp_to_char_boundary(text, cursor_offset);
let before_cursor = &text[..safe_cursor];
let after_cursor = &text[safe_cursor..];
if let Ok((w, h)) = image::image_dimensions(&path_buf) {
// Remove the current @token (mirror logic from insert_selected_path without inserting text)
// using the flat text and byte-offset cursor API.
let cursor_offset = self.textarea.cursor();
let text = self.textarea.text();
// Clamp to a valid char boundary to avoid panics when slicing.
let safe_cursor = Self::clamp_to_char_boundary(text, cursor_offset);
let before_cursor = &text[..safe_cursor];
let after_cursor = &text[safe_cursor..];
// Determine token boundaries in the full text.
let start_idx = before_cursor
.char_indices()
.rfind(|(_, c)| c.is_whitespace())
.map(|(idx, c)| idx + c.len_utf8())
.unwrap_or(0);
let end_rel_idx = after_cursor
.char_indices()
.find(|(_, c)| c.is_whitespace())
.map(|(idx, _)| idx)
.unwrap_or(after_cursor.len());
let end_idx = safe_cursor + end_rel_idx;
// Determine token boundaries in the full text.
let start_idx = before_cursor
.char_indices()
.rfind(|(_, c)| c.is_whitespace())
.map(|(idx, c)| idx + c.len_utf8())
.unwrap_or(0);
let end_rel_idx = after_cursor
.char_indices()
.find(|(_, c)| c.is_whitespace())
.map(|(idx, _)| idx)
.unwrap_or(after_cursor.len());
let end_idx = safe_cursor + end_rel_idx;
self.textarea.replace_range(start_idx..end_idx, "");
self.textarea.set_cursor(start_idx);
self.textarea.replace_range(start_idx..end_idx, "");
self.textarea.set_cursor(start_idx);
self.attach_image(path_buf);
// Add a trailing space to keep typing fluid.
self.textarea.insert_str(" ");
}
Err(err) => {
tracing::trace!("image dimensions lookup failed: {err}");
// Fallback to plain path insertion if metadata read fails.
self.insert_selected_path(&sel_path);
}
let format_label = match Path::new(&sel_path)
.extension()
.and_then(|e| e.to_str())
.map(str::to_ascii_lowercase)
{
Some(ext) if ext == "png" => "PNG",
Some(ext) if ext == "jpg" || ext == "jpeg" => "JPEG",
_ => "IMG",
};
self.attach_image(path_buf, w, h, format_label);
// Add a trailing space to keep typing fluid.
self.textarea.insert_str(" ");
} else {
// Fallback to plain path insertion if metadata read fails.
self.insert_selected_path(&sel_path);
}
} else {
// Non-image: inserting file path.
@@ -1443,29 +1464,20 @@ impl ChatComposer {
}
}
// Backspace at the start of an image placeholder should delete that placeholder (rather
// than deleting content before it). Do this without scanning the full text by consulting
// the textarea's element list.
if matches!(input.code, KeyCode::Backspace)
&& self.try_remove_image_element_at_cursor_start()
// For non-char inputs (or after flushing), handle normally.
// Special handling for backspace on placeholders
if let KeyEvent {
code: KeyCode::Backspace,
..
} = input
&& self.try_remove_any_placeholder_at_cursor()
{
return (InputResult::None, true);
}
// For non-char inputs (or after flushing), handle normally.
// Track element removals so we can drop any corresponding placeholders without scanning
// the full text. (Placeholders are atomic elements; when deleted, the element disappears.)
let elements_before = if self.pending_pastes.is_empty() && self.attached_images.is_empty() {
None
} else {
Some(self.textarea.element_payloads())
};
// Normal input handling
self.textarea.input(input);
if let Some(elements_before) = elements_before {
self.reconcile_deleted_elements(elements_before);
}
let text_after = self.textarea.text();
// Update paste-burst heuristic for plain Char (no Ctrl/Alt) events.
let crossterm::event::KeyEvent {
@@ -1487,69 +1499,176 @@ impl ChatComposer {
}
}
// Check if any placeholders were removed and remove their corresponding pending pastes
self.pending_pastes
.retain(|(placeholder, _)| text_after.contains(placeholder));
// Keep attached images in proportion to how many matching placeholders exist in the text.
// This handles duplicate placeholders that share the same visible label.
if !self.attached_images.is_empty() {
let mut needed: HashMap<String, usize> = HashMap::new();
for img in &self.attached_images {
needed
.entry(img.placeholder.clone())
.or_insert_with(|| text_after.matches(&img.placeholder).count());
}
let mut used: HashMap<String, usize> = HashMap::new();
let mut kept: Vec<AttachedImage> = Vec::with_capacity(self.attached_images.len());
for img in self.attached_images.drain(..) {
let total_needed = *needed.get(&img.placeholder).unwrap_or(&0);
let used_count = used.entry(img.placeholder.clone()).or_insert(0);
if *used_count < total_needed {
kept.push(img);
*used_count += 1;
}
}
self.attached_images = kept;
}
(InputResult::None, true)
}
fn try_remove_image_element_at_cursor_start(&mut self) -> bool {
if self.attached_images.is_empty() {
return false;
}
/// Attempts to remove an image or paste placeholder if the cursor is at the end of one.
/// Returns true if a placeholder was removed.
fn try_remove_any_placeholder_at_cursor(&mut self) -> bool {
// Clamp the cursor to a valid char boundary to avoid panics when slicing.
let text = self.textarea.text();
let p = Self::clamp_to_char_boundary(text, self.textarea.cursor());
let p = self.textarea.cursor();
let Some(payload) = self.textarea.element_payload_starting_at(p) else {
return false;
};
let Some(idx) = self
.attached_images
.iter()
.position(|img| img.placeholder == payload)
else {
return false;
};
self.textarea.replace_range(p..p + payload.len(), "");
self.attached_images.remove(idx);
self.relabel_attached_images_and_update_placeholders();
true
}
fn reconcile_deleted_elements(&mut self, elements_before: Vec<String>) {
let elements_after: HashSet<String> =
self.textarea.element_payloads().into_iter().collect();
let mut removed_any_image = false;
for removed in elements_before
.into_iter()
.filter(|payload| !elements_after.contains(payload))
{
self.pending_pastes.retain(|(ph, _)| ph != &removed);
if let Some(idx) = self
.attached_images
.iter()
.position(|img| img.placeholder == removed)
{
self.attached_images.remove(idx);
removed_any_image = true;
// Try image placeholders first
let mut out: Option<(usize, String)> = None;
// Detect if the cursor is at the end of any image placeholder.
// If duplicates exist, remove the specific occurrence's mapping.
for (i, img) in self.attached_images.iter().enumerate() {
let ph = &img.placeholder;
if p < ph.len() {
continue;
}
}
if removed_any_image {
self.relabel_attached_images_and_update_placeholders();
}
}
fn relabel_attached_images_and_update_placeholders(&mut self) {
for idx in 0..self.attached_images.len() {
let expected = local_image_label_text(idx + 1);
let current = self.attached_images[idx].placeholder.clone();
if current == expected {
let start = p - ph.len();
if text.get(start..p) != Some(ph.as_str()) {
continue;
}
self.attached_images[idx].placeholder = expected.clone();
let _renamed = self.textarea.replace_element_payload(&current, &expected);
// Count the number of occurrences of `ph` before `start`.
let mut occ_before = 0usize;
let mut search_pos = 0usize;
while search_pos < start {
let segment = match text.get(search_pos..start) {
Some(s) => s,
None => break,
};
if let Some(found) = segment.find(ph) {
occ_before += 1;
search_pos += found + ph.len();
} else {
break;
}
}
// Remove the occ_before-th attached image that shares this placeholder label.
out = if let Some((remove_idx, _)) = self
.attached_images
.iter()
.enumerate()
.filter(|(_, img2)| img2.placeholder == *ph)
.nth(occ_before)
{
Some((remove_idx, ph.clone()))
} else {
Some((i, ph.clone()))
};
break;
}
if let Some((idx, placeholder)) = out {
self.textarea.replace_range(p - placeholder.len()..p, "");
self.attached_images.remove(idx);
return true;
}
// Also handle when the cursor is at the START of an image placeholder.
// let result = 'out: {
let out: Option<(usize, String)> = 'out: {
for (i, img) in self.attached_images.iter().enumerate() {
let ph = &img.placeholder;
if p + ph.len() > text.len() {
continue;
}
if text.get(p..p + ph.len()) != Some(ph.as_str()) {
continue;
}
// Count occurrences of `ph` before `p`.
