Move freshness smoke to Rust
This commit is contained in:
parent
a76a35f91d
commit
b7212a1bf3
8 changed files with 424 additions and 247 deletions
|
|
@ -480,6 +480,34 @@ impl AppServerClient {
|
|||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
pub fn wait_for_notification(
|
||||
&mut self,
|
||||
method: &str,
|
||||
start_index: usize,
|
||||
timeout: Duration,
|
||||
) -> Result<Value> {
|
||||
let deadline = Instant::now() + timeout;
|
||||
while Instant::now() < deadline {
|
||||
if let Some(status) = self.child.try_wait()? {
|
||||
return Err(anyhow!(
|
||||
"app-server exited with {} before notification {}",
|
||||
status,
|
||||
method
|
||||
));
|
||||
}
|
||||
if let Ok(notifications) = self.notifications.lock()
|
||||
&& let Some(message) = notifications
|
||||
.iter()
|
||||
.skip(start_index)
|
||||
.find(|message| message.get("method").and_then(Value::as_str) == Some(method))
|
||||
{
|
||||
return Ok(message.clone());
|
||||
}
|
||||
thread::sleep(Duration::from_millis(100));
|
||||
}
|
||||
Err(anyhow!("timed out waiting for notification {method}"))
|
||||
}
|
||||
}
|
||||
|
||||
impl Drop for AppServerClient {
|
||||
|
|
|
|||
389
epiphany-core/src/bin/epiphany-phase6-freshness-smoke.rs
Normal file
389
epiphany-core/src/bin/epiphany-phase6-freshness-smoke.rs
Normal file
|
|
@ -0,0 +1,389 @@
|
|||
use anyhow::Context;
|
||||
use anyhow::Result;
|
||||
use anyhow::anyhow;
|
||||
use serde_json::Value;
|
||||
use serde_json::json;
|
||||
use std::env;
|
||||
use std::fs;
|
||||
use std::path::{Path, PathBuf};
|
||||
use std::time::Duration;
|
||||
|
||||
#[allow(dead_code)]
|
||||
#[path = "epiphany-mvp-status.rs"]
|
||||
mod status_cli;
|
||||
|
||||
const DEFAULT_APP_SERVER: &str = r"C:\Users\Meta\.cargo-target-codex\debug\codex-app-server.exe";
|
||||
|
||||
fn main() -> Result<()> {
|
||||
let args = Args::parse()?;
|
||||
let result = run_smoke(&args)?;
|
||||
println!("{}", serde_json::to_string_pretty(&result)?);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[derive(Debug)]
|
||||
struct Args {
|
||||
app_server: PathBuf,
|
||||
codex_home: PathBuf,
|
||||
result: PathBuf,
|
||||
transcript: PathBuf,
|
||||
stderr: PathBuf,
|
||||
}
|
||||
|
||||
impl Args {
|
||||
fn parse() -> Result<Self> {
|
||||
let root = env::current_dir().context("failed to resolve current dir")?;
|
||||
let mut parsed = Self {
|
||||
app_server: PathBuf::from(DEFAULT_APP_SERVER),
|
||||
codex_home: root
|
||||
.join(".epiphany-smoke")
|
||||
.join("phase6-freshness-codex-home"),
|
||||
result: root
|
||||
.join(".epiphany-smoke")
|
||||
.join("phase6-freshness-smoke-result.json"),
|
||||
transcript: root
|
||||
.join(".epiphany-smoke")
|
||||
.join("phase6-freshness-smoke-transcript.jsonl"),
|
||||
stderr: root
|
||||
.join(".epiphany-smoke")
|
||||
.join("phase6-freshness-smoke-server.stderr.log"),
|
||||
};
|
||||
let mut args = env::args().skip(1);
|
||||
while let Some(arg) = args.next() {
|
||||
match arg.as_str() {
|
||||
"--app-server" => parsed.app_server = take_path(&mut args, "--app-server")?,
|
||||
"--codex-home" => parsed.codex_home = take_path(&mut args, "--codex-home")?,
|
||||
"--result" => parsed.result = take_path(&mut args, "--result")?,
|
||||
"--transcript" => parsed.transcript = take_path(&mut args, "--transcript")?,
|
||||
