mirror of
https://github.com/pchuan98/codex.git
synced 2026-07-01 00:31:56 +08:00
## Description This PR adds an optional `metadata` field to `ResponseItem` for Responses API calls. Only mechanical plumbing, no actual values populated and sent yet. Turns out just adding a new field to `ResponseItem` has quite a large blast radius already. This change is backwards compatible because `metadata` is optional and omitted when absent, so existing response items and rollout history without it still deserialize and requests that do not set it keep the same wire shape. For provider compatibility, we strip out `metadata` before non-OpenAI Responses requests so Azure and AWS Bedrock never see this field. My followup PR here will actually make use of it to start storing and passing along `turn_id`: https://github.com/openai/codex/pull/28360 ## What changed - Added `ResponseItemMetadata` with optional `turn_id`, plus optional `metadata` on Responses API item variants and inter-agent communication. - Preserved item metadata through response-item rewrites such as truncation, missing tool-output synthesis, compaction history rebuilding, visible-history conversion, rollout/resume, and generated app-server schemas/types. - Strip item metadata from non-OpenAI Responses requests while preserving it for OpenAI-shaped requests. - Updated the mechanical fixture/test construction churn required by the new optional field.
329 lines
10 KiB
Rust
329 lines
10 KiB
Rust
use codex_api::ImageBackground;
|
|
use codex_api::ImageEditRequest;
|
|
use codex_api::ImageGenerationRequest;
|
|
use codex_api::ImageQuality;
|
|
use codex_api::ImageUrl;
|
|
use codex_core::context::extension_image_generation_output_hint;
|
|
use codex_extension_api::ToolOutput;
|
|
use codex_extension_api::ToolPayload;
|
|
use codex_extension_api::ToolSpec;
|
|
use codex_protocol::models::ContentItem;
|
|
use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
|
|
use codex_protocol::models::FunctionCallOutputBody;
|
|
use codex_protocol::models::FunctionCallOutputContentItem;
|
|
use codex_protocol::models::FunctionCallOutputPayload;
|
|
use codex_protocol::models::ResponseInputItem;
|
|
use codex_protocol::models::ResponseItem;
|
|
use codex_tools::ResponsesApiNamespaceTool;
|
|
use pretty_assertions::assert_eq;
|
|
|
|
use super::GeneratedImageOutput;
|
|
use super::ImageRequest;
|
|
use super::ImagegenArgs;
|
|
use super::imagegen_tool_spec;
|
|
use super::request_for_call_args;
|
|
use crate::IMAGE_GEN_NAMESPACE;
|
|
use crate::IMAGEGEN_TOOL_NAME;
|
|
|
|
const RESULT: &str = "cG5n";
|
|
|
|
#[test]
|
|
fn uses_reserved_image_gen_namespace() {
|
|
let ToolSpec::Namespace(spec) = imagegen_tool_spec() else {
|
|
panic!("imagegen should advertise a namespace tool");
|
|
};
|
|
assert_eq!(spec.name, IMAGE_GEN_NAMESPACE);
|
|
let ResponsesApiNamespaceTool::Function(function) = &spec.tools[0];
|
|
assert_eq!(function.name, IMAGEGEN_TOOL_NAME);
|
|
}
|
|
|
|
#[tokio::test]
|
|
async fn omitted_references_generate_with_fixed_defaults() {
|
|
assert_eq!(
|
|
request_for_call_args(
|
|
&ImagegenArgs {
|
|
prompt: "paint a moonlit lake".to_string(),
|
|
referenced_image_paths: None,
|
|
num_last_images_to_include: None,
|
|
},
|
|
&[],
|
|
&[],
|
|
)
|
|
.await
|
|
.expect("generation request should build"),
|
|
ImageRequest::Generate(ImageGenerationRequest {
|
|
prompt: "paint a moonlit lake".to_string(),
|
|
background: Some(ImageBackground::Auto),
|
|
model: "gpt-image-2".to_string(),
|
|
n: None,
|
|
quality: Some(ImageQuality::Auto),
|
|
size: Some("auto".to_string()),
|
|
})
|
|
);
|
|
}
|
|
|
|
#[tokio::test]
|
|
async fn recent_image_fallback_selects_newest_images_in_chronological_order() {
|
|
let history = vec![
|
|
ResponseItem::Message {
|
|
id: None,
|
|
role: "user".to_string(),
|
|
content: vec![
|
|
input_image("user-1"),
|
|
input_image("user-2"),
|
|
ContentItem::InputText {
|
|
text: "edit these".to_string(),
|
