fix(protocol): align trace schemas with v2/v3 classify rules

RecordedTrace validation now follows classify_value — numeric version
selects v3, otherwise v2 — so mixed pages/states and v2 envelopes with
a version field fail Schema checks before Recall import. Keep standalone
trace_step as a v2|v3 union for tools that validate a single step.
This commit is contained in:
“haonan”
2026-08-23 21:26:39 +08:00
committed by GitHub
parent e64f4cac9f
commit c2a225377b
11 changed files with 1784 additions and 320 deletions
+224 -14
View File
@@ -16,6 +16,7 @@ pub use super::record_v3::{
VOM_FORMAT_VERSION,
};
/// Logical v2 identifier. Not a wire field — v2 envelopes omit `version`.
pub const TRACE_VERSION_V2: u32 = 2;
// ---------------------------------------------------------------------------
@@ -45,6 +46,10 @@ pub struct RecordStopParams {
}
/// Wire trace payload — v2 (legacy `pages[]`) or v3 (`version: 3`, `states[]`).
///
/// Classification matches [`RecordedTrace::classify_value`]: a numeric `version`
/// selects v3 (and forbids `pages[]`); otherwise the envelope is v2 and must
/// not include `states[]`.
#[derive(Debug, Clone, PartialEq)]
pub enum RecordedTrace {
V2(TraceV2),
@@ -104,12 +109,134 @@ impl JsonSchema for RecordedTrace {
}
fn json_schema(generator: &mut schemars::SchemaGenerator) -> schemars::schema::Schema {
// Mirror classify_value: a numeric `version` selects v3; otherwise v2.
// TraceV2 stays open for unknown extension fields but forbids `states[]`
// and a numeric `version`. TraceV3 already denies `pages[]`.
let mut version_props = schemars::Map::new();
version_props.insert(
"version".into(),
schemars::schema::SchemaObject {
instance_type: Some(schemars::schema::InstanceType::Integer.into()),
..Default::default()
}
.into(),
);
let mut required = schemars::Set::new();
required.insert("version".into());
schemars::schema::SchemaObject {
subschemas: Some(Box::new(schemars::schema::SubschemaValidation {
one_of: Some(vec![
generator.subschema_for::<TraceV2>(),
generator.subschema_for::<TraceV3>(),
]),
if_schema: Some(Box::new(
schemars::schema::SchemaObject {
object: Some(Box::new(schemars::schema::ObjectValidation {
properties: version_props,
required,
..Default::default()
})),
..Default::default()
}
.into(),
)),
then_schema: Some(Box::new(generator.subschema_for::<TraceV3>())),
else_schema: Some(Box::new(generator.subschema_for::<TraceV2>())),
..Default::default()
})),
..Default::default()
}
.into()
}
}
/// One recorded step — v2 (`page`) or v3 (`state` / `result.state`).
///
/// Classification matches [`RecordedStep::classify_value`]: `state` selects v3
/// (and forbids `page`); otherwise the step is v2 and must include `page`.
#[derive(Debug, Clone, PartialEq)]
pub enum RecordedStep {
V2(StepV2),
V3(StepV3),
}
impl RecordedStep {
pub fn classify_value(v: &serde_json::Value) -> Result<Self, String> {
if v.get("state").is_some() {
if v.get("page").is_some() {
return Err("step v3 must not include legacy page".into());
}
return serde_json::from_value(v.clone())
.map(RecordedStep::V3)
.map_err(|e| e.to_string());
}
if v.get("page").is_some() {
return serde_json::from_value(v.clone())
.map(RecordedStep::V2)
.map_err(|e| e.to_string());
}
Err("ambiguous or unparseable step".into())
}
}
impl Serialize for RecordedStep {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
match self {
RecordedStep::V2(s) => s.serialize(serializer),
RecordedStep::V3(s) => s.serialize(serializer),
}
}
}
impl<'de> Deserialize<'de> for RecordedStep {
fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
let value = serde_json::Value::deserialize(deserializer)?;
Self::classify_value(&value).map_err(serde::de::Error::custom)
}
}
impl JsonSchema for RecordedStep {
fn schema_name() -> String {
"RecordedStep".into()
}
fn json_schema(generator: &mut schemars::SchemaGenerator) -> schemars::schema::Schema {
// Mirror classify_value: `state` selects v3; otherwise v2.
