diff --git a/codex-rs/core/tests/suite/compact.rs b/codex-rs/core/tests/suite/compact.rs index 22a6c72076..08850ec38f 100644 --- a/codex-rs/core/tests/suite/compact.rs +++ b/codex-rs/core/tests/suite/compact.rs @@ -3038,6 +3038,131 @@ async fn snapshot_request_shape_pre_turn_compaction_including_incoming_user_mess #[tokio::test(flavor = "multi_thread", worker_threads = 2)] // TODO(ccunningham): Update once pre-turn compaction context-overflow handling includes incoming // user input and emits richer oversized-input messaging. +async fn snapshot_request_shape_pre_turn_compaction_strips_incoming_model_switch() { + skip_if_no_network!(); + + let server = start_mock_server().await; + let previous_model = "gpt-5.1-codex-max"; + let next_model = "gpt-5.2-codex"; + + let request_log = mount_sse_sequence( + &server, + vec![ + sse(vec![ + ev_assistant_message("m1", "BEFORE_SWITCH_REPLY"), + ev_completed_with_tokens("r1", 500), + ]), + sse(vec![ + ev_assistant_message("m2", "PRETURN_SWITCH_SUMMARY"), + ev_completed_with_tokens("r2", 100), + ]), + sse(vec![ + ev_assistant_message("m3", "AFTER_SWITCH_REPLY"), + ev_completed_with_tokens("r3", 100), + ]), + ], + ) + .await; + + let model_provider = non_openai_model_provider(&server); + let test = test_codex() + .with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing()) + .with_model(previous_model) + .with_config(move |config| { + config.model_provider = model_provider; + set_test_compact_prompt(config); + config + .features + .enable(codex_core::features::Feature::RemoteModels); + config.model_auto_compact_token_limit = Some(200); + }) + .build(&server) + .await + .expect("build codex"); + + test.codex + .submit(Op::UserTurn { + items: vec![UserInput::Text { + text: "BEFORE_SWITCH_USER".into(), + text_elements: Vec::new(), + }], + final_output_json_schema: None, + cwd: test.cwd.path().to_path_buf(), + approval_policy: AskForApproval::Never, + sandbox_policy: SandboxPolicy::DangerFullAccess, + model: previous_model.to_string(), + effort: None, + summary: ReasoningSummary::Auto, + collaboration_mode: None, + personality: None, + }) + .await + .expect("submit first user turn"); + wait_for_event(&test.codex, |event| { + matches!(event, EventMsg::TurnComplete(_)) + }) + .await; + + test.codex + .submit(Op::UserTurn { + items: vec![UserInput::Text { + text: "AFTER_SWITCH_USER".into(), + text_elements: Vec::new(), + }], + final_output_json_schema: None, + cwd: test.cwd.path().to_path_buf(), + approval_policy: AskForApproval::Never, + sandbox_policy: SandboxPolicy::DangerFullAccess, + model: next_model.to_string(), + effort: None, + summary: ReasoningSummary::Auto, + collaboration_mode: None, + personality: None, + }) + .await + .expect("submit second user turn"); + wait_for_event(&test.codex, |event| { + matches!(event, EventMsg::TurnComplete(_)) + }) + .await; + + let requests = request_log.requests(); + assert_eq!( + requests.len(), + 3, + "expected first turn, pre-turn compact, and post-compact follow-up requests" + ); + + let compact_body = requests[1].body_json().to_string(); + assert!( + body_contains_text(&compact_body, SUMMARIZATION_PROMPT), + "pre-turn compaction request should include summarization prompt" + ); + assert!( + !compact_body.contains(""), + "pre-turn compaction request should strip incoming model-switch update item" + ); + + let follow_up_body = requests[2].body_json().to_string(); + assert!( + follow_up_body.contains(""), + "post-compaction follow-up should include model-switch update item" + ); + + insta::assert_snapshot!( + "pre_turn_compaction_strips_incoming_model_switch_shapes", + format_labeled_requests_snapshot( + "Pre-turn compaction during model switch (without pre-sampling model-switch compaction): current behavior strips incoming from the compact request and restores it in the post-compaction follow-up request.", + &[ + ("Initial Request (Previous Model)", &requests[0]), + ("Local Compaction Request", &requests[1]), + ("Local Post-Compaction History Layout", &requests[2]), + ] + ) + ); +} + +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] async fn snapshot_request_shape_pre_turn_compaction_context_window_exceeded() { skip_if_no_network!