|
},
|
|
],
|
|
phase: None,
|
|
metadata: None,
|
|
},
|
|
ResponseItem::FunctionCall {
|
|
id: None,
|
|
name: "mcp_image".to_string(),
|
|
namespace: None,
|
|
arguments: "{}".to_string(),
|
|
call_id: "mcp-call".to_string(),
|
|
metadata: None,
|
|
},
|
|
ResponseItem::FunctionCallOutput {
|
|
call_id: "mcp-call".to_string(),
|
|
output: image_output("mcp"),
|
|
metadata: None,
|
|
},
|
|
ResponseItem::CustomToolCall {
|
|
id: None,
|
|
status: Some("completed".to_string()),
|
|
call_id: "code-mode-call".to_string(),
|
|
name: "exec".to_string(),
|
|
input: String::new(),
|
|
metadata: None,
|
|
},
|
|
ResponseItem::CustomToolCallOutput {
|
|
call_id: "code-mode-call".to_string(),
|
|
name: Some("exec".to_string()),
|
|
output: image_output("code-mode"),
|
|
metadata: None,
|
|
},
|
|
ResponseItem::ImageGenerationCall {
|
|
id: "generated-call".to_string(),
|
|
status: "completed".to_string(),
|
|
revised_prompt: None,
|
|
result: "generated".to_string(),
|
|
metadata: None,
|
|
},
|
|
ResponseItem::FunctionCallOutput {
|
|
call_id: "orphan-call".to_string(),
|
|
output: image_output("orphan"),
|
|
metadata: None,
|
|
},
|
|
];
|
|
|
|
assert_eq!(
|
|
request_for_call_args(
|
|
&ImagegenArgs {
|
|
prompt: "change the lighting".to_string(),
|
|
referenced_image_paths: None,
|
|
num_last_images_to_include: Some(4),
|
|
},
|
|
&history,
|
|
&[],
|
|
)
|
|
.await
|
|
.expect("history-backed edit request should build"),
|
|
ImageRequest::Edit(expected_edit_request(
|
|
"change the lighting",
|
|
&["user-2", "mcp", "code-mode", "generated"],
|
|
))
|
|
);
|
|
}
|
|
|
|
#[tokio::test]
|
|
async fn conflicting_image_selectors_return_tool_error() {
|
|
let error = request_for_call_args(
|
|
&ImagegenArgs {
|
|
prompt: "change the lighting".to_string(),
|
|
referenced_image_paths: Some(vec![
|
|
"/tmp/image.png"
|
|
.try_into()
|
|
.expect("test path should be absolute"),
|
|
]),
|
|
num_last_images_to_include: Some(1),
|
|
},
|
|
&[],
|
|
&[],
|
|
)
|
|
.await
|
|
.expect_err("conflicting selectors should fail");
|
|
|
|
assert_eq!(
|
|
error.to_string(),
|
|
"provide only one of `referenced_image_paths` or `num_last_images_to_include`"
|
|
);
|
|
}
|
|
|
|
#[tokio::test]
|
|
async fn too_many_referenced_image_paths_return_tool_error() {
|
|
let error = request_for_call_args(
|
|
&ImagegenArgs {
|
|
prompt: "change the lighting".to_string(),
|
|
referenced_image_paths: Some(
|
|
(0..6)
|
|
.map(|index| {
|
|
format!("/tmp/image-{index}.png")
|
|
.try_into()
|
|
.expect("test path should be absolute")
|
|
})
|
|
.collect(),
|
|
),
|
|
num_last_images_to_include: None,
|
|
},
|
|
&[],
|
|
&[],
|
|
)
|
|
.await
|
|
.expect_err("too many paths should fail before reading files");
|
|
|
|
assert_eq!(
|
|
error.to_string(),
|
|
"`referenced_image_paths` must contain at most 5 paths"
|
|
);
|
|
}
|
|
|
|
#[tokio::test]
|
|
async fn recent_image_fallback_requires_requested_count() {
|
|
let error = request_for_call_args(
|
|
&ImagegenArgs {
|
|
prompt: "change the lighting".to_string(),
|
|
referenced_image_paths: None,
|
|
num_last_images_to_include: Some(2),
|
|
},
|
|
&[ResponseItem::Message {
|
|
id: None,
|
|
role: "user".to_string(),
|
|
content: vec![input_image("only-image")],
|
|
phase: None,
|
|
metadata: None,
|
|
}],
|
|
&[],
|
|
)
|
|
.await
|
|
.expect_err("history-backed edit should require the requested image count");
|
|
|
|
assert_eq!(
|
|
error.to_string(),
|
|
"requested the last 2 conversation images, but only 1 were available"
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn generated_output_returns_image_input_and_output_hint() {
|
|
let output_hint =
|
|
extension_image_generation_output_hint("/tmp", "/tmp/call-1.png").expect("hint should fit");
|
|
let output = GeneratedImageOutput {
|
|
result: RESULT.to_string(),
|
|
output_hint: Some(output_hint.clone()),
|
|
};
|
|
|
|
let ResponseInputItem::FunctionCallOutput {
|
|
output: response_output,
|
|
..