"--stderr" => parsed.stderr = take_path(&mut args, "--stderr")?,
|
||||
other => return Err(anyhow!("unknown argument: {other}")),
|
||||
}
|
||||
}
|
||||
Ok(parsed)
|
||||
}
|
||||
}
|
||||
|
||||
fn run_smoke(args: &Args) -> Result<Value> {
|
||||
let root = env::current_dir().context("failed to resolve current dir")?;
|
||||
let app_server = status_cli::absolute_path(&args.app_server)?;
|
||||
if !app_server.exists() {
|
||||
return Err(anyhow!(
|
||||
"codex app-server binary not found: {}",
|
||||
app_server.display()
|
||||
));
|
||||
}
|
||||
let codex_home = status_cli::absolute_path(&args.codex_home)?;
|
||||
let result_path = status_cli::absolute_path(&args.result)?;
|
||||
let transcript_path = status_cli::absolute_path(&args.transcript)?;
|
||||
let stderr_path = status_cli::absolute_path(&args.stderr)?;
|
||||
reset_smoke_paths(
|
||||
&root,
|
||||
&[
|
||||
codex_home.clone(),
|
||||
result_path.clone(),
|
||||
transcript_path.clone(),
|
||||
stderr_path.clone(),
|
||||
],
|
||||
)?;
|
||||
fs::create_dir_all(&codex_home)
|
||||
.with_context(|| format!("failed to create {}", codex_home.display()))?;
|
||||
|
||||
let mut client = status_cli::AppServerClient::start(
|
||||
&app_server,
|
||||
&codex_home,
|
||||
&transcript_path,
|
||||
&stderr_path,
|
||||
)?;
|
||||
client.send(
|
||||
"initialize",
|
||||
Some(json!({
|
||||
"clientInfo": {
|
||||
"name": "epiphany-phase6-freshness-smoke",
|
||||
"title": "Epiphany Phase 6 Freshness Smoke",
|
||||
"version": "0.1.0",
|
||||
},
|
||||
"capabilities": {"experimentalApi": true},
|
||||
})),
|
||||
true,
|
||||
)?;
|
||||
client.send("initialized", None, false)?;
|
||||
let started = client.send(
|
||||
"thread/start",
|
||||
Some(json!({"cwd": root.join("epiphany-core"), "ephemeral": true})),
|
||||
true,
|
||||
)?;
|
||||
let thread_id = started
|
||||
.pointer("/thread/id")
|
||||
.and_then(Value::as_str)
|
||||
.ok_or_else(|| anyhow!("thread/start response missing thread.id"))?
|
||||
.to_string();
|
||||
|
||||
let missing_notification_start = client.notification_len();
|
||||
let missing_response = client.send(
|
||||
"thread/epiphany/freshness",
|
||||
Some(json!({"threadId": thread_id})),
|
||||
true,
|
||||
)?;
|
||||
require(
|
||||
missing_response.get("threadId").and_then(Value::as_str) == Some(&thread_id),
|
||||
"freshness response should echo thread id",
|
||||
)?;
|
||||
assert_missing_state_freshness(&missing_response)?;
|
||||
client.require_no_notification(
|
||||
"thread/epiphany/stateUpdated",
|
||||
missing_notification_start,
|
||||
Duration::from_secs(1),
|
||||
)?;
|
||||
|
||||
let update_notification_start = client.notification_len();
|
||||
let update = client.send(
|
||||
"thread/epiphany/update",
|
||||
Some(json!({
|
||||
"threadId": thread_id,
|
||||
"expectedRevision": 0,
|
||||
"patch": freshness_patch(),
|
||||
})),
|
||||
true,
|
||||
)?;
|
||||
require(
|
||||
update.get("revision").and_then(Value::as_u64) == Some(1),
|
||||
"freshness smoke patch should advance revision to 1",
|
||||
)?;
|
||||
client.wait_for_notification(
|
||||
"thread/epiphany/stateUpdated",
|
||||
update_notification_start,
|
||||
Duration::from_secs(10),
|
||||
)?;
|
||||
|
||||
let ready_notification_start = client.notification_len();
|
||||
let ready_response = client.send(
|
||||