(); diff --git a/codex-rs/core/tests/suite/compact_remote.rs b/codex-rs/core/tests/suite/compact_remote.rs index 6c31b8dbeb..6677bc227c 100644 --- a/codex-rs/core/tests/suite/compact_remote.rs +++ b/codex-rs/core/tests/suite/compact_remote.rs @@ -1424,6 +1424,137 @@ async fn snapshot_request_shape_remote_pre_turn_compaction_including_incoming_us Ok(()) } +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] +async fn snapshot_request_shape_remote_pre_turn_compaction_strips_incoming_model_switch() +-> Result<()> { + skip_if_no_network!(Ok(())); + + let previous_model = "gpt-5.1-codex-max"; + let next_model = "gpt-5.2-codex"; + let harness = TestCodexHarness::with_builder( + test_codex() + .with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing()) + .with_model(previous_model) + .with_config(|config| { + config.model_auto_compact_token_limit = Some(200); + }), + ) + .await?; + let codex = harness.test().codex.clone(); + + let responses_mock = responses::mount_sse_sequence( + harness.server(), + vec![ + responses::sse(vec![ + responses::ev_assistant_message("m1", "BEFORE_SWITCH_REPLY"), + responses::ev_completed_with_tokens("r1", 500), + ]), + responses::sse(vec![ + responses::ev_assistant_message("m2", "AFTER_SWITCH_REPLY"), + responses::ev_completed_with_tokens("r2", 80), + ]), + ], + ) + .await; + let compact_mock = responses::mount_compact_json_once( + harness.server(), + serde_json::json!({ + "output": [ + responses::user_message_item("BEFORE_SWITCH_USER"), + responses::user_message_item("AFTER_SWITCH_USER"), + responses::user_message_item(&summary_with_prefix("REMOTE_SWITCH_SUMMARY")) + ] + }), + ) + .await; + + codex + .submit(Op::UserInput { + items: vec![UserInput::Text { + text: "BEFORE_SWITCH_USER".to_string(), + text_elements: Vec::new(), + }], + final_output_json_schema: None, + }) + .await?; + wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await; + + codex + .submit(Op::OverrideTurnContext { + cwd: None, + approval_policy: None, + sandbox_policy: None, + windows_sandbox_level: None, + model: Some(next_model.to_string()), + effort: None, + summary: None, + collaboration_mode: None, + personality: None, + }) + .await?; + codex + .submit(Op::UserInput { + items: vec![UserInput::Text { + text: "AFTER_SWITCH_USER".to_string(), + text_elements: Vec::new(), + }], + final_output_json_schema: None, + }) + .await?; + wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await; + + assert_eq!( + compact_mock.requests().len(), + 1, + "expected a single remote pre-turn compaction request" + ); + let requests = responses_mock.requests(); + assert_eq!( + requests.len(), + 2, + "expected initial turn request and post-compaction follow-up request" + ); + + let compact_request = compact_mock.single_request(); + let compact_body = compact_request.body_json().to_string(); + assert!( + compact_body.contains("AFTER_SWITCH_USER"), + "pre-turn remote compaction request should include incoming user message" + ); + assert!( + !compact_body.contains(""), + "pre-turn remote compaction request should strip incoming model-switch update item" + ); + + let follow_up_body = requests[1].body_json().to_string(); + assert!( + follow_up_body.contains("BEFORE_SWITCH_USER"), + "post-compaction follow-up should preserve older user messages when they fit" + ); + assert!( + follow_up_body.contains("AFTER_SWITCH_USER"), + "post-compaction follow-up should preserve incoming user message from compact output" + ); + assert!( + follow_up_body.contains(""), + "post-compaction follow-up should include the model-switch update item" + ); + + insta::assert_snapshot!