Files
llmx/codex-rs/core/src/models.rs
Michael Bolin b323d10ea7 feat: add ZDR support to Rust implementation (#642)
This adds support for the `--disable-response-storage` flag across our
multiple Rust CLIs to support customers who have opted into Zero-Data
Retention (ZDR). The analogous changes to the TypeScript CLI were:

* https://github.com/openai/codex/pull/481
* https://github.com/openai/codex/pull/543

For a client using ZDR, `previous_response_id` will never be available,
so the `input` field of an API request must include the full transcript
of the conversation thus far. As such, this PR changes the type of
`Prompt.input` from `Vec<ResponseInputItem>` to `Vec<ResponseItem>`.

Practically speaking, `ResponseItem` was effectively a "superset" of
`ResponseInputItem` already. The main difference for us is that
`ResponseItem` includes the `FunctionCall` variant that we have to
include as part of the conversation history in the ZDR case.

Another key change in this PR is modifying `try_run_turn()` so that it
returns the `Vec<ResponseItem>` for the turn in addition to the
`Vec<ResponseInputItem>` produced by `try_run_turn()`. This is because
the caller of `run_turn()` needs to record the `Vec<ResponseItem>` when
ZDR is enabled.

To that end, this PR introduces `ZdrTranscript` (and adds
`zdr_transcript: Option<ZdrTranscript>` to `struct State` in `codex.rs`)
to take responsibility for maintaining the conversation transcript in
the ZDR case.
2025-04-25 12:08:18 -07:00

187 lines
6.6 KiB
Rust
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
use base64::Engine;
use serde::ser::Serializer;
use serde::Deserialize;
use serde::Serialize;
use crate::protocol::InputItem;
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ResponseInputItem {
Message {
role: String,
content: Vec<ContentItem>,
},
FunctionCallOutput {
call_id: String,
output: FunctionCallOutputPayload,
},
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ContentItem {
InputText { text: String },
InputImage { image_url: String },
OutputText { text: String },
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ResponseItem {
Message {
role: String,
content: Vec<ContentItem>,
},
FunctionCall {
name: String,
// The Responses API returns the function call arguments as a *string* that contains
// JSON, not as an alreadyparsed object. We keep it as a raw string here and let
// Session::handle_function_call parse it into a Value. This exactly matches the
// Chat Completions + Responses API behavior.
arguments: String,
call_id: String,
},
// NOTE: The input schema for `function_call_output` objects that clients send to the
// OpenAI /v1/responses endpoint is NOT the same shape as the objects the server returns on the
// SSE stream. When *sending* we must wrap the string output inside an object that includes a
// required `success` boolean. The upstream TypeScript CLI does this implicitly. To ensure we
// serialize exactly the expected shape we introduce a dedicated payload struct and flatten it
// here.
FunctionCallOutput {
call_id: String,
output: FunctionCallOutputPayload,
},
#[serde(other)]
Other,
}
impl From<ResponseInputItem> for ResponseItem {
fn from(item: ResponseInputItem) -> Self {
match item {
ResponseInputItem::Message { role, content } => Self::Message { role, content },
ResponseInputItem::FunctionCallOutput { call_id, output } => {
Self::FunctionCallOutput { call_id, output }
}
}
}
}
impl From<Vec<InputItem>> for ResponseInputItem {
fn from(items: Vec<InputItem>) -> Self {
Self::Message {
role: "user".to_string(),
content: items
.into_iter()
.filter_map(|c| match c {
InputItem::Text { text } => Some(ContentItem::InputText { text }),
InputItem::Image { image_url } => Some(ContentItem::InputImage { image_url }),
InputItem::LocalImage { path } => match std::fs::read(&path) {
Ok(bytes) => {
let mime = mime_guess::from_path(&path)
.first()
.map(|m| m.essence_str().to_owned())
.unwrap_or_else(|| "application/octet-stream".to_string());
let encoded = base64::engine::general_purpose::STANDARD.encode(bytes);
Some(ContentItem::InputImage {
image_url: format!("data:{};base64,{}", mime, encoded),
})
}
Err(err) => {
tracing::warn!(
"Skipping image {} could not read file: {}",
path.display(),
err
);
None
}
},
})
.collect::<Vec<ContentItem>>(),
}
}
}
#[expect(dead_code)]
#[derive(Deserialize, Debug, Clone)]
pub struct FunctionCallOutputPayload {
pub content: String,
pub success: Option<bool>,
}
// The Responses API expects two *different* shapes depending on success vs failure:
// • success → output is a plain string (no nested object)
// • failure → output is an object { content, success:false }
// The upstream TypeScript CLI implements this by specialcasing the serialize path.
// We replicate that behavior with a manual Serialize impl.
impl Serialize for FunctionCallOutputPayload {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: Serializer,
{
// The upstream TypeScript CLI always serializes `output` as a *plain string* regardless
// of whether the function call succeeded or failed. The boolean is purely informational
// for local bookkeeping and is NOT sent to the OpenAI endpoint. Sending the nested object
// form `{ content, success:false }` triggers the 400 we are still seeing. Mirror the JS CLI
// exactly: always emit a bare string.
serializer.serialize_str(&self.content)
}
}
// Implement Display so callers can treat the payload like a plain string when logging or doing
// trivial substring checks in tests (existing tests call `.contains()` on the output). Display
// returns the raw `content` field.
impl std::fmt::Display for FunctionCallOutputPayload {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.write_str(&self.content)
}
}
impl std::ops::Deref for FunctionCallOutputPayload {
type Target = str;
fn deref(&self) -> &Self::Target {
&self.content
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn serializes_success_as_plain_string() {
let item = ResponseInputItem::FunctionCallOutput {
call_id: "call1".into(),
output: FunctionCallOutputPayload {
content: "ok".into(),
success: None,
},
};
let json = serde_json::to_string(&item).unwrap();
let v: serde_json::Value = serde_json::from_str(&json).unwrap();
// Success case -> output should be a plain string
assert_eq!(v.get("output").unwrap().as_str().unwrap(), "ok");
}
#[test]
fn serializes_failure_as_string() {
let item = ResponseInputItem::FunctionCallOutput {
call_id: "call1".into(),
output: FunctionCallOutputPayload {
content: "bad".into(),
success: Some(false),
},
};
let json = serde_json::to_string(&item).unwrap();
let v: serde_json::Value = serde_json::from_str(&json).unwrap();
assert_eq!(v.get("output").unwrap().as_str().unwrap(), "bad");
}
}