let mut occ_before = 0usize;
let mut search_pos = 0usize;
while search_pos < p {
let segment = match text.get(search_pos..p) {
Some(s) => s,
None => break 'out None,
};
if let Some(found) = segment.find(ph) {
occ_before += 1;
search_pos += found + ph.len();
} else {
break 'out None;
}
}
if let Some((remove_idx, _)) = self
.attached_images
.iter()
.enumerate()
.filter(|(_, img2)| img2.placeholder == *ph)
.nth(occ_before)
{
break 'out Some((remove_idx, ph.clone()));
} else {
break 'out Some((i, ph.clone()));
}
}
None
};
if let Some((idx, placeholder)) = out {
self.textarea.replace_range(p..p + placeholder.len(), "");
self.attached_images.remove(idx);
return true;
}
// Then try pasted-content placeholders
if let Some(placeholder) = self.pending_pastes.iter().find_map(|(ph, _)| {
if p < ph.len() {
return None;
}
let start = p - ph.len();
if text.get(start..p) == Some(ph.as_str()) {
Some(ph.clone())
} else {
None
}
}) {
self.textarea.replace_range(p - placeholder.len()..p, "");
self.pending_pastes.retain(|(ph, _)| ph != &placeholder);
return true;
}
// Also handle when the cursor is at the START of a pasted-content placeholder.
if let Some(placeholder) = self.pending_pastes.iter().find_map(|(ph, _)| {
if p + ph.len() > text.len() {
return None;
}
if text.get(p..p + ph.len()) == Some(ph.as_str()) {
Some(ph.clone())
} else {
None
}
}) {
self.textarea.replace_range(p..p + placeholder.len(), "");
self.pending_pastes.retain(|(ph, _)| ph != &placeholder);
return true;
}
false
}
fn handle_shortcut_overlay_key(&mut self, key_event: &KeyEvent) -> bool {
@@ -3358,12 +3477,12 @@ mod tests {
false,
);
let path = PathBuf::from("/tmp/image1.png");
composer.attach_image(path.clone());
composer.attach_image(path.clone(), 32, 16, "PNG");
composer.handle_paste(" hi".into());
let (result, _) =
composer.handle_key_event(KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE));
match result {
InputResult::Submitted(text) => assert_eq!(text, "[Image #1] hi"),
InputResult::Submitted(text) => assert_eq!(text, "[image1.png 32x16] hi"),
_ => panic!("expected Submitted"),
}
let imgs = composer.take_recent_submission_images();
@@ -3382,11 +3501,11 @@ mod tests {
false,
);
let path = PathBuf::from("/tmp/image2.png");
composer.attach_image(path.clone());
composer.attach_image(path.clone(), 10, 5, "PNG");
let (result, _) =
composer.handle_key_event(KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE));
match result {
InputResult::Submitted(text) => assert_eq!(text, "[Image #1]"),
InputResult::Submitted(text) => assert_eq!(text, "[image2.png 10x5]"),
_ => panic!("expected Submitted"),
}
let imgs = composer.take_recent_submission_images();
@@ -3407,15 +3526,21 @@ mod tests {
false,
);
let path = PathBuf::from("/tmp/image_dup.png");
composer.attach_image(path.clone());
composer.attach_image(path.clone(), 10, 5, "PNG");
composer.handle_paste(" ".into());
composer.attach_image(path);
composer.attach_image(path, 10, 5, "PNG");
let text = composer.textarea.text().to_string();
assert!(text.contains("[Image #1]"));
assert!(text.contains("[Image #2]"));
assert_eq!(composer.attached_images[0].placeholder, "[Image #1]");
assert_eq!(composer.attached_images[1].placeholder, "[Image #2]");
assert!(text.contains("[image_dup.png 10x5]"));
assert!(text.contains("[image_dup.png 10x5 #2]"));
assert_eq!(
composer.attached_images[0].placeholder,
"[image_dup.png 10x5]"
);
assert_eq!(
composer.attached_images[1].placeholder,
"[image_dup.png 10x5 #2]"
);
}
#[test]
@@ -3430,7 +3555,7 @@ mod tests {
false,
);
let path = PathBuf::from("/tmp/image3.png");
composer.attach_image(path.clone());
composer.attach_image(path.clone(), 20, 10, "PNG");
let placeholder = composer.attached_images[0].placeholder.clone();
// Case 1: backspace at end
@@ -3441,7 +3566,7 @@ mod tests {
// Re-add and test backspace in middle: should break the placeholder string
// and drop the image mapping (same as text placeholder behavior).
composer.attach_image(path);
composer.attach_image(path, 20, 10, "PNG");
let placeholder2 = composer.attached_images[0].placeholder.clone();
// Move cursor to roughly middle of placeholder
if let Some(start_pos) = composer.textarea.text().find(&placeholder2) {
@@ -3473,7 +3598,7 @@ mod tests {
// Insert an image placeholder at the start
let path = PathBuf::from("/tmp/image_multibyte.png");
composer.attach_image(path);
composer.attach_image(path, 10, 5, "PNG");
// Add multibyte text after the placeholder
composer.textarea.insert_str("日本語");
@@ -3482,11 +3607,16 @@ mod tests {
composer.handle_key_event(KeyEvent::new(KeyCode::Backspace, KeyModifiers::NONE));
assert_eq!(composer.attached_images.len(), 1);
assert!(composer.textarea.text().starts_with("[Image #1]"));
assert!(
composer
.textarea
.text()
.starts_with("[image_multibyte.png 10x5]")
);
}
#[test]
fn deleting_one_of_duplicate_image_placeholders_removes_one_entry() {
fn deleting_one_of_duplicate_image_placeholders_removes_matching_entry() {
let (tx, _rx) = unbounded_channel::<AppEvent>();
let sender = AppEventSender::new(tx);
let mut composer = ChatComposer::new(
@@ -3500,10 +3630,10 @@ mod tests {
let path1 = PathBuf::from("/tmp/image_dup1.png");
let path2 = PathBuf::from("/tmp/image_dup2.png");
composer.attach_image(path1);
composer.attach_image(path1, 10, 5, "PNG");
// separate placeholders with a space for clarity
composer.handle_paste(" ".into());
composer.attach_image(path2.clone());
composer.attach_image(path2.clone(), 10, 5, "PNG");
let placeholder1 = composer.attached_images[0].placeholder.clone();
let placeholder2 = composer.attached_images[1].placeholder.clone();
@@ -3516,67 +3646,26 @@ mod tests {
composer.handle_key_event(KeyEvent::new(KeyCode::Backspace, KeyModifiers::NONE));
let new_text = composer.textarea.text().to_string();
assert_eq!(
1,
new_text.matches(&placeholder1).count(),
"one placeholder remains after deletion"
);
assert_eq!(
0,
new_text.matches(&placeholder2).count(),
"second placeholder was relabeled"
new_text.matches(&placeholder1).count(),
"first placeholder removed"
);
assert_eq!(
1,
new_text.matches("[Image #1]").count(),
"remaining placeholder relabeled to #1"
new_text.matches(&placeholder2).count(),
"second placeholder remains"
);
assert_eq!(
vec![AttachedImage {
path: path2,
placeholder: "[Image #1]".to_string()
placeholder: "[image_dup2.png 10x5]".to_string()
}],
composer.attached_images,
"one image mapping remains"
);
}
#[test]
fn deleting_first_text_element_renumbers_following_text_element() {
use crossterm::event::KeyCode;
use crossterm::event::KeyEvent;
use crossterm::event::KeyModifiers;
let (tx, _rx) = unbounded_channel::<AppEvent>();
let sender = AppEventSender::new(tx);
let mut composer = ChatComposer::new(
true,
sender,
false,
"Ask Codex to do anything".to_string(),
false,
);
let path1 = PathBuf::from("/tmp/image_first.png");
let path2 = PathBuf::from("/tmp/image_second.png");
// Insert two adjacent atomic elements.
composer.attach_image(path1);
composer.attach_image(path2.clone());
assert_eq!(composer.textarea.text(), "[Image #1][Image #2]");
assert_eq!(composer.attached_images.len(), 2);
// Delete the first element using normal textarea editing (Delete at cursor start).
composer.textarea.set_cursor(0);
composer.handle_key_event(KeyEvent::new(KeyCode::Delete, KeyModifiers::NONE));
// Remaining image should be renumbered and the textarea element updated.
assert_eq!(composer.attached_images.len(), 1);
assert_eq!(composer.attached_images[0].path, path2);
assert_eq!(composer.attached_images[0].placeholder, "[Image #1]");
assert_eq!(composer.textarea.text(), "[Image #1]");
}
#[test]
fn pasting_filepath_attaches_image() {
let tmp = tempdir().expect("create TempDir");
@@ -3597,7 +3686,12 @@ mod tests {
let needs_redraw = composer.handle_paste(tmp_path.to_string_lossy().to_string());
assert!(needs_redraw);
assert!(composer.textarea.text().starts_with("[Image #1] "));
assert!(
composer
.textarea
.text()
.starts_with("[codex_tui_test_paste_image.png 3x2] ")
);
let imgs = composer.take_recent_submission_images();
assert_eq!(imgs, vec![tmp_path]);