"thread/epiphany/freshness",
|
||||
Some(json!({"threadId": thread_id})),
|
||||
true,
|
||||
)?;
|
||||
assert_ready_freshness(&ready_response)?;
|
||||
client.require_no_notification(
|
||||
"thread/epiphany/stateUpdated",
|
||||
ready_notification_start,
|
||||
Duration::from_secs(1),
|
||||
)?;
|
||||
|
||||
let final_read = client.send(
|
||||
"thread/read",
|
||||
Some(json!({"threadId": thread_id, "includeTurns": false})),
|
||||
true,
|
||||
)?;
|
||||
require(
|
||||
final_read
|
||||
.pointer("/thread/epiphanyState/revision")
|
||||
.and_then(Value::as_u64)
|
||||
== Some(1),
|
||||
"freshness reflection should not mutate state revision",
|
||||
)?;
|
||||
|
||||
let result = json!({
|
||||
"threadId": thread_id,
|
||||
"codexHome": codex_home,
|
||||
"missingRetrievalStatus": missing_response["retrieval"]["status"],
|
||||
"missingGraphStatus": missing_response["graph"]["status"],
|
||||
"readyRevision": ready_response["stateRevision"],
|
||||
"readyRetrievalStatus": ready_response["retrieval"]["status"],
|
||||
"readyGraphStatus": ready_response["graph"]["status"],
|
||||
"readyDirtyPathCount": ready_response["graph"]["dirtyPathCount"],
|
||||
"freshnessNotificationCount": client.notification_count(
|
||||
"thread/epiphany/stateUpdated",
|
||||
ready_notification_start,
|
||||
),
|
||||
"finalReadRevision": final_read["thread"]["epiphanyState"]["revision"],
|
||||
});
|
||||
write_json(&result_path, &result)?;
|
||||
Ok(result)
|
||||
}
|
||||
|
||||
fn freshness_patch() -> Value {
|
||||
json!({
|
||||
"objective": "Expose read-only Epiphany freshness reflection without inventing watcher-driven invalidation.",
|
||||
"activeSubgoalId": "phase6-freshness-smoke",
|
||||
"subgoals": [{
|
||||
"id": "phase6-freshness-smoke",
|
||||
"title": "Live-smoke freshness reflection",
|
||||
"status": "active",
|
||||
"summary": "The freshness surface should reflect graph staleness and live retrieval state without mutation."
|
||||
}],
|
||||
"graphs": {
|
||||
"architecture": {
|
||||
"nodes": [{
|
||||
"id": "freshness-surface",
|
||||
"title": "Freshness reflection surface",
|
||||
"purpose": "Expose exact retrieval and graph freshness pressure without becoming a scheduler.",
|
||||
"code_refs": [{
|
||||
"path": "app-server/src/codex_message_processor.rs",
|
||||
"start_line": 4200,
|
||||
"end_line": 4340,
|
||||
"symbol": "thread_epiphany_freshness"
|
||||
}]
|
||||
}]
|
||||
},
|
||||
"dataflow": {"nodes": []},
|
||||
"links": []
|
||||
},
|
||||
"graphFrontier": {
|
||||
"active_node_ids": ["freshness-surface"],
|
||||
"dirty_paths": ["app-server/src/codex_message_processor.rs"],
|
||||
"open_question_ids": ["q-freshness-gap"]
|
||||
},
|
||||
"graphCheckpoint": {
|
||||
"checkpoint_id": "ck-freshness-1",
|
||||
"graph_revision": 1,
|
||||
"summary": "Freshness smoke checkpoint",
|
||||
"frontier_node_ids": ["freshness-surface"],
|
||||
"open_question_ids": ["q-freshness-gap"]
|
||||
},
|
||||
"churn": {
|
||||
"understanding_status": "ready",
|
||||
"diff_pressure": "low",
|
||||
"graph_freshness": "stale",
|
||||
"unexplained_writes": 0
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
fn assert_missing_state_freshness(response: &Value) -> Result<()> {
|
||||
require(
|
||||
response.get("source").and_then(Value::as_str) == Some("live"),
|
||||
