openai/codex
1
0
Fork 0
mirror of https://github.com/openai/codex.git synced 2026-09-06 15:29:32 +00:00
Code Issues Packages Projects Releases Wiki Activity

clarify response model metadata selection

Browse Source
This commit is contained in:
GLENNA ZHANG
2026-06-12 11:35:02 -04:00
parent 47f0fb5ae9
commit eb4ff63c7e
2 changed files with 15 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
codex-rs/codex-api/src/endpoint/responses_websocket.rs
View File

@@ -693,7 +693,7 @@ async fn run_websocket_response_stream(
continue;
}
let model_metadata = event.response_model_metadata();
if let Some(model) = model_metadata.warning_model
if let Some(model) = model_metadata.response_model
&& last_server_model.as_deref() != Some(model.as_str())
{
let _ = tx_event

23
codex-rs/codex-api/src/sse/responses.rs
View File

@@ -179,7 +179,12 @@ pub struct ResponsesStreamEvent {
#[derive(Debug, PartialEq, Eq)]
pub(crate) struct ResponseModelMetadata {
pub(crate) warning_model: Option<String>,
/// Effective model reported in the response metadata, if present.
///
/// Selection precedence preserves the existing model-mismatch/reroute warning behavior:
/// 1. `response.headers` on standard Responses stream events.
/// 2. Top-level `headers` on WebSocket metadata events.
pub(crate) response_model: Option<String>,
pub(crate) execution_identity: ResponseModelIdentity,
}
@@ -188,7 +193,7 @@ impl ResponsesStreamEvent {
&self.kind
}
/// Parses model headers once while preserving their two selection rules.
/// Parses model headers once for the response-model and execution-identity selections.
pub(crate) fn response_model_metadata(&self) -> ResponseModelMetadata {
let response_identity = self
.response
@@ -199,7 +204,7 @@ impl ResponsesStreamEvent {
.headers
.as_ref()
.and_then(response_model_identity_from_json);
let warning_model = response_identity
let response_model = response_identity
.as_ref()
.and_then(|identity| identity.final_model.clone())
.or_else(|| {
@@ -212,7 +217,7 @@ impl ResponsesStreamEvent {
let execution_identity = response_identity.or(top_level_identity).unwrap_or_default();
ResponseModelMetadata {
warning_model,
response_model,
execution_identity,
}
}
@@ -500,7 +505,7 @@ pub async fn process_sse(
let turn_moderation_metadata = event.turn_moderation_metadata();
let model_metadata = event.response_model_metadata();
if let Some(model) = model_metadata.warning_model
if let Some(model) = model_metadata.response_model
&& last_server_model.as_deref() != Some(model.as_str())
{
if tx_event
@@ -1382,7 +1387,7 @@ mod tests {
assert_eq!(
ev.response_model_metadata(),
ResponseModelMetadata {
warning_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
response_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
execution_identity: ResponseModelIdentity {
final_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
model_snapshot: Some(MODEL_SNAPSHOT_FOR_TESTS.to_string()),
@@ -1412,7 +1417,7 @@ mod tests {
assert_eq!(
ev.response_model_metadata(),
ResponseModelMetadata {
warning_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
response_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
execution_identity: ResponseModelIdentity {
final_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
model_snapshot: Some(MODEL_SNAPSHOT_FOR_TESTS.to_string()),
@@ -1440,7 +1445,7 @@ mod tests {
assert_eq!(
ev.response_model_metadata(),
ResponseModelMetadata {
warning_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
response_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
execution_identity: ResponseModelIdentity {
final_model: Some(CYBER_RESTRICTED_MODEL_FOR_TESTS.to_string()),
model_snapshot: None,
@@ -1468,7 +1473,7 @@ mod tests {
assert_eq!(
ev.response_model_metadata(),
ResponseModelMetadata {
warning_model: Some("top-level-model".to_string()),
response_model: Some("top-level-model".to_string()),
execution_identity: ResponseModelIdentity {
final_model: None,
model_snapshot: Some(MODEL_SNAPSHOT_FOR_TESTS.to_string()),