View File

@@ -542,9 +542,16 @@ impl BottomPane {
self.request_redraw();
}
pub(crate) fn attach_image(&mut self, path: PathBuf) {
pub(crate) fn attach_image(
&mut self,
path: PathBuf,
width: u32,
height: u32,
format_label: &str,
) {
if self.view_stack.is_empty() {
self.composer.attach_image(path);
self.composer
.attach_image(path, width, height, format_label);
self.request_redraw();
}
}

View File

@@ -715,88 +715,6 @@ impl TextArea {
// ===== Text elements support =====
pub fn element_payloads(&self) -> Vec<String> {
self.elements
.iter()
.filter_map(|e| self.text.get(e.range.clone()).map(str::to_string))
.collect()
}
pub fn element_payload_starting_at(&self, pos: usize) -> Option<String> {
let pos = pos.min(self.text.len());
let elem = self.elements.iter().find(|e| e.range.start == pos)?;
self.text.get(elem.range.clone()).map(str::to_string)
}
/// Renames a single text element in-place, keeping it atomic.
///
/// This is intended for cases where the element payload is an identifier (e.g. a placeholder)
/// that must be updated without converting the element back into normal text.
pub fn replace_element_payload(&mut self, old: &str, new: &str) -> bool {
let Some(idx) = self
.elements
.iter()
.position(|e| self.text.get(e.range.clone()) == Some(old))
else {
return false;
};
let range = self.elements[idx].range.clone();
let start = range.start;
let end = range.end;
if start > end || end > self.text.len() {
return false;
}
let removed_len = end - start;
let inserted_len = new.len();
let diff = inserted_len as isize - removed_len as isize;
self.text.replace_range(range, new);
self.wrap_cache.replace(None);
self.preferred_col = None;
// Update the modified element's range.
self.elements[idx].range = start..(start + inserted_len);
// Shift element ranges that occur after the replaced element.
if diff != 0 {
for (j, e) in self.elements.iter_mut().enumerate() {
if j == idx {
continue;
}
if e.range.end <= start {
continue;
}
if e.range.start >= end {
e.range.start = ((e.range.start as isize) + diff) as usize;
e.range.end = ((e.range.end as isize) + diff) as usize;
continue;
}
// Elements should not partially overlap each other; degrade gracefully by
// snapping anything intersecting the replaced range to the new bounds.
e.range.start = start.min(e.range.start);
e.range.end = (start + inserted_len).max(e.range.end.saturating_add_signed(diff));
}
}
// Update the cursor position to account for the edit.
self.cursor_pos = if self.cursor_pos < start {
self.cursor_pos
} else if self.cursor_pos <= end {
start + inserted_len
} else {
((self.cursor_pos as isize) + diff) as usize
};
self.cursor_pos = self.clamp_pos_to_nearest_boundary(self.cursor_pos);
// Keep element ordering deterministic.
self.elements.sort_by_key(|e| e.range.start);
true
}
pub fn insert_element(&mut self, text: &str) {
let start = self.clamp_pos_for_insertion(self.cursor_pos);
self.insert_str_at(start, text);