"freshness should report live source for a loaded thread",
|
||||
)?;
|
||||
require(
|
||||
response.get("stateRevision").is_none(),
|
||||
"missing-state freshness should not invent a revision",
|
||||
)?;
|
||||
let retrieval = &response["retrieval"];
|
||||
require(
|
||||
retrieval.get("status").and_then(Value::as_str) == Some("ready"),
|
||||
"live retrieval freshness should be available",
|
||||
)?;
|
||||
require(
|
||||
retrieval.get("note").and_then(Value::as_str) == Some("Retrieval catalog is ready."),
|
||||
"fresh live retrieval should report ready note",
|
||||
)?;
|
||||
let graph = &response["graph"];
|
||||
require(
|
||||
graph.get("status").and_then(Value::as_str) == Some("missing"),
|
||||
"missing Epiphany state should block graph freshness",
|
||||
)?;
|
||||
require(
|
||||
graph.get("note").and_then(Value::as_str)
|
||||
== Some("Epiphany state is missing, so graph freshness cannot be assessed."),
|
||||
"missing graph freshness should explain itself",
|
||||
)
|
||||
}
|
||||
|
||||
fn assert_ready_freshness(response: &Value) -> Result<()> {
|
||||
require(
|
||||
response.get("source").and_then(Value::as_str) == Some("live"),
|
||||
"ready freshness should report live source",
|
||||
)?;
|
||||
require(
|
||||
response.get("stateRevision").and_then(Value::as_u64) == Some(1),
|
||||
"freshness should preserve state revision identity",
|
||||
)?;
|
||||
let retrieval = &response["retrieval"];
|
||||
require(
|
||||
retrieval.get("status").and_then(Value::as_str) == Some("ready"),
|
||||
"live retrieval freshness should stay ready for the smoke workspace",
|
||||
)?;
|
||||
require(
|
||||
retrieval.get("semanticAvailable").and_then(Value::as_bool) == Some(true),
|
||||
"live retrieval freshness should report semantic availability",
|
||||
)?;
|
||||
let graph = &response["graph"];
|
||||
require(
|
||||
graph.get("status").and_then(Value::as_str) == Some("stale"),
|
||||
"dirty graph frontier should report stale freshness",
|
||||
)?;
|
||||
require(
|
||||
graph.get("graphFreshness").and_then(Value::as_str) == Some("stale"),
|
||||
"graph freshness should expose the churn hint",
|
||||
)?;
|
||||
require(
|
||||
graph.get("checkpointId").and_then(Value::as_str) == Some("ck-freshness-1"),
|
||||
"graph freshness should expose the checkpoint id",
|
||||
)?;
|
||||
require(
|
||||
graph.get("dirtyPathCount").and_then(Value::as_u64) == Some(1),
|
||||
"graph freshness should count dirty paths",
|
||||
)?;
|
||||
require(
|
||||
string_array_eq(
|
||||
graph.get("dirtyPaths").and_then(Value::as_array),
|
||||
&["app-server/src/codex_message_processor.rs"],
|
||||
),
|
||||
"graph freshness should expose dirty paths",
|
||||
)?;
|
||||
require(
|
||||
graph.get("openQuestionCount").and_then(Value::as_u64) == Some(1),
|
||||
"graph freshness should count open questions",
|
||||
)?;
|
||||
require(
|
||||
graph.get("openGapCount").and_then(Value::as_u64) == Some(0),
|
||||
"graph freshness should count open gaps",
|
||||
)
|
||||
}
|
||||
|
||||
fn string_array_eq(values: Option<&Vec<Value>>, expected: &[&str]) -> bool {
|
||||
values.is_some_and(|values| {
|
||||
values.len() == expected.len()
|
||||
&& values
|
||||
.iter()
|
||||
.zip(expected)
|
||||
.all(|(value, expected)| value.as_str() == Some(*expected))
|
||||
})
|
||||
}
|
||||
|
||||
