Standardize shell execution on unified exec (#39772)

## What changed

- Use `exec_command` and `write_stdin` as the shell tool surface.
- Treat legacy `default`, `local`, and `shell_command` model metadata as `unified_exec`.
- Remove obsolete shell-selection configuration and runtime paths while preserving the feature and policy gates for zsh fork execution.

GitOrigin-RevId: d743cbe598630d73052f1fecad680c4cde17977d
This commit is contained in:
jif
2026-08-20 18:29:35 +00:00
committed by copyberry
parent 5bcd7b0fbc
commit bce5f2fcfc
40 changed files with 189 additions and 992 deletions
+6 -183
View File
@@ -112,7 +112,6 @@ impl ToolCallRuntime {
let router = &self.step_context.tool_router;
let supports_parallel = router.tool_supports_parallel(&call);
let tool_runtime = router.tool_runtime(&call);
let wait_for_runtime_cancellation = router.tool_waits_for_runtime_cancellation(&call);
let router = Arc::clone(router);
let session = Arc::clone(&self.session);
let step_context = Arc::clone(&self.step_context);
@@ -185,24 +184,11 @@ impl ToolCallRuntime {
} else {
let secs = started.elapsed().as_secs_f32().max(0.1);
abort_dispatch_span.record("aborted", true);
if wait_for_runtime_cancellation {
if terminal_outcome_reached.swap(true, Ordering::AcqRel) {
return dispatch_handle.await.map_err(Self::tool_task_join_error)?;
}
// The abort owns the terminal outcome; await only so
// the runtime can finish process teardown.
match dispatch_handle.await {
Ok(_) => {}
Err(err) if err.is_cancelled() => {}
Err(err) => return Err(Self::tool_task_join_error(err)),
}
} else {
dispatch_handle.abort();
match dispatch_handle.await {
Ok(result) => return result,
Err(err) if err.is_cancelled() => {}
Err(err) => return Err(Self::tool_task_join_error(err)),
}
dispatch_handle.abort();
match dispatch_handle.await {
Ok(result) => return result,
Err(err) if err.is_cancelled() => {}
Err(err) => return Err(Self::tool_task_join_error(err)),
}
let response = Self::aborted_response(&call, secs);
notify_tool_aborted(
@@ -266,12 +252,7 @@ impl ToolCallRuntime {
}
fn abort_message(call: &ToolCall, secs: f32) -> String {
if call.tool_name.is_default_namespace()
&& matches!(
call.tool_name.name.as_str(),
"exec_command" | "unified_exec"
)
{
if call.tool_name.is_default_namespace() && call.tool_name.name == "exec_command" {
format!("Wall time: {secs:.1} seconds\naborted by user")
} else {
format!("aborted by user after {secs:.1}s")
@@ -568,90 +549,6 @@ mod tests {
impl CoreToolRuntime for ImmediateHandler {}
struct CancellationCleanupHandler {
tool_name: codex_tools::ToolName,
started: std::sync::Mutex<Option<oneshot::Sender<()>>>,
cleanup_started: std::sync::Mutex<Option<oneshot::Sender<()>>>,
allow_cleanup: Arc<Notify>,
}
impl ToolExecutor<ToolInvocation> for CancellationCleanupHandler {
fn tool_name(&self) -> codex_tools::ToolName {
self.tool_name.clone()
}
fn spec(&self) -> codex_tools::ToolSpec {
codex_tools::ToolSpec::Function(codex_tools::ResponsesApiTool {
name: self.tool_name.name.clone(),
description: "Cancellation cleanup test tool.".to_string(),
strict: false,
defer_loading: None,
parameters: codex_tools::JsonSchema::default(),
output_schema: None,
})
}
fn handle(&self, invocation: ToolInvocation) -> codex_tools::ToolExecutorFuture<'_> {
Box::pin(self.handle_call(invocation))
}
}
impl CancellationCleanupHandler {
async fn handle_call(
&self,
invocation: ToolInvocation,
) -> Result<Box<dyn crate::tools::context::ToolOutput>, FunctionCallError> {
let started = self
.started
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.take();
if let Some(started) = started {
let _ = started.send(());
}
invocation.cancellation_token.cancelled().await;
let cleanup_started = self
.cleanup_started