// Mixed page + state is rejected at this layer because StepV2/StepV3
// still allow unknown extension fields.
let mut required = schemars::Set::new();
required.insert("state".into());
let mut mixed_keys = schemars::Set::new();
mixed_keys.insert("page".into());
mixed_keys.insert("state".into());
schemars::schema::SchemaObject {
subschemas: Some(Box::new(schemars::schema::SubschemaValidation {
if_schema: Some(Box::new(
schemars::schema::SchemaObject {
object: Some(Box::new(schemars::schema::ObjectValidation {
required,
..Default::default()
})),
..Default::default()
}
.into(),
)),
then_schema: Some(Box::new(generator.subschema_for::<StepV3>())),
else_schema: Some(Box::new(generator.subschema_for::<StepV2>())),
not: Some(Box::new(
schemars::schema::SchemaObject {
object: Some(Box::new(schemars::schema::ObjectValidation {
required: mixed_keys,
..Default::default()
})),
..Default::default()
}
.into(),
)),
..Default::default()
})),
..Default::default()
@@ -204,26 +331,43 @@ mod tests {
assert!(serde_json::from_value::<TraceV3>(unsupported).is_err());
}
fn assert_classifies_by_integer_version(schema: &serde_json::Value) {
assert_eq!(
schema["if"]["required"],
json!(["version"]),
"numeric version is the RecordedTrace classification key"
);
assert_eq!(schema["if"]["properties"]["version"]["type"], "integer");
assert_eq!(schema["then"]["$ref"], "#/definitions/TraceV3");
assert_eq!(schema["else"]["$ref"], "#/definitions/TraceV2");
}
#[test]
fn recorded_trace_schema_includes_v2_and_v3() {
let schema = serde_json::to_value(schemars::schema_for!(RecordStopResult)).unwrap();
let variants = schema["definitions"]["RecordedTrace"]["oneOf"]
.as_array()
.expect("RecordedTrace schema should use oneOf");
assert_classifies_by_integer_version(&schema["definitions"]["RecordedTrace"]);
assert_eq!(
schema["definitions"]["TraceV3"]["properties"]["version"]["const"],
3
);
}
assert_eq!(variants.len(), 2);
let trace_schema = schema["definitions"]["TraceV3"].clone();
assert_eq!(trace_schema["properties"]["version"]["const"], 3);
#[test]
fn recorded_trace_schema_follows_classify_value() {
let standalone = serde_json::to_value(schemars::schema_for!(RecordedTrace)).unwrap();
assert_classifies_by_integer_version(&standalone);
assert!(standalone["definitions"].get("TraceV2").is_some());
assert!(standalone["definitions"].get("TraceV3").is_some());
let stop_result = serde_json::to_value(schemars::schema_for!(RecordStopResult)).unwrap();
assert_classifies_by_integer_version(&stop_result["definitions"]["RecordedTrace"]);
}
#[test]
fn standalone_trace_schema_is_recorded_trace_union() {
let schema = serde_json::to_value(schemars::schema_for!(RecordedTrace)).unwrap();
assert_eq!(schema["title"], "RecordedTrace");
let variants = schema["oneOf"]
.as_array()
.expect("standalone trace schema should be a v2|v3 oneOf");
assert_eq!(variants.len(), 2);
assert_classifies_by_integer_version(&schema);
assert!(schema["definitions"].get("TraceV2").is_some());
assert!(schema["definitions"].get("TraceV3").is_some());
}
@@ -242,4 +386,70 @@ mod tests {
other => panic!("expected v2, got {other:?}"),
}
}
fn v2_click() -> serde_json::Value {
json!({
"op": "click",
"id": 1,
"page": "p1",
"target": { "tag": "button", "role": "button", "name": "发布" }
})
}
fn v3_click() -> serde_json::Value {
json!({
"op": "click",
"id": 1,
"state": "s1",
"result": { "state": "s2" },
"target": { "ref": "e1", "role": "button", "name": "发布" }
})
}
fn assert_step_classifies_by_state(schema: &serde_json::Value) {
assert_eq!(
schema["if"]["required"],
json!(["state"]),
"state is the RecordedStep classification key"
);
assert_eq!(schema["then"]["$ref"], "#/definitions/StepV3");
assert_eq!(schema["else"]["$ref"], "#/definitions/StepV2");
let mut forbidden = schema["not"]["required"]
.as_array()
.expect("RecordedStep schema must reject mixed page + state steps")
.iter()
.filter_map(|v| v.as_str())
.collect::<Vec<_>>();
forbidden.sort_unstable();
assert_eq!(forbidden, ["page", "state"]);
}
#[test]
fn recorded_step_classifies_v2_and_v3() {
match RecordedStep::classify_value(&v2_click()).unwrap() {
RecordedStep::V2(StepV2::Click { .. }) => {}
other => panic!("expected v2 click, got {other:?}"),
}
match RecordedStep::classify_value(&v3_click()).unwrap() {
RecordedStep::V3(StepV3::Click { .. }) => {}
other => panic!("expected v3 click, got {other:?}"),
}
}
#[test]
fn recorded_step_rejects_mixed_page_and_state() {
let mut mixed = v2_click();
mixed["state"] = json!("s1");
mixed["result"] = json!({ "state": "s2" });
assert!(RecordedStep::classify_value(&mixed).is_err());
}
#[test]
fn recorded_step_schema_follows_classify_value() {
let schema = serde_json::to_value(schemars::schema_for!(RecordedStep)).unwrap();
assert_eq!(schema["title"], "RecordedStep");
assert_step_classifies_by_state(&schema);
assert!(schema["definitions"].get("StepV2").is_some());
assert!(schema["definitions"].get("StepV3").is_some());
}
}
+81 -2
View File
@@ -107,8 +107,9 @@ pub enum StepV2 {
/// Persisted user-action trace exported by legacy `tool.record_stop` / `await`.