( + "remote_pre_turn_compaction_strips_incoming_model_switch_shapes", + format_labeled_requests_snapshot( + "Remote pre-turn compaction during model switch strips incoming from the compact request and restores it in the post-compaction follow-up request.", + &[ + ("Initial Request (Previous Model)", &requests[0]), + ("Remote Compaction Request", &compact_request), + ("Remote Post-Compaction History Layout", &requests[1]), + ] + ) + ); + + Ok(()) +} + #[tokio::test(flavor = "multi_thread", worker_threads = 2)] // TODO(ccunningham): Update once remote pre-turn compaction context-overflow handling includes // incoming user input and emits richer oversized-input messaging. @@ -1599,6 +1730,153 @@ async fn snapshot_request_shape_remote_mid_turn_continuation_compaction() -> Res Ok(()) } +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] +async fn snapshot_request_shape_remote_mid_turn_compaction_summary_only_reinjects_context() +-> Result<()> { + skip_if_no_network!(Ok(())); + + let harness = TestCodexHarness::with_builder( + test_codex() + .with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing()) + .with_config(|config| { + config.model_auto_compact_token_limit = Some(200); + }), + ) + .await?; + let codex = harness.test().codex.clone(); + + let responses_mock = responses::mount_sse_sequence( + harness.server(), + vec![ + responses::sse(vec![ + responses::ev_function_call("call-remote-summary-only", DUMMY_FUNCTION_NAME, "{}"), + responses::ev_completed_with_tokens("r1", 500), + ]), + responses::sse(vec![ + responses::ev_assistant_message("m2", "REMOTE_SUMMARY_ONLY_FINAL_REPLY"), + responses::ev_completed_with_tokens("r2", 80), + ]), + ], + ) + .await; + + let compacted_history = vec![responses::user_message_item(&summary_with_prefix( + "REMOTE_SUMMARY_ONLY", + ))]; + let compact_mock = responses::mount_compact_json_once( + harness.server(), + serde_json::json!({ "output": compacted_history }), + ) + .await; + + codex + .submit(Op::UserInput { + items: vec![UserInput::Text { + text: "USER_ONE".to_string(), + text_elements: Vec::new(), + }], + final_output_json_schema: None, + }) + .await?; + wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await; + + assert_eq!(compact_mock.requests().len(), 1); + let requests = responses_mock.requests(); + assert_eq!( + requests.len(), + 2, + "expected initial and post-compact requests" + ); + + let compact_request = compact_mock.single_request(); + insta::assert_snapshot!( + "remote_mid_turn_compaction_summary_only_reinjects_context_shapes", + format_labeled_requests_snapshot( + "Remote mid-turn compaction where compact output has only summary user content: continuation layout reinjects canonical context before that summary.", + &[ + ("Remote Compaction Request", &compact_request), + ("Remote Post-Compaction History Layout", &requests[1]), + ] + ) + ); + + Ok(()) +} + +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] +async fn snapshot_request_shape_remote_mid_turn_compaction_multi_summary_reinjects_above_last_summary() +-> Result<()> { + skip_if_no_network!(Ok(())); + + let harness = TestCodexHarness::with_builder( + test_codex() + .with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing()) + .with_config(|config| { + config.model_auto_compact_token_limit = Some(200); + }), + ) + .await?; + let codex = harness.test().codex.clone(); + + let responses_mock = responses::mount_sse_sequence( + harness.server(), + vec![ + responses::sse(vec![ + responses::ev_function_call("call-remote-multi-summary", DUMMY_FUNCTION_NAME, "{}"), + responses::ev_completed_with_tokens("r1", 500), + ]), + responses::sse(vec![ + responses::ev_assistant_message("m2", "REMOTE_MULTI_SUMMARY_FINAL_REPLY"), + responses::ev_completed_with_tokens("r2", 80), + ]), + ], + ) + .await; + + let compacted_history = vec![ + responses::user_message_item(&summary_with_prefix("REMOTE_OLDER_SUMMARY")), + responses::user_message_item(&summary_with_prefix("REMOTE_LATEST_SUMMARY")), + ]; + let compact_mock = responses::mount_compact_json_once( + harness.server(), + serde_json::json!({ "output": compacted_history }), + ) + .await; + + codex + .submit(Op::UserInput { + items: vec![UserInput::Text { + text: "USER_ONE".to_string(), + text_elements: Vec::new(), + }], + final_output_json_schema: None, + }) + .await?; + wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await; + + assert_eq!(compact_mock.requests().len(), 1); + let requests = responses_mock.requests(); + assert_eq!( + requests.len(), + 2, + "expected initial and post-compact requests" + ); + + let compact_request = compact_mock.single_request(); + insta::assert_snapshot!( + "remote_mid_turn_compaction_multi_summary_reinjects_above_last_summary_shapes", + format_labeled_requests_snapshot( + "Remote mid-turn compaction where compact output has multiple summary-only user messages: continuation layout reinjects canonical context above the latest summary.", + &[ + ("Remote Compaction Request", &compact_request), + ("Remote Post-Compaction History Layout", &requests[1]), + ] + ) + ); + + Ok(()) +} + #[tokio::test(flavor = "multi_thread", worker_threads = 2)] // TODO(ccunningham): Update once manual remote /compact with no prior user turn becomes a no-op. async fn snapshot_request_shape_remote_manual_compact_without_previous_user_messages() -> Result<()> diff --git a/codex-rs/core/tests/suite/snapshots/all__suite__compact__pre_turn_compaction_strips_incoming_model_switch_shapes.snap b/codex-rs/core/tests/suite/snapshots/all__suite__compact__pre_turn_compaction_strips_incoming_model_switch_shapes.snap new file mode 100644 index 0000000000..48b721f2c2 --- /dev/null +++ b/codex-rs/core/tests/suite/snapshots/all__suite__compact__pre_turn_compaction_strips_incoming_model_switch_shapes.snap @@ -0,0 +1,32 @@ +--- +source: core/tests/suite/compact.rs +expression: "format_labeled_requests_snapshot(\"Pre-turn compaction during model switch (without pre-sampling model-switch compaction): current behavior strips incoming from the compact request and restores it in the post-compaction follow-up request.\",\n&[(\"Initial Request (Previous Model)\", &requests[0]),\n(\"Local Compaction Request\", &requests[1]),\n(\"Local Post-Compaction History Layout\", &requests[2]),])" +--- +Scenario: Pre-turn compaction during model switch (without pre-sampling model-switch compaction): current behavior strips incoming from the compact request and restores it in the post-compaction follow-up request. + +## Initial Request (Previous Model) +00:message/developer: +01:message/user: +02:message/user:> +03:message/developer: +04:message/user:BEFORE_SWITCH_USER + +## Local Compaction Request +00:message/developer: +01:message/user: +02:message/user:> +03:message/developer: +04:message/user:BEFORE_SWITCH_USER +05:message/assistant:BEFORE_SWITCH_REPLY +06:message/user:AFTER_SWITCH_USER +07:message/user: + +## Local Post-Compaction History Layout +00:message/user:BEFORE_SWITCH_USER +01:message/developer: +02:message/developer: The user has requested a new communication st... +03:message/user: +04:message/user:> +05:message/user:AFTER_SWITCH_USER +06:message/user:\nPRETURN_SWITCH_SUMMARY +07:message/developer:\nThe user was previously using a different model.... diff --git a/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_mid_turn_compaction_multi_summary_reinjects_above_last_summary_shapes.snap b/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_mid_turn_compaction_multi_summary_reinjects_above_last_summary_shapes.snap new file mode 100644 index 0000000000..f90337f824 --- /dev/null +++ b/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_mid_turn_compaction_multi_summary_reinjects_above_last_summary_shapes.snap @@ -0,0 +1,20 @@ +--- +source: core/tests/suite/compact_remote.rs +expression: "format_labeled_requests_snapshot(\"Remote mid-turn compaction where compact output has multiple summary-only user messages: continuation layout reinjects canonical context above the latest summary.