View File

@@ -24,7 +24,6 @@ use codex_core::protocol::AgentReasoningRawContentDeltaEvent;
use codex_core::protocol::AgentReasoningRawContentEvent;
use codex_core::protocol::ApplyPatchApprovalRequestEvent;
use codex_core::protocol::BackgroundEventEvent;
use codex_core::protocol::CollabInteractionEvent;
use codex_core::protocol::CreditsSnapshot;
use codex_core::protocol::DeprecationNoticeEvent;
use codex_core::protocol::ErrorEvent;
@@ -103,7 +102,6 @@ use crate::bottom_pane::SelectionViewParams;
use crate::bottom_pane::custom_prompt_view::CustomPromptView;
use crate::bottom_pane::popup_consts::standard_popup_hint_line;
use crate::clipboard_paste::paste_image_to_temp_png;
use crate::collab_event_cell;
use crate::diff_render::display_path_for;
use crate::exec_cell::CommandOutput;
use crate::exec_cell::ExecCell;
@@ -1075,11 +1073,6 @@ impl ChatWidget {
self.set_status_header(message);
}
fn on_collab_interaction(&mut self, event: CollabInteractionEvent) {
self.add_to_history(collab_event_cell::new_collab_interaction(event));
self.request_redraw();
}
fn on_undo_started(&mut self, event: UndoStartedEvent) {
self.bottom_pane.ensure_status_indicator();
self.bottom_pane.set_interrupt_hint_visible(false);
@@ -1600,13 +1593,12 @@ impl ChatWidget {
{
match paste_image_to_temp_png() {
Ok((path, info)) => {
tracing::debug!(
"pasted image size={}x{} format={}",
self.attach_image(
path,
info.width,
info.height,
info.encoded_format.label()
info.encoded_format.label(),
);
self.attach_image(path);
}
Err(err) => {
tracing::warn!("failed to paste image: {err}");
@@ -1659,9 +1651,18 @@ impl ChatWidget {
}
}
pub(crate) fn attach_image(&mut self, path: PathBuf) {
tracing::info!("attach_image path={path:?}");
self.bottom_pane.attach_image(path);
pub(crate) fn attach_image(
&mut self,
path: PathBuf,
width: u32,
height: u32,
format_label: &str,
) {
tracing::info!(
"attach_image path={path:?} width={width} height={height} format={format_label}",
);
self.bottom_pane
.attach_image(path, width, height, format_label);
self.request_redraw();
}
@@ -2011,14 +2012,14 @@ impl ChatWidget {
return;
}
for path in image_paths {
items.push(UserInput::LocalImage { path });
}
if !text.is_empty() {
items.push(UserInput::Text { text: text.clone() });
}
for path in image_paths {
items.push(UserInput::LocalImage { path });
}
if let Some(skills) = self.bottom_pane.skills() {
let skill_mentions = find_skill_mentions(&text, skills);
for skill in skill_mentions {
@@ -2189,7 +2190,6 @@ impl ChatWidget {
}
EventMsg::ExitedReviewMode(review) => self.on_exited_review_mode(review),
EventMsg::ContextCompacted(_) => self.on_agent_message("Context compacted".to_owned()),
EventMsg::CollabInteraction(event) => self.on_collab_interaction(event),
EventMsg::ThreadRolledBack(_) => {}
EventMsg::RawResponseItem(_)
| EventMsg::ItemStarted(_)

View File

@@ -1114,8 +1114,8 @@ async fn ctrl_c_cleared_prompt_is_recoverable_via_history() {
chat.bottom_pane.insert_str("draft message ");
chat.bottom_pane
.attach_image(PathBuf::from("/tmp/preview.png"));
let placeholder = "[Image #1]";
.attach_image(PathBuf::from("/tmp/preview.png"), 24, 42, "png");
let placeholder = "[preview.png 24x42]";
assert!(
chat.bottom_pane.composer_text().ends_with(placeholder),
"expected placeholder {placeholder:?} in composer text"

View File

@@ -1,140 +0,0 @@
use codex_core::protocol::AgentStatus;
use codex_core::protocol::CollabInteractionEvent;
use ratatui::style::Stylize;
use ratatui::text::Line;
use crate::history_cell::HistoryCell;
use crate::text_formatting::truncate_text;
use crate::wrapping::RtOptions;
use crate::wrapping::word_wrap_lines;
const COLLAB_PROMPT_MAX_GRAPHEMES: usize = 120;
#[derive(Debug)]
pub(crate) struct CollabInteractionCell {
summary: Line<'static>,
detail: Option<Line<'static>>,
}
impl CollabInteractionCell {
fn new(summary: Line<'static>, detail: Option<Line<'static>>) -> Self {
Self { summary, detail }
}
}
impl HistoryCell for CollabInteractionCell {
fn display_lines(&self, width: u16) -> Vec<Line<'static>> {
let wrap_width = width.max(1) as usize;
let mut lines = word_wrap_lines(
std::iter::once(self.summary.clone()),
RtOptions::new(wrap_width)
.initial_indent("".dim().into())
.subsequent_indent(" ".into()),
);
if let Some(detail) = &self.detail {
let detail_lines = word_wrap_lines(
std::iter::once(detail.clone()),
RtOptions::new(wrap_width)
.initial_indent("".dim().into())
.subsequent_indent(" ".into()),
);
lines.extend(detail_lines);
}
lines
}
}
fn collab_status_label(status: &AgentStatus) -> String {
match status {
AgentStatus::PendingInit => "pending init".to_string(),
AgentStatus::Running => "running".to_string(),
AgentStatus::Completed(message) => format!("completed: {message:?}"),
AgentStatus::Errored(_) => "errored".to_string(),
AgentStatus::Shutdown => "shutdown".to_string(),
AgentStatus::NotFound => "not found".to_string(),
}
}
fn collab_detail_line(label: &str, message: &str) -> Option<Line<'static>> {
let trimmed = message.trim();
if trimmed.is_empty() {
return None;
}
let collapsed = trimmed
.lines()
.map(str::trim)
.filter(|line| !line.is_empty())
.collect::<Vec<_>>()
.join(" ");
let truncated = truncate_text(&collapsed, COLLAB_PROMPT_MAX_GRAPHEMES);
let label = format!("{label}: ");
Some(Line::from(vec![label.dim(), truncated.into()]))
}
pub(crate) fn new_collab_interaction(event: CollabInteractionEvent) -> CollabInteractionCell {
let (summary, detail) = match event {
CollabInteractionEvent::AgentSpawned { new_id, prompt, .. } => {
let summary = Line::from(vec![
"Spawned agent".bold(),
" ".into(),
new_id.to_string().dim(),
]);
let detail = collab_detail_line("Prompt", &prompt);
(summary, detail)
}
CollabInteractionEvent::AgentInteraction {
receiver_id,
prompt,
..
} => {
let summary = Line::from(vec![
"Sent to agent".bold(),
" ".into(),
receiver_id.to_string().dim(),
]);
let detail = collab_detail_line("Message", &prompt);
(summary, detail)
}
CollabInteractionEvent::WaitingBegin { receiver_id, .. } => {
let summary = Line::from(vec![
"Waiting on agent".bold(),
" ".into(),
receiver_id.to_string().dim(),
]);
(summary, None)
}
CollabInteractionEvent::WaitingEnd {
receiver_id,
status,
..
} => {
let summary = Line::from(vec![
"Wait ended for agent".bold(),
" ".into(),
receiver_id.to_string().dim(),
" · ".dim(),
collab_status_label(&status).dim(),
]);
(summary, None)
}
CollabInteractionEvent::Close {
receiver_id,
status,
..
} => {
let summary = Line::from(vec![
"Closed agent".bold(),
" ".into(),
receiver_id.to_string().dim(),
" · ".dim(),
collab_status_label(&status).dim(),
]);
(summary, None)
}
};
CollabInteractionCell::new(summary, detail)
}

