Stop auto-compact loop on compact failure

codex-rs/core/src/codex.rs

@@ -3654,8 +3654,8 @@ pub(crate) async fn run_turn(
         collaboration_mode_kind: turn_context.collaboration_mode.mode,
     });
     sess.send_event(&turn_context, event).await;
-    if total_usage_tokens >= auto_compact_limit {
-        run_auto_compact(&sess, &turn_context).await;
+    if total_usage_tokens >= auto_compact_limit && !run_auto_compact(&sess, &turn_context).await {
+        return None;
     }
 
     let skills_outcome = Some(
@@ -3837,7 +3837,9 @@ pub(crate) async fn run_turn(
 
                 // as long as compaction works well in getting us way below the token limit, we shouldn't worry about being in an infinite loop.
                 if token_limit_reached && needs_follow_up {
-                    run_auto_compact(&sess, &turn_context).await;
+                    if !run_auto_compact(&sess, &turn_context).await {
+                        break;
+                    }
                     continue;
                 }
 
@@ -3895,11 +3897,11 @@ pub(crate) async fn run_turn(
     last_agent_message
 }
 
-async fn run_auto_compact(sess: &Arc<Session>, turn_context: &Arc<TurnContext>) {
+async fn run_auto_compact(sess: &Arc<Session>, turn_context: &Arc<TurnContext>) -> bool {
     if should_use_remote_compact_task(&turn_context.provider) {
-        run_inline_remote_auto_compact_task(Arc::clone(sess), Arc::clone(turn_context)).await;
+        run_inline_remote_auto_compact_task(Arc::clone(sess), Arc::clone(turn_context)).await
     } else {
-        run_inline_auto_compact_task(Arc::clone(sess), Arc::clone(turn_context)).await;
+        run_inline_auto_compact_task(Arc::clone(sess), Arc::clone(turn_context)).await
     }
 }
 

codex-rs/core/src/compact.rs

@@ -40,7 +40,7 @@ pub(crate) fn should_use_remote_compact_task(provider: &ModelProviderInfo) -> bo
 pub(crate) async fn run_inline_auto_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
-) {
+) -> bool {
     let prompt = turn_context.compact_prompt().to_string();
     let input = vec![UserInput::Text {
         text: prompt,
@@ -48,7 +48,7 @@ pub(crate) async fn run_inline_auto_compact_task(
         text_elements: Vec::new(),
     }];
 
-    run_compact_task_inner(sess, turn_context, input).await;
+    run_compact_task_inner(sess, turn_context, input).await
 }
 
 pub(crate) async fn run_compact_task(
@@ -61,14 +61,14 @@ pub(crate) async fn run_compact_task(
         collaboration_mode_kind: turn_context.collaboration_mode.mode,
     });
     sess.send_event(&turn_context, start_event).await;
-    run_compact_task_inner(sess.clone(), turn_context, input).await;
+    let _ = run_compact_task_inner(sess.clone(), turn_context, input).await;
 }
 
 async fn run_compact_task_inner(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
     input: Vec<UserInput>,
-) {
+) -> bool {
     let compaction_item = TurnItem::ContextCompaction(ContextCompactionItem::new());
     sess.emit_turn_item_started(&turn_context, &compaction_item)
         .await;
@@ -143,7 +143,7 @@ async fn run_compact_task_inner(
                 break;
             }
             Err(CodexErr::Interrupted) => {
-                return;
+                return false;
             }
             Err(e @ CodexErr::ContextWindowExceeded) => {
                 if turn_input_len > 1 {
@@ -159,7 +159,7 @@ async fn run_compact_task_inner(
                 sess.set_total_tokens_full(turn_context.as_ref()).await;
                 let event = EventMsg::Error(e.to_error_event(None));
                 sess.send_event(&turn_context, event).await;
-                return;
+                return false;
             }
             Err(e) => {
                 if retries < max_retries {
@@ -176,7 +176,7 @@ async fn run_compact_task_inner(
                 } else {
                     let event = EventMsg::Error(e.to_error_event(None));
                     sess.send_event(&turn_context, event).await;
-                    return;
+                    return false;
                 }
             }
         }
@@ -211,6 +211,7 @@ async fn run_compact_task_inner(
         message: "Heads up: Long threads and multiple compactions can cause the model to be less accurate. Start a new thread when possible to keep threads small and targeted.".to_string(),
     });
     sess.send_event(&turn_context, warning).await;
+    true
 }
 