\",\n&[(\"Remote Compaction Request\", &compact_request),\n(\"Remote Post-Compaction History Layout\", &requests[1]),])" +--- +Scenario: Remote mid-turn compaction where compact output has multiple summary-only user messages: continuation layout reinjects canonical context above the latest summary. + +## Remote Compaction Request +00:message/developer: +01:message/user: +02:message/user:> +03:message/user:USER_ONE +04:function_call/test_tool +05:function_call_output:unsupported call: test_tool + +## Remote Post-Compaction History Layout +00:message/user:\nREMOTE_OLDER_SUMMARY +01:message/developer: +02:message/user: +03:message/user:> +04:message/user:\nREMOTE_LATEST_SUMMARY diff --git a/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_mid_turn_compaction_summary_only_reinjects_context_shapes.snap b/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_mid_turn_compaction_summary_only_reinjects_context_shapes.snap new file mode 100644 index 0000000000..4ede5c3cad --- /dev/null +++ b/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_mid_turn_compaction_summary_only_reinjects_context_shapes.snap @@ -0,0 +1,19 @@ +--- +source: core/tests/suite/compact_remote.rs +expression: "format_labeled_requests_snapshot(\"Remote mid-turn compaction where compact output has only summary user content: continuation layout reinjects canonical context before that summary.\",\n&[(\"Remote Compaction Request\", &compact_request),\n(\"Remote Post-Compaction History Layout\", &requests[1]),])" +--- +Scenario: Remote mid-turn compaction where compact output has only summary user content: continuation layout reinjects canonical context before that summary. + +## Remote Compaction Request +00:message/developer: +01:message/user: +02:message/user:> +03:message/user:USER_ONE +04:function_call/test_tool +05:function_call_output:unsupported call: test_tool + +## Remote Post-Compaction History Layout +00:message/developer: +01:message/user: +02:message/user:> +03:message/user:\nREMOTE_SUMMARY_ONLY diff --git a/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_pre_turn_compaction_strips_incoming_model_switch_shapes.snap b/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_pre_turn_compaction_strips_incoming_model_switch_shapes.snap new file mode 100644 index 0000000000..0a789f5b69 --- /dev/null +++ b/codex-rs/core/tests/suite/snapshots/all__suite__compact_remote__remote_pre_turn_compaction_strips_incoming_model_switch_shapes.snap @@ -0,0 +1,29 @@ +--- +source: core/tests/suite/compact_remote.rs +expression: "format_labeled_requests_snapshot(\"Remote pre-turn compaction during model switch strips incoming from the compact request and restores it in the post-compaction follow-up request.\",\n&[(\"Initial Request (Previous Model)\", &requests[0]),\n(\"Remote Compaction Request\", &compact_request),\n(\"Remote Post-Compaction History Layout\", &requests[1]),])" +--- +Scenario: Remote pre-turn compaction during model switch strips incoming from the compact request and restores it in the post-compaction follow-up request. + +## Initial Request (Previous Model) +00:message/developer: +01:message/user: +02:message/user:> +03:message/user:BEFORE_SWITCH_USER + +## Remote Compaction Request +00:message/developer: +01:message/user: +02:message/user:> +03:message/user:BEFORE_SWITCH_USER +04:message/assistant:BEFORE_SWITCH_REPLY +05:message/user:AFTER_SWITCH_USER + +## Remote Post-Compaction History Layout +00:message/user:BEFORE_SWITCH_USER +01:message/developer: +02:message/developer: The user has requested a new communication st... +03:message/user: +04:message/user:> +05:message/user:AFTER_SWITCH_USER +06:message/user:\nREMOTE_SWITCH_SUMMARY +07:message/developer:\nThe user was previously using a different model....