|
|
} = output.to_response_item("call-1", &function_payload())
|
|
else {
|
|
panic!("imagegen should return function tool output");
|
|
};
|
|
let FunctionCallOutputBody::ContentItems(content_items) = response_output.body else {
|
|
panic!("imagegen output should contain generated image bytes");
|
|
};
|
|
assert_eq!(
|
|
content_items,
|
|
vec![
|
|
FunctionCallOutputContentItem::InputImage {
|
|
image_url: format!("data:image/png;base64,{RESULT}"),
|
|
detail: Some(DEFAULT_IMAGE_DETAIL),
|
|
},
|
|
FunctionCallOutputContentItem::InputText { text: output_hint },
|
|
]
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn generated_output_returns_generated_image_helper_input_in_code_mode() {
|
|
let output = GeneratedImageOutput {
|
|
result: RESULT.to_string(),
|
|
output_hint: Some("generated image save hint".to_string()),
|
|
};
|
|
|
|
assert_eq!(
|
|
output.code_mode_result(&function_payload()),
|
|
serde_json::json!({
|
|
"image_url": format!("data:image/png;base64,{RESULT}"),
|
|
"output_hint": "generated image save hint",
|
|
})
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn generated_output_omits_oversized_output_hint() {
|
|
let long_path = "x".repeat(1024);
|
|
let output = GeneratedImageOutput {
|
|
result: RESULT.to_string(),
|
|
output_hint: extension_image_generation_output_hint("/tmp", long_path),
|
|
};
|
|
|
|
let ResponseInputItem::FunctionCallOutput {
|
|
output: response_output,
|
|
..
|
|
} = output.to_response_item("call-1", &function_payload())
|
|
else {
|
|
panic!("imagegen should return function tool output");
|
|
};
|
|
let FunctionCallOutputBody::ContentItems(content_items) = response_output.body else {
|
|
panic!("imagegen output should contain generated image bytes");
|
|
};
|
|
assert_eq!(
|
|
content_items,
|
|
vec![FunctionCallOutputContentItem::InputImage {
|
|
image_url: format!("data:image/png;base64,{RESULT}"),
|
|
detail: Some(DEFAULT_IMAGE_DETAIL),
|
|
}]
|
|
);
|
|
}
|
|
|
|
fn input_image(image: &str) -> ContentItem {
|
|
ContentItem::InputImage {
|
|
image_url: format!("data:image/png;base64,{image}"),
|
|
detail: None,
|
|
}
|
|
}
|
|
|
|
fn image_output(image: &str) -> FunctionCallOutputPayload {
|
|
FunctionCallOutputPayload::from_content_items(vec![FunctionCallOutputContentItem::InputImage {
|
|
image_url: format!("data:image/png;base64,{image}"),
|
|
detail: None,
|
|
}])
|
|
}
|
|
|
|
fn expected_edit_request(prompt: &str, images: &[&str]) -> ImageEditRequest {
|
|
ImageEditRequest {
|
|
images: images
|
|
.iter()
|
|
.map(|image| ImageUrl {
|
|
image_url: format!("data:image/png;base64,{image}"),
|
|
})
|
|
.collect(),
|
|
prompt: prompt.to_string(),
|
|
background: Some(ImageBackground::Auto),
|
|
model: "gpt-image-2".to_string(),
|
|
n: None,
|
|
quality: Some(ImageQuality::Auto),
|
|
size: Some("auto".to_string()),
|
|
}
|
|
}
|
|
|
|
fn function_payload() -> ToolPayload {
|
|
ToolPayload::Function {
|
|
arguments: "{}".to_string(),
|
|
}
|
|
}
|