View File

@@ -43,7 +43,6 @@ mod bottom_pane;
mod chatwidget;
mod cli;
mod clipboard_paste;
mod collab_event_cell;
mod color;
pub mod custom_terminal;
mod diff_render;

View File

@@ -49,7 +49,6 @@ use crate::style::user_message_style;
use codex_common::fuzzy_match::fuzzy_match;
use codex_protocol::custom_prompts::CustomPrompt;
use codex_protocol::custom_prompts::PROMPTS_CMD_PREFIX;
use codex_protocol::models::local_image_label_text;
use crate::app_event::AppEvent;
use crate::app_event_sender::AppEventSender;
@@ -63,7 +62,7 @@ use codex_core::skills::model::SkillMetadata;
use codex_file_search::FileMatch;
use std::cell::RefCell;
use std::collections::HashMap;
use std::collections::HashSet;
use std::path::Path;
use std::path::PathBuf;
use std::time::Duration;
use std::time::Instant;
@@ -288,12 +287,10 @@ impl ChatComposer {
// normalize_pasted_path already handles Windows → WSL path conversion,
// so we can directly try to read the image dimensions.
match image::image_dimensions(&path_buf) {
Ok((width, height)) => {
Ok((w, h)) => {
tracing::info!("OK: {pasted}");
tracing::debug!("image dimensions={}x{}", width, height);
let format = pasted_image_format(&path_buf);
tracing::debug!("attached image format={}", format.label());
self.attach_image(path_buf);
let format_label = pasted_image_format(&path_buf).label();
self.attach_image(path_buf, w, h, format_label);
true
}
Err(err) => {
@@ -345,9 +342,12 @@ impl ChatComposer {
}
/// Attempt to start a burst by retro-capturing recent chars before the cursor.
pub fn attach_image(&mut self, path: PathBuf) {
let image_number = self.attached_images.len() + 1;
let placeholder = local_image_label_text(image_number);
pub fn attach_image(&mut self, path: PathBuf, width: u32, height: u32, _format_label: &str) {
let file_label = path
.file_name()
.map(|name| name.to_string_lossy().into_owned())
.unwrap_or_else(|| "image".to_string());
let placeholder = format!("[{file_label} {width}x{height}]");
// Insert as an element to match large paste placeholder behavior:
// styled distinctly and treated atomically for cursor/mutations.
self.textarea.insert_element(&placeholder);
@@ -735,43 +735,47 @@ impl ChatComposer {
if is_image {
// Determine dimensions; if that fails fall back to normal path insertion.
let path_buf = PathBuf::from(&sel_path);
match image::image_dimensions(&path_buf) {
Ok((width, height)) => {
tracing::debug!("selected image dimensions={}x{}", width, height);
// Remove the current @token (mirror logic from insert_selected_path without inserting text)
// using the flat text and byte-offset cursor API.
let cursor_offset = self.textarea.cursor();
let text = self.textarea.text();
// Clamp to a valid char boundary to avoid panics when slicing.
let safe_cursor = Self::clamp_to_char_boundary(text, cursor_offset);
let before_cursor = &text[..safe_cursor];
let after_cursor = &text[safe_cursor..];
if let Ok((w, h)) = image::image_dimensions(&path_buf) {
// Remove the current @token (mirror logic from insert_selected_path without inserting text)
// using the flat text and byte-offset cursor API.
let cursor_offset = self.textarea.cursor();
let text = self.textarea.text();
// Clamp to a valid char boundary to avoid panics when slicing.
let safe_cursor = Self::clamp_to_char_boundary(text, cursor_offset);
let before_cursor = &text[..safe_cursor];
let after_cursor = &text[safe_cursor..];
// Determine token boundaries in the full text.
let start_idx = before_cursor
.char_indices()
.rfind(|(_, c)| c.is_whitespace())
.map(|(idx, c)| idx + c.len_utf8())
.unwrap_or(0);
let end_rel_idx = after_cursor
.char_indices()
.find(|(_, c)| c.is_whitespace())
.map(|(idx, _)| idx)
.unwrap_or(after_cursor.len());
let end_idx = safe_cursor + end_rel_idx;
// Determine token boundaries in the full text.
let start_idx = before_cursor
.char_indices()
.rfind(|(_, c)| c.is_whitespace())
.map(|(idx, c)| idx + c.len_utf8())
.unwrap_or(0);
let end_rel_idx = after_cursor
.char_indices()
.find(|(_, c)| c.is_whitespace())
.map(|(idx, _)| idx)
.unwrap_or(after_cursor.len());
let end_idx = safe_cursor + end_rel_idx;
self.textarea.replace_range(start_idx..end_idx, "");
self.textarea.set_cursor(start_idx);
self.textarea.replace_range(start_idx..end_idx, "");
self.textarea.set_cursor(start_idx);
self.attach_image(path_buf);
// Add a trailing space to keep typing fluid.
self.textarea.insert_str(" ");
}
Err(err) => {
tracing::trace!("image dimensions lookup failed: {err}");
// Fallback to plain path insertion if metadata read fails.
self.insert_selected_path(&sel_path);
}
let format_label = match Path::new(&sel_path)
.extension()
.and_then(|e| e.to_str())
.map(str::to_ascii_lowercase)
{
Some(ext) if ext == "png" => "PNG",
Some(ext) if ext == "jpg" || ext == "jpeg" => "JPEG",
_ => "IMG",
};
self.attach_image(path_buf, w, h, format_label);
// Add a trailing space to keep typing fluid.
self.textarea.insert_str(" ");
} else {
// Fallback to plain path insertion if metadata read fails.
self.insert_selected_path(&sel_path);
}
} else {
// Non-image: inserting file path.
@@ -1377,28 +1381,20 @@ impl ChatComposer {
}
}
// Backspace at the start of an image placeholder should delete that placeholder (rather
// than deleting content before it). Do this without scanning the full text by consulting
// the textarea's element list.
if matches!(input.code, KeyCode::Backspace)
&& self.try_remove_image_element_at_cursor_start()
// For non-char inputs (or after flushing), handle normally.
// Special handling for backspace on placeholders
if let KeyEvent {
code: KeyCode::Backspace,
..
} = input
&& self.try_remove_any_placeholder_at_cursor()
{
return (InputResult::None, true);
}
// Track element removals so we can drop any corresponding placeholders without scanning
// the full text. (Placeholders are atomic elements; when deleted, the element disappears.)