///
/// Unknown extension fields are ignored so older traces remain readable.
/// Mixed v2/v3 envelopes are rejected by `RecordedTrace` classification.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, JsonSchema)]
/// `states[]` and a numeric `version` are reserved for Trace v3 / `RecordedTrace`
/// classification — serde still ignores them, but the JSON Schema rejects them.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
pub struct TraceV2 {
/// RFC 3339 timestamp when recording stopped.
pub recorded_at: String,
@@ -121,6 +122,70 @@ pub struct TraceV2 {
pub steps: Vec<StepV2>,
}
fn constrain_trace_v2_schema(mut schema: schemars::schema::Schema) -> schemars::schema::Schema {
let schemars::schema::Schema::Object(obj) = &mut schema else {
return schema;
};
obj.metadata().title = Some("TraceV2".into());
obj.metadata().description = Some(
"Persisted user-action trace exported by legacy `tool.record_stop` / `await`.\n\n\
Unknown extension fields are ignored so older traces remain readable. \
`states[]` and a numeric `version` are reserved for Trace v3 / `RecordedTrace` classification."
.into(),
);
let object = obj.object.get_or_insert_with(Default::default);
object
.properties
.insert("states".into(), schemars::schema::Schema::Bool(false));
object.properties.insert(
"version".into(),
schemars::schema::SchemaObject {
metadata: Some(Box::new(schemars::schema::Metadata {
description: Some(
"Numeric version selects Trace v3. Legacy v2 envelopes omit this field.".into(),
),
..Default::default()
})),
subschemas: Some(Box::new(schemars::schema::SubschemaValidation {
not: Some(Box::new(
schemars::schema::SchemaObject {
instance_type: Some(schemars::schema::InstanceType::Integer.into()),
..Default::default()
}
.into(),
)),
..Default::default()
})),
..Default::default()
}
.into(),
);
schema
}
impl JsonSchema for TraceV2 {
fn schema_name() -> String {
"TraceV2".into()
}
fn json_schema(generator: &mut schemars::SchemaGenerator) -> schemars::schema::Schema {
#[derive(JsonSchema)]
#[allow(dead_code)]
struct TraceV2Shape {
/// RFC 3339 timestamp when recording stopped.
recorded_at: String,
started_at: Option<String>,
purpose: Option<String>,
entry: TraceEntry,
pages: Vec<PageRefV2>,
steps: Vec<StepV2>,
}
constrain_trace_v2_schema(TraceV2Shape::json_schema(generator))
}
}
#[cfg(test)]
mod tests {
use super::*;
@@ -248,4 +313,18 @@ mod tests {
"Trace v2 must keep accepting traces with unknown extension fields"
);
}
#[test]
fn trace_v2_schema_forbids_classification_keys() {
let schema = serde_json::to_value(schemars::schema_for!(TraceV2)).unwrap();
assert_eq!(
schema["properties"]["states"],
json!(false),
"Trace v2 schema must reject states[] (reserved for v3)"
);
assert_eq!(
schema["properties"]["version"]["not"]["type"], "integer",
"Trace v2 schema must reject a numeric version (that selects the v3 path)"
);
}
}