 pub fn content_items_to_text(content: &[ContentItem]) -> Option<String> {

codex-rs/core/src/compact_remote.rs

@@ -19,8 +19,8 @@ use tracing::info;
 pub(crate) async fn run_inline_remote_auto_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
-) {
-    run_remote_compact_task_inner(&sess, &turn_context).await;
+) -> bool {
+    run_remote_compact_task_inner(&sess, &turn_context).await
 }
 
 pub(crate) async fn run_remote_compact_task(sess: Arc<Session>, turn_context: Arc<TurnContext>) {
@@ -30,16 +30,21 @@ pub(crate) async fn run_remote_compact_task(sess: Arc<Session>, turn_context: Ar
     });
     sess.send_event(&turn_context, start_event).await;
 
-    run_remote_compact_task_inner(&sess, &turn_context).await;
+    let _ = run_remote_compact_task_inner(&sess, &turn_context).await;
 }
 
-async fn run_remote_compact_task_inner(sess: &Arc<Session>, turn_context: &Arc<TurnContext>) {
+async fn run_remote_compact_task_inner(
+    sess: &Arc<Session>,
+    turn_context: &Arc<TurnContext>,
+) -> bool {
     if let Err(err) = run_remote_compact_task_inner_impl(sess, turn_context).await {
         let event = EventMsg::Error(
             err.to_error_event(Some("Error running remote compact task".to_string())),
         );
         sess.send_event(turn_context, event).await;
+        return false;
     }
+    true
 }
 
 async fn run_remote_compact_task_inner_impl(

codex-rs/core/tests/suite/compact.rs

@@ -2226,6 +2226,75 @@ async fn auto_compact_triggers_after_function_call_over_95_percent_usage() {
     );
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_failure_stops_follow_up_sampling_loop() {
+    skip_if_no_network!();
+
+    let server = start_mock_server().await;
+
+    let context_window = 100;
+    let limit = context_window * 90 / 100;
+    let over_limit_tokens = context_window * 95 / 100 + 1;
+
+    let first_turn = sse(vec![
+        ev_function_call(DUMMY_CALL_ID, DUMMY_FUNCTION_NAME, "{}"),
+        ev_completed_with_tokens("r1", over_limit_tokens),
+    ]);
+    let follow_up_turn = sse(vec![
+        ev_assistant_message("m2", "SHOULD_NOT_BE_REQUESTED"),
+        ev_completed_with_tokens("r2", 1),
+    ]);
+
+    mount_sse_once(&server, first_turn).await;
+    let follow_up_mock = mount_sse_once(&server, follow_up_turn).await;
+    let compact_mock =
+        mount_compact_json_once(&server, serde_json::json!({ "bad": "shape" })).await;
+
+    let codex = test_codex()
+        .with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
+        .with_config(move |config| {
+            set_test_compact_prompt(config);
+            config.model_context_window = Some(context_window);
+            config.model_auto_compact_token_limit = Some(limit);
+        })
+        .build(&server)
+        .await
+        .expect("build codex")
+        .codex;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![UserInput::Text {
+                text: FUNCTION_CALL_LIMIT_MSG.into(),
+                text_elements: Vec::new(),
+            }],
+            final_output_json_schema: None,
+        })
+        .await
+        .unwrap();
+
+    let error_message = wait_for_event_match(&codex, |msg| match msg {
+        EventMsg::Error(err) => Some(err.message.clone()),
+        _ => None,
+    })
+    .await;
+    assert!(
+        error_message.contains("Error running remote compact task"),
+        "expected remote auto compact failure message, got: {error_message}"
+    );
+    wait_for_event(&codex, |msg| matches!(msg, EventMsg::TurnComplete(_))).await;
+
+    assert_eq!(
+        compact_mock.requests().len(),
+        1,
+        "auto compact should run once before failing"
+    );
+    assert!(
+        follow_up_mock.requests().is_empty(),
+        "turn should stop after auto compact failure and avoid follow-up sampling"
+    );
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn auto_compact_counts_encrypted_reasoning_before_last_user() {
     skip_if_no_network!();

codex-rs/core/tests/suite/compact_remote.rs

@@ -245,6 +245,7 @@ async fn remote_compact_trims_function_call_history_to_fit_context_window() -> R
             .with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
             .with_config(|config| {
                 config.model_context_window = Some(2_000);
+                config.model_auto_compact_token_limit = Some(i64::MAX);
             }),
     )
     .await?;