let elements_before = if self.pending_pastes.is_empty() && self.attached_images.is_empty() {
None
} else {
Some(self.textarea.element_payloads())
};
// Normal input handling
self.textarea.input(input);
if let Some(elements_before) = elements_before {
self.reconcile_deleted_elements(elements_before);
}
let text_after = self.textarea.text();
// Update paste-burst heuristic for plain Char (no Ctrl/Alt) events.
let crossterm::event::KeyEvent {
@@ -1420,69 +1416,176 @@ impl ChatComposer {
}
}
// Check if any placeholders were removed and remove their corresponding pending pastes
self.pending_pastes
.retain(|(placeholder, _)| text_after.contains(placeholder));
// Keep attached images in proportion to how many matching placeholders exist in the text.
// This handles duplicate placeholders that share the same visible label.
if !self.attached_images.is_empty() {
let mut needed: HashMap<String, usize> = HashMap::new();
for img in &self.attached_images {
needed
.entry(img.placeholder.clone())
.or_insert_with(|| text_after.matches(&img.placeholder).count());
}
let mut used: HashMap<String, usize> = HashMap::new();
let mut kept: Vec<AttachedImage> = Vec::with_capacity(self.attached_images.len());
for img in self.attached_images.drain(..) {
let total_needed = *needed.get(&img.placeholder).unwrap_or(&0);
let used_count = used.entry(img.placeholder.clone()).or_insert(0);
if *used_count < total_needed {
kept.push(img);
*used_count += 1;
}
}
self.attached_images = kept;
}
(InputResult::None, true)
}
fn try_remove_image_element_at_cursor_start(&mut self) -> bool {
if self.attached_images.is_empty() {
return false;
}
/// Attempts to remove an image or paste placeholder if the cursor is at the end of one.
/// Returns true if a placeholder was removed.
fn try_remove_any_placeholder_at_cursor(&mut self) -> bool {
// Clamp the cursor to a valid char boundary to avoid panics when slicing.
let text = self.textarea.text();
let p = Self::clamp_to_char_boundary(text, self.textarea.cursor());
let p = self.textarea.cursor();
let Some(payload) = self.textarea.element_payload_starting_at(p) else {
return false;
};
let Some(idx) = self
.attached_images
.iter()
.position(|img| img.placeholder == payload)
else {
return false;
};
self.textarea.replace_range(p..p + payload.len(), "");
self.attached_images.remove(idx);
self.relabel_attached_images_and_update_placeholders();
true
}
fn reconcile_deleted_elements(&mut self, elements_before: Vec<String>) {
let elements_after: HashSet<String> =
self.textarea.element_payloads().into_iter().collect();
let mut removed_any_image = false;
for removed in elements_before
.into_iter()
.filter(|payload| !elements_after.contains(payload))
{
self.pending_pastes.retain(|(ph, _)| ph != &removed);
if let Some(idx) = self
.attached_images
.iter()
.position(|img| img.placeholder == removed)
{
self.attached_images.remove(idx);
removed_any_image = true;
// Try image placeholders first
let mut out: Option<(usize, String)> = None;
// Detect if the cursor is at the end of any image placeholder.
// If duplicates exist, remove the specific occurrence's mapping.
for (i, img) in self.attached_images.iter().enumerate() {
let ph = &img.placeholder;
if p < ph.len() {
continue;
}
}
if removed_any_image {
self.relabel_attached_images_and_update_placeholders();
}
}
fn relabel_attached_images_and_update_placeholders(&mut self) {
for idx in 0..self.attached_images.len() {
let expected = local_image_label_text(idx + 1);
let current = self.attached_images[idx].placeholder.clone();
if current == expected {
let start = p - ph.len();
if text.get(start..p) != Some(ph.as_str()) {
continue;
}
self.attached_images[idx].placeholder = expected.clone();
let _renamed = self.textarea.replace_element_payload(&current, &expected);
// Count the number of occurrences of `ph` before `start`.
let mut occ_before = 0usize;
let mut search_pos = 0usize;
while search_pos < start {
let segment = match text.get(search_pos..start) {
Some(s) => s,
None => break,
};
if let Some(found) = segment.find(ph) {
occ_before += 1;
search_pos += found + ph.len();
} else {
break;
}
}
// Remove the occ_before-th attached image that shares this placeholder label.
out = if let Some((remove_idx, _)) = self
.attached_images
.iter()
.enumerate()
.filter(|(_, img2)| img2.placeholder == *ph)
.nth(occ_before)
{
Some((remove_idx, ph.clone()))
} else {
Some((i, ph.clone()))
};
break;
}
if let Some((idx, placeholder)) = out {
self.textarea.replace_range(p - placeholder.len()..p, "");
self.attached_images.remove(idx);
return true;
}
// Also handle when the cursor is at the START of an image placeholder.
// let result = 'out: {
let out: Option<(usize, String)> = 'out: {
for (i, img) in self.attached_images.iter().enumerate() {
let ph = &img.placeholder;
if p + ph.len() > text.len() {
continue;
}
if text.get(p..p + ph.len()) != Some(ph.as_str()) {
continue;
}
// Count occurrences of `ph` before `p`.
let mut occ_before = 0usize;
let mut search_pos = 0usize;
while search_pos < p {
let segment = match text.get(search_pos..p) {
Some(s) => s,
None => break 'out None,
};
if let Some(found) = segment.find(ph) {
occ_before += 1;
search_pos += found + ph.len();
} else {
break 'out None;
}
}
if let Some((remove_idx, _)) = self
.attached_images
.iter()
.enumerate()
.filter(|(_, img2)| img2.placeholder == *ph)
.nth(occ_before)
{
break 'out Some((remove_idx, ph.clone()));
} else {
break 'out Some((i, ph.clone()));
}
}
None
};
if let Some((idx, placeholder)) = out {
self.textarea.replace_range(p..p + placeholder.len(), "");
self.attached_images.remove(idx);
return true;
}
// Then try pasted-content placeholders
if let Some(placeholder) = self.pending_pastes.iter().find_map(|(ph, _)| {
if p < ph.len() {
return None;
}
let start = p - ph.len();
if text.get(start..p) == Some(ph.as_str()) {
Some(ph.clone())
} else {
None
