feat: land unified_exec (#10641)
Land `unified_exec` for all non-windows OS
This commit is contained in:
parent
0efd33f7f4
commit
49dd67a260
5 changed files with 118 additions and 82 deletions
|
|
@ -3862,7 +3862,7 @@ model_verbosity = "high"
|
|||
forced_login_method: None,
|
||||
include_apply_patch_tool: false,
|
||||
web_search_mode: None,
|
||||
use_experimental_unified_exec_tool: false,
|
||||
use_experimental_unified_exec_tool: !cfg!(windows),
|
||||
ghost_snapshot: GhostSnapshotConfig::default(),
|
||||
features: Features::with_defaults(),
|
||||
suppress_unstable_features_warning: false,
|
||||
|
|
@ -3947,7 +3947,7 @@ model_verbosity = "high"
|
|||
forced_login_method: None,
|
||||
include_apply_patch_tool: false,
|
||||
web_search_mode: None,
|
||||
use_experimental_unified_exec_tool: false,
|
||||
use_experimental_unified_exec_tool: !cfg!(windows),
|
||||
ghost_snapshot: GhostSnapshotConfig::default(),
|
||||
features: Features::with_defaults(),
|
||||
suppress_unstable_features_warning: false,
|
||||
|
|
@ -4047,7 +4047,7 @@ model_verbosity = "high"
|
|||
forced_login_method: None,
|
||||
include_apply_patch_tool: false,
|
||||
web_search_mode: None,
|
||||
use_experimental_unified_exec_tool: false,
|
||||
use_experimental_unified_exec_tool: !cfg!(windows),
|
||||
ghost_snapshot: GhostSnapshotConfig::default(),
|
||||
features: Features::with_defaults(),
|
||||
suppress_unstable_features_warning: false,
|
||||
|
|
@ -4133,7 +4133,7 @@ model_verbosity = "high"
|
|||
forced_login_method: None,
|
||||
include_apply_patch_tool: false,
|
||||
web_search_mode: None,
|
||||
use_experimental_unified_exec_tool: false,
|
||||
use_experimental_unified_exec_tool: !cfg!(windows),
|
||||
ghost_snapshot: GhostSnapshotConfig::default(),
|
||||
features: Features::with_defaults(),
|
||||
suppress_unstable_features_warning: false,
|
||||
|
|
|
|||
|
|
@ -406,6 +406,12 @@ pub const FEATURES: &[FeatureSpec] = &[
|
|||
stage: Stage::Stable,
|
||||
default_enabled: true,
|
||||
},
|
||||
FeatureSpec {
|
||||
id: Feature::UnifiedExec,
|
||||
key: "unified_exec",
|
||||
stage: Stage::Stable,
|
||||
default_enabled: !cfg!(windows),
|
||||
},
|
||||
FeatureSpec {
|
||||
id: Feature::WebSearchRequest,
|
||||
key: "web_search_request",
|
||||
|
|
@ -419,16 +425,6 @@ pub const FEATURES: &[FeatureSpec] = &[
|
|||
default_enabled: false,
|
||||
},
|
||||
// Experimental program. Rendered in the `/experimental` menu for users.
|
||||
FeatureSpec {
|
||||
id: Feature::UnifiedExec,
|
||||
key: "unified_exec",
|
||||
stage: Stage::Experimental {
|
||||
name: "Background terminal",
|
||||
menu_description: "Run long-running terminal commands in the background.",
|
||||
announcement: "NEW! Try Background terminals for long-running commands. Enable in /experimental!",
|
||||
},
|
||||
default_enabled: false,
|
||||
},
|
||||
FeatureSpec {
|
||||
id: Feature::ShellSnapshot,
|
||||
key: "shell_snapshot",
|
||||
|
|
|
|||
|
|
@ -1687,6 +1687,22 @@ mod tests {
|
|||
assert_eq!(&tool_names, &expected_tools,);
|
||||
}
|
||||
|
||||
fn assert_default_model_tools(
|
||||
model_slug: &str,
|
||||
features: &Features,
|
||||
web_search_mode: Option<WebSearchMode>,
|
||||
shell_tool: &'static str,
|
||||
expected_tail: &[&str],
|
||||
) {
|
||||
let mut expected = if features.enabled(Feature::UnifiedExec) {
|
||||
vec!["exec_command", "write_stdin"]
|
||||
} else {
|
||||
vec![shell_tool]
|
||||
};
|
||||
expected.extend(expected_tail);
|
||||
assert_model_tools(model_slug, features, web_search_mode, &expected);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn web_search_mode_cached_sets_external_web_access_false() {
|
||||
let config = test_config();
|
||||
|
|
@ -1735,12 +1751,12 @@ mod tests {
|
|||
fn test_build_specs_gpt5_codex_default() {
|
||||
let mut features = Features::with_defaults();
|
||||
features.enable(Feature::CollaborationModes);
|
||||
assert_model_tools(
|
||||
assert_default_model_tools(
|
||||
"gpt-5-codex",
|
||||