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.take();
if let Some(cleanup_started) = cleanup_started {
let _ = cleanup_started.send(());
}
self.allow_cleanup.notified().await;
Ok(Box::new(FunctionToolOutput::from_text(
"cleanup complete".to_string(),
Some(false),
)) as Box<dyn crate::tools::context::ToolOutput>)
}
}
impl CoreToolRuntime for CancellationCleanupHandler {
fn waits_for_runtime_cancellation(&self) -> bool {
true
}
}
struct FinishRecorder {
records: Arc<std::sync::Mutex<Vec<ToolCallOutcome>>>,
}
impl codex_extension_api::ToolLifecycleContributor for FinishRecorder {
fn on_tool_finish<'a>(
&'a self,
input: codex_extension_api::ToolFinishInput<'a>,
) -> codex_extension_api::ToolLifecycleFuture<'a> {
let records = Arc::clone(&self.records);
let outcome = input.outcome;
Box::pin(async move {
records
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.push(outcome);
})
}
}
struct BlockingFinishContributor {
records: Arc<std::sync::Mutex<Vec<ToolCallOutcome>>>,
finish_started: std::sync::Mutex<Option<oneshot::Sender<()>>>,
@@ -756,78 +653,4 @@ mod tests {
Ok(())
}
#[tokio::test]
async fn cancellation_waiting_for_runtime_cleanup_emits_only_aborted_lifecycle()
-> anyhow::Result<()> {
let (mut session, turn_context) = crate::session::tests::make_session_and_context().await;
let records = Arc::new(std::sync::Mutex::new(Vec::new()));
let mut builder =
codex_extension_api::ExtensionRegistryBuilder::<crate::config::Config>::new();
builder.tool_lifecycle_contributor(Arc::new(FinishRecorder {
records: Arc::clone(&records),
}));
session.services.extensions = Arc::new(builder.build());
let session = Arc::new(session);
let turn_context = Arc::new(turn_context);
let tool_name = codex_tools::ToolName::plain("cleanup_tool");
let (started_tx, started_rx) = oneshot::channel();
let (cleanup_started_tx, cleanup_started_rx) = oneshot::channel();
let allow_cleanup = Arc::new(Notify::new());
let handler = Arc::new(CancellationCleanupHandler {
tool_name: tool_name.clone(),
started: std::sync::Mutex::new(Some(started_tx)),
cleanup_started: std::sync::Mutex::new(Some(cleanup_started_tx)),
allow_cleanup: Arc::clone(&allow_cleanup),
}) as Arc<dyn CoreToolRuntime>;
let step_context = StepContext::for_test(Arc::clone(&turn_context));
let router = Arc::new(ToolRouter::from_parts(
ToolRegistry::from_tools([handler]),
Vec::new(),
));
let step_context = step_context.with_tool_router_for_test(router);
let tracker = Arc::new(tokio::sync::Mutex::new(TurnDiffTracker::new()));
let runtime = ToolCallRuntime::new(session, step_context, tracker);
let cancellation_token = CancellationToken::new();
let call = ToolCall {
tool_name,
call_id: "call-1".to_string(),
payload: ToolPayload::Function {
arguments: "{}".to_string(),
},
encrypted_function_args: None,
};
let response_task =
tokio::spawn(runtime.handle_tool_call(call, cancellation_token.clone()));
started_rx.await.expect("handler should start");
cancellation_token.cancel();
cleanup_started_rx
.await
.expect("handler should start cleanup");
tokio::time::sleep(Duration::from_millis(10)).await;
allow_cleanup.notify_one();
let response = tokio::time::timeout(Duration::from_secs(1), response_task)
.await
.expect("timed out waiting for tool response")
.expect("tool response task should join")?;
let ResponseInputItem::FunctionCallOutput { output, .. } = response else {
anyhow::bail!("cancelled tool should return function output");
};
let FunctionCallOutputBody::Text(text) = output.body else {
anyhow::bail!("cancelled tool output should be text");
};
assert!(text.contains("aborted by user"));
let actual = records
.lock()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.drain(..)
.collect::<Vec<_>>();
assert_eq!(vec![ToolCallOutcome::Aborted], actual);
Ok(())
}
}