}
}) {
self.textarea.replace_range(p - placeholder.len()..p, "");
self.pending_pastes.retain(|(ph, _)| ph != &placeholder);
return true;
}
// Also handle when the cursor is at the START of a pasted-content placeholder.
if let Some(placeholder) = self.pending_pastes.iter().find_map(|(ph, _)| {
if p + ph.len() > text.len() {
return None;
}
if text.get(p..p + ph.len()) == Some(ph.as_str()) {
Some(ph.clone())
} else {
None
}
}) {
self.textarea.replace_range(p..p + placeholder.len(), "");
self.pending_pastes.retain(|(ph, _)| ph != &placeholder);
return true;
}
false
}
fn handle_shortcut_overlay_key(&mut self, key_event: &KeyEvent) -> bool {
@@ -3305,12 +3408,12 @@ mod tests {
false,
);
let path = PathBuf::from("/tmp/image1.png");
composer.attach_image(path.clone());
composer.attach_image(path.clone(), 32, 16, "PNG");
composer.handle_paste(" hi".into());
let (result, _) =
composer.handle_key_event(KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE));
match result {
InputResult::Submitted(text) => assert_eq!(text, "[Image #1] hi"),
InputResult::Submitted(text) => assert_eq!(text, "[image1.png 32x16] hi"),
_ => panic!("expected Submitted"),
}
let imgs = composer.take_recent_submission_images();
@@ -3329,11 +3432,11 @@ mod tests {
false,
);
let path = PathBuf::from("/tmp/image2.png");
composer.attach_image(path.clone());
composer.attach_image(path.clone(), 10, 5, "PNG");
let (result, _) =
composer.handle_key_event(KeyEvent::new(KeyCode::Enter, KeyModifiers::NONE));
match result {
InputResult::Submitted(text) => assert_eq!(text, "[Image #1]"),
InputResult::Submitted(text) => assert_eq!(text, "[image2.png 10x5]"),
_ => panic!("expected Submitted"),
}
let imgs = composer.take_recent_submission_images();
@@ -3354,7 +3457,7 @@ mod tests {
false,
);
let path = PathBuf::from("/tmp/image3.png");
composer.attach_image(path.clone());
composer.attach_image(path.clone(), 20, 10, "PNG");
let placeholder = composer.attached_images[0].placeholder.clone();
// Case 1: backspace at end
@@ -3365,7 +3468,7 @@ mod tests {
// Re-add and test backspace in middle: should break the placeholder string
// and drop the image mapping (same as text placeholder behavior).
composer.attach_image(path);
composer.attach_image(path, 20, 10, "PNG");
let placeholder2 = composer.attached_images[0].placeholder.clone();
// Move cursor to roughly middle of placeholder
if let Some(start_pos) = composer.textarea.text().find(&placeholder2) {
@@ -3397,7 +3500,7 @@ mod tests {
// Insert an image placeholder at the start
let path = PathBuf::from("/tmp/image_multibyte.png");
composer.attach_image(path);
composer.attach_image(path, 10, 5, "PNG");
// Add multibyte text after the placeholder
composer.textarea.insert_str("日本語");
@@ -3406,7 +3509,12 @@ mod tests {
composer.handle_key_event(KeyEvent::new(KeyCode::Backspace, KeyModifiers::NONE));
assert_eq!(composer.attached_images.len(), 1);
assert!(composer.textarea.text().starts_with("[Image #1]"));
assert!(
composer
.textarea
.text()
.starts_with("[image_multibyte.png 10x5]")
);
}
#[test]
@@ -3424,10 +3532,10 @@ mod tests {
let path1 = PathBuf::from("/tmp/image_dup1.png");
let path2 = PathBuf::from("/tmp/image_dup2.png");
composer.attach_image(path1);
composer.attach_image(path1, 10, 5, "PNG");
// separate placeholders with a space for clarity
composer.handle_paste(" ".into());
composer.attach_image(path2.clone());
composer.attach_image(path2.clone(), 10, 5, "PNG");
let placeholder1 = composer.attached_images[0].placeholder.clone();
let placeholder2 = composer.attached_images[1].placeholder.clone();
@@ -3442,60 +3550,24 @@ mod tests {
let new_text = composer.textarea.text().to_string();
assert_eq!(
0,
new_text.matches(&placeholder2).count(),
"second placeholder was relabeled"
new_text.matches(&placeholder1).count(),
"first placeholder removed"
);
assert_eq!(
1,
new_text.matches("[Image #1]").count(),
"remaining placeholder relabeled to #1"
new_text.matches(&placeholder2).count(),
"second placeholder remains"
);
assert_eq!(
vec![AttachedImage {
path: path2,
placeholder: "[Image #1]".to_string()
placeholder: "[image_dup2.png 10x5]".to_string()
}],
composer.attached_images,
"one image mapping remains"
);
}
#[test]
fn deleting_first_text_element_renumbers_following_text_element() {
use crossterm::event::KeyCode;
use crossterm::event::KeyEvent;
use crossterm::event::KeyModifiers;
let (tx, _rx) = unbounded_channel::<AppEvent>();
let sender = AppEventSender::new(tx);
let mut composer = ChatComposer::new(
true,
sender,
false,
"Ask Codex to do anything".to_string(),
false,
);
let path1 = PathBuf::from("/tmp/image_first.png");
let path2 = PathBuf::from("/tmp/image_second.png");
// Insert two adjacent atomic elements.
composer.attach_image(path1);
composer.attach_image(path2.clone());
assert_eq!(composer.textarea.text(), "[Image #1][Image #2]");
assert_eq!(composer.attached_images.len(), 2);
// Delete the first element using normal textarea editing (Delete at cursor start).
composer.textarea.set_cursor(0);
composer.handle_key_event(KeyEvent::new(KeyCode::Delete, KeyModifiers::NONE));
// Remaining image should be renumbered and the textarea element updated.
assert_eq!(composer.attached_images.len(), 1);
assert_eq!(composer.attached_images[0].path, path2);
assert_eq!(composer.attached_images[0].placeholder, "[Image #1]");
assert_eq!(composer.textarea.text(), "[Image #1]");
}
#[test]
fn pasting_filepath_attaches_image() {
let tmp = tempdir().expect("create TempDir");
@@ -3516,7 +3588,12 @@ mod tests {
let needs_redraw = composer.handle_paste(tmp_path.to_string_lossy().to_string());
assert!(needs_redraw);
assert!(composer.textarea.text().starts_with("[Image #1] "));
assert!(
composer
.textarea
.text()
.starts_with("[codex_tui_test_paste_image.png 3x2] ")
);
let imgs = composer.take_recent_submission_images();
assert_eq!(imgs, vec![tmp_path]);

View File

@@ -529,9 +529,16 @@ impl BottomPane {
self.request_redraw();
}
pub(crate) fn attach_image(&mut self, path: PathBuf) {
pub(crate) fn attach_image(
&mut self,
path: PathBuf,
width: u32,
height: u32,
format_label: &str,
) {
if self.view_stack.is_empty() {
self.composer.attach_image(path);
self.composer
.attach_image(path, width, height, format_label);
self.request_redraw();
}
}

View File

@@ -715,88 +715,6 @@ impl TextArea {
// ===== Text elements support =====
pub fn element_payloads(&self) -> Vec<String> {
self.elements
.iter()
.filter_map(|e| self.text.get(e.range.clone()).map(str::to_string))
.collect()
}
pub fn element_payload_starting_at(&self, pos: usize) -> Option<String> {
let pos = pos.min(self.text.len());
let elem = self.elements.iter().find(|e| e.range.start == pos)?;
self.text.get(elem.range.clone()).map(str::to_string)
}
/// Renames a single text element in-place, keeping it atomic.
///
/// This is intended for cases where the element payload is an identifier (e.g. a placeholder)
/// that must be updated without converting the element back into normal text.
pub fn replace_element_payload(&mut self, old: &str, new: &str) -> bool {
let Some(idx) = self
.elements
.iter()
.position(|e| self.text.get(e.range.clone()) == Some(old))
else {
return false;
};
let range = self.elements[idx].range.clone();
let start = range.start;
let end = range.end;
if start > end || end > self.text.len() {
return false;
}
let removed_len = end - start;
let inserted_len = new.len();
let diff = inserted_len as isize - removed_len as isize;
self.text.replace_range(range, new);
self.wrap_cache.replace(None);
self.preferred_col = None;
// Update the modified element's range.
self.elements[idx].range = start..(start + inserted_len);
// Shift element ranges that occur after the replaced element.
if diff != 0 {
for (j, e) in self.elements.iter_mut().enumerate() {
if j == idx {
continue;
}
if e.range.end <= start {
continue;
}
if e.range.start >= end {
e.range.start = ((e.range.start as isize) + diff) as usize;
e.range.end = ((e.range.end as isize) + diff) as usize;
continue;
}
// Elements should not partially overlap each other; degrade gracefully by
// snapping anything intersecting the replaced range to the new bounds.
e.range.start = start.min(e.range.start);
e.range.end = (start + inserted_len).max(e.range.end.saturating_add_signed(diff));
}
}
// Update the cursor position to account for the edit.
self.cursor_pos = if self.cursor_pos < start {
self.cursor_pos
} else if self.cursor_pos <= end {
start + inserted_len
} else {
((self.cursor_pos as isize) + diff) as usize
};
self.cursor_pos = self.clamp_pos_to_nearest_boundary(self.cursor_pos);
// Keep element ordering deterministic.
self.elements.sort_by_key(|e| e.range.start);
true
}
pub fn insert_element(&mut self, text: &str) {
let start = self.clamp_pos_for_insertion(self.cursor_pos);
self.insert_str_at(start, text);

View File

@@ -1452,13 +1452,12 @@ impl ChatWidget {
{
match paste_image_to_temp_png() {
Ok((path, info)) => {
tracing::debug!(
"pasted image size={}x{} format={}",
self.attach_image(
path,
info.width,
info.height,
info.encoded_format.label()
info.encoded_format.label(),
);
self.attach_image(path);
}
Err(err) => {
tracing::warn!("failed to paste image: {err}");
@@ -1511,9 +1510,18 @@ impl ChatWidget {
}
}
pub(crate) fn attach_image(&mut self, path: PathBuf) {
tracing::info!("attach_image path={path:?}");
self.bottom_pane.attach_image(path);
pub(crate) fn attach_image(
&mut self,
path: PathBuf,
width: u32,
height: u32,
format_label: &str,
) {
tracing::info!(
"attach_image path={path:?} width={width} height={height} format={format_label}",
);
self.bottom_pane
.attach_image(path, width, height, format_label);
self.request_redraw();
}
@@ -1810,14 +1818,14 @@ impl ChatWidget {
return;
}
for path in image_paths {
items.push(UserInput::LocalImage { path });
}
if !text.is_empty() {
items.push(UserInput::Text { text: text.clone() });
}
for path in image_paths {
items.push(UserInput::LocalImage { path });
}
if let Some(skills) = self.bottom_pane.skills() {
let skill_mentions = find_skill_mentions(&text, skills);
for skill in skill_mentions {
@@ -1988,9 +1996,6 @@ impl ChatWidget {
}
EventMsg::ExitedReviewMode(review) => self.on_exited_review_mode(review),
EventMsg::ContextCompacted(_) => self.on_agent_message("Context compacted".to_owned()),
EventMsg::CollabInteraction(_) => {
// TODO(jif) handle collab tools.
}
EventMsg::RawResponseItem(_)
| EventMsg::ThreadRolledBack(_)
| EventMsg::ItemStarted(_)

View File

@@ -1065,8 +1065,8 @@ async fn ctrl_c_cleared_prompt_is_recoverable_via_history() {
chat.bottom_pane.insert_str("draft message ");
chat.bottom_pane
.attach_image(PathBuf::from("/tmp/preview.png"));
let placeholder = "[Image #1]";
.attach_image(PathBuf::from("/tmp/preview.png"), 24, 42, "png");
let placeholder = "[preview.png 24x42]";
assert!(
chat.bottom_pane.composer_text().ends_with(placeholder),
"expected placeholder {placeholder:?} in composer text"

42
rbe.bzl
View File

@@ -1,42 +0,0 @@
def _rbe_platform_repo_impl(rctx):
arch = rctx.os.arch
if arch in ["x86_64", "amd64"]:
cpu = "x86_64"
exec_arch = "amd64"
image_sha = "8c9ff94187ea7c08a31e9a81f5fe8046ea3972a6768983c955c4079fa30567fb"
elif arch in ["aarch64", "arm64"]:
cpu = "aarch64"
exec_arch = "arm64"
image_sha = "ad9506086215fccfc66ed8d2be87847324be56790ae6a1964c241c28b77ef141"
else:
fail("Unsupported host arch for rbe platform: {}".format(arch))
rctx.file("BUILD.bazel", """\
platform(
name = "rbe_platform",
constraint_values = [
"@platforms//cpu:{cpu}",
"@platforms//os:linux",
"@bazel_tools//tools/cpp:clang",
"@toolchains_llvm_bootstrapped//constraints/libc:gnu.2.28",
],
exec_properties = {{
# Ubuntu-based image that includes git, python3, dotslash, and other
# tools that various integration tests need.
# Verify at https://hub.docker.com/layers/mbolin491/codex-bazel/latest/images/sha256:{image_sha}
"container-image": "docker://docker.io/mbolin491/codex-bazel@sha256:{image_sha}",
"Arch": "{arch}",
"OSFamily": "Linux",
}},
visibility = ["//visibility:public"],
)
""".format(
cpu = cpu,
arch = exec_arch,
image_sha = image_sha
))
rbe_platform_repository = repository_rule(
implementation = _rbe_platform_repo_impl,
doc = "Sets up a platform for remote builds with an Arch exec_property matching the host.",
)