fn reset_smoke_paths(root: &Path, paths: &[PathBuf]) -> Result<()> {
|
||||
let smoke_root = root.join(".epiphany-smoke").canonicalize().or_else(|_| {
|
||||
let smoke_root = root.join(".epiphany-smoke");
|
||||
fs::create_dir_all(&smoke_root)?;
|
||||
smoke_root.canonicalize()
|
||||
})?;
|
||||
for path in paths {
|
||||
if !path.exists() {
|
||||
continue;
|
||||
}
|
||||
let resolved = path.canonicalize()?;
|
||||
if resolved == smoke_root || !resolved.starts_with(&smoke_root) {
|
||||
return Err(anyhow!(
|
||||
"refusing to delete non-smoke path: {}",
|
||||
path.display()
|
||||
));
|
||||
}
|
||||
if path.is_dir() {
|
||||
fs::remove_dir_all(path)?;
|
||||
} else {
|
||||
fs::remove_file(path)?;
|
||||
}
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn write_json(path: &Path, value: &Value) -> Result<()> {
|
||||
if let Some(parent) = path.parent() {
|
||||
fs::create_dir_all(parent)?;
|
||||
}
|
||||
fs::write(path, format!("{}\n", serde_json::to_string_pretty(value)?))?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn require(condition: bool, message: &str) -> Result<()> {
|
||||
if condition {
|
||||
Ok(())
|
||||
} else {
|
||||
Err(anyhow!("{message}"))
|
||||
}
|
||||
}
|
||||
|
||||
fn take_path(args: &mut impl Iterator<Item = String>, name: &str) -> Result<PathBuf> {
|
||||
args.next()
|
||||
.map(PathBuf::from)
|
||||
.ok_or_else(|| anyhow!("{name} requires a value"))
|
||||
}
|
||||
|
|
@ -1502,7 +1502,7 @@ Current tests cover the landed flows at useful seams:
|
|||
- reusable app-server smoke for the richer Phase 5 chain in `tools/epiphany_phase5_smoke.py`, including update/promote response metadata, `thread/epiphany/stateUpdated` notifications with source, revision, changed-field checks, retrieval-summary presence in successful write response/notification states, the verifier-only promotion edge where `Evidence` changes because verifier evidence is appended, invalid direct observation/evidence and replacement update rejection without mutation or notification, and explicit zero-notification assertions for rejected promotions.
|
||||
- live app-server smoke for the first Phase 6 scene boundary in `tools/epiphany_phase6_scene_smoke.py`.
|
||||
- live app-server smoke for the first Phase 6 jobs boundary in `tools/epiphany_phase6_jobs_smoke.py`.
|
||||
- live app-server smoke for the first Phase 6 freshness boundary in `tools/epiphany_phase6_freshness_smoke.py`.
|
||||
- live app-server smoke for the first Phase 6 freshness boundary through native `epiphany-phase6-freshness-smoke`.
|
||||
- live app-server smoke for watcher-backed invalidation inputs inside the Phase 6 freshness boundary in `tools/epiphany_phase6_invalidation_smoke.py`.
|
||||
- live app-server smoke for the first Phase 6 context boundary in `tools/epiphany_phase6_context_smoke.py`.
|
||||
- live app-server smoke for the first Phase 6 graph traversal boundary in `tools/epiphany_phase6_graph_query_smoke.py`.
|
||||
|
|
@ -1526,7 +1526,7 @@ $env:CARGO_TARGET_DIR='C:\Users\Meta\.cargo-target-codex'; cargo test -p codex-a
|
|||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase5_smoke.py'
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_scene_smoke.py'
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_jobs_smoke.py'
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_freshness_smoke.py