&features,
|
||||
Some(WebSearchMode::Cached),
|
||||
"shell_command",
|
||||
&[
|
||||
"shell_command",
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
|
|
@ -1757,12 +1773,12 @@ mod tests {
|
|||
fn test_build_specs_gpt51_codex_default() {
|
||||
let mut features = Features::with_defaults();
|
||||
features.enable(Feature::CollaborationModes);
|
||||
assert_model_tools(
|
||||
assert_default_model_tools(
|
||||
"gpt-5.1-codex",
|
||||
&features,
|
||||
Some(WebSearchMode::Cached),
|
||||
"shell_command",
|
||||
&[
|
||||
"shell_command",
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
|
|
@ -1827,12 +1843,12 @@ mod tests {
|
|||
fn test_codex_mini_defaults() {
|
||||
let mut features = Features::with_defaults();
|
||||
features.enable(Feature::CollaborationModes);
|
||||
assert_model_tools(
|
||||
assert_default_model_tools(
|
||||
"codex-mini-latest",
|
||||
&features,
|
||||
Some(WebSearchMode::Cached),
|
||||
"local_shell",
|
||||
&[
|
||||
"local_shell",
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
|
|
@ -1848,12 +1864,12 @@ mod tests {
|
|||
fn test_codex_5_1_mini_defaults() {
|
||||
let mut features = Features::with_defaults();
|
||||
features.enable(Feature::CollaborationModes);
|
||||
assert_model_tools(
|
||||
assert_default_model_tools(
|
||||
"gpt-5.1-codex-mini",
|
||||
&features,
|
||||
Some(WebSearchMode::Cached),
|
||||
"shell_command",
|
||||
&[
|
||||
"shell_command",
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
|
|
@ -1870,12 +1886,12 @@ mod tests {
|
|||
fn test_gpt_5_defaults() {
|
||||
let mut features = Features::with_defaults();
|
||||
features.enable(Feature::CollaborationModes);
|
||||
assert_model_tools(
|
||||
assert_default_model_tools(
|
||||
"gpt-5",
|
||||
&features,
|
||||
Some(WebSearchMode::Cached),
|
||||
"shell",
|
||||
&[
|
||||
"shell",
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
|
|
@ -1891,12 +1907,12 @@ mod tests {
|
|||
fn test_gpt_5_1_defaults() {
|
||||
let mut features = Features::with_defaults();
|
||||
features.enable(Feature::CollaborationModes);
|
||||
assert_model_tools(
|
||||
assert_default_model_tools(
|
||||
"gpt-5.1",
|
||||
&features,
|
||||
Some(WebSearchMode::Cached),
|
||||
"shell_command",
|
||||
&[
|
||||
"shell_command",
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
|
|
|
|||
|
|
@ -50,6 +50,16 @@ async fn collect_tool_identifiers_for_model(model: &str) -> Vec<String> {
|
|||
tool_identifiers(&body)
|
||||
}
|
||||
|
||||
fn expected_default_tools(shell_tool: &str, tail: &[&str]) -> Vec<String> {
|
||||
let mut tools = if cfg!(windows) {
|
||||
vec![shell_tool.to_string()]
|
||||
} else {
|
||||
vec!["exec_command".to_string(), "write_stdin".to_string()]
|
||||
};
|
||||
tools.extend(tail.iter().map(|tool| (*tool).to_string()));
|
||||
tools
|
||||
}
|
||||
|
||||
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
|
||||
async fn model_selects_expected_tools() {
|
||||
skip_if_no_network!();
|
||||
|
|
@ -58,83 +68,93 @@ async fn model_selects_expected_tools() {
|
|||
let codex_tools = collect_tool_identifiers_for_model("codex-mini-latest").await;
|
||||
assert_eq!(
|
||||
codex_tools,
|
||||
vec![
|
||||
"local_shell".to_string(),
|
||||
"list_mcp_resources".to_string(),
|
||||
"list_mcp_resource_templates".to_string(),
|
||||
"read_mcp_resource".to_string(),
|
||||
"update_plan".to_string(),
|
||||
"request_user_input".to_string(),
|
||||
"web_search".to_string(),
|
||||
"view_image".to_string()
|
||||
],
|
||||
expected_default_tools(
|
||||
"local_shell",
|
||||
&[
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
"update_plan",
|
||||
"request_user_input",
|
||||
"web_search",
|
||||
"view_image",
|
||||
],
|
||||
),
|
||||
"codex-mini-latest should expose the local shell tool",
|
||||
);
|
||||
|
||||
let gpt5_codex_tools = collect_tool_identifiers_for_model("gpt-5-codex").await;
|
||||
assert_eq!(
|
||||
gpt5_codex_tools,
|
||||
vec![
|
||||
"shell_command".to_string(),
|
||||
"list_mcp_resources".to_string(),
|
||||
"list_mcp_resource_templates".to_string(),
|
||||
"read_mcp_resource".to_string(),
|
||||
"update_plan".to_string(),
|
||||
"request_user_input".to_string(),
|
||||
"apply_patch".to_string(),
|
||||
"web_search".to_string(),
|
||||
"view_image".to_string()
|
||||
],
|
||||
expected_default_tools(
|
||||
"shell_command",
|
||||
&[
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
"update_plan",
|
||||
"request_user_input",
|
||||
"apply_patch",
|
||||
"web_search",
|
||||
"view_image",
|
||||
],
|
||||
),
|
||||
"gpt-5-codex should expose the apply_patch tool",
|
||||
);
|
||||
|
||||
let gpt51_codex_tools = collect_tool_identifiers_for_model("gpt-5.1-codex").await;
|
||||
assert_eq!(
|
||||
gpt51_codex_tools,
|
||||
vec![
|
||||
"shell_command".to_string(),
|
||||
"list_mcp_resources".to_string(),
|
||||
"list_mcp_resource_templates".to_string(),
|
||||
"read_mcp_resource".to_string(),
|
||||
"update_plan".to_string(),
|
||||
"request_user_input".to_string(),
|
||||
"apply_patch".to_string(),
|
||||
"web_search".to_string(),
|
||||
"view_image".to_string()
|
||||
],
|
||||
expected_default_tools(
|
||||
"shell_command",
|
||||
&[
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
"update_plan",
|
||||
"request_user_input",
|
||||
"apply_patch",
|
||||
"web_search",
|
||||
"view_image",
|
||||
],
|
||||
),
|
||||
"gpt-5.1-codex should expose the apply_patch tool",
|
||||
);
|
||||
|
||||
let gpt5_tools = collect_tool_identifiers_for_model("gpt-5").await;
|
||||
assert_eq!(
|
||||
gpt5_tools,
|
||||
vec![
|
||||
"shell".to_string(),
|
||||
"list_mcp_resources".to_string(),
|
||||
"list_mcp_resource_templates".to_string(),
|
||||
"read_mcp_resource".to_string(),
|
||||
"update_plan".to_string(),
|
||||
"request_user_input".to_string(),
|
||||
"web_search".to_string(),
|
||||
"view_image".to_string()
|
||||
],
|
||||
expected_default_tools(
|
||||
"shell",
|
||||
&[
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
"update_plan",
|
||||
"request_user_input",
|
||||
"web_search",
|
||||
"view_image",
|
||||
],
|
||||
),
|
||||
"gpt-5 should expose the apply_patch tool",
|
||||
);
|
||||
|
||||
let gpt51_tools = collect_tool_identifiers_for_model("gpt-5.1").await;
|
||||
assert_eq!(
|
||||
gpt51_tools,
|
||||
vec![
|
||||
"shell_command".to_string(),
|
||||
"list_mcp_resources".to_string(),
|
||||
"list_mcp_resource_templates".to_string(),
|
||||
"read_mcp_resource".to_string(),
|
||||
"update_plan".to_string(),
|
||||
"request_user_input".to_string(),
|
||||
"apply_patch".to_string(),
|
||||
"web_search".to_string(),
|
||||
"view_image".to_string()
|
||||
],
|
||||
expected_default_tools(
|
||||
"shell_command",
|
||||
&[
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
"update_plan",
|
||||
"request_user_input",
|
||||
"apply_patch",
|
||||
"web_search",
|
||||
"view_image",
|
||||
],
|
||||
),
|
||||
"gpt-5.1 should expose the apply_patch tool",
|
||||
);
|
||||
let exp_tools = collect_tool_identifiers_for_model("exp-5.1").await;
|
||||
|
|
|
|||
|
|
@ -136,8 +136,12 @@ async fn prompt_tools_are_consistent_across_requests() -> anyhow::Result<()> {
|
|||
.await?;
|
||||
wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await;
|
||||
|
||||
let expected_tools_names = vec![
|
||||
"shell_command",
|
||||
let mut expected_tools_names = if cfg!(windows) {
|
||||
vec!["shell_command"]
|
||||
} else {
|
||||
vec!["exec_command", "write_stdin"]
|
||||
};
|
||||
expected_tools_names.extend([
|
||||
"list_mcp_resources",
|
||||
"list_mcp_resource_templates",
|
||||
"read_mcp_resource",
|
||||
|
|
@ -146,7 +150,7 @@ async fn prompt_tools_are_consistent_across_requests() -> anyhow::Result<()> {
|
|||
"apply_patch",
|
||||
"web_search",
|
||||
"view_image",
|
||||
];
|
||||
]);
|
||||
let body0 = req1.single_request().body_json();
|
||||
|
||||
let expected_instructions = if expected_tools_names.contains(&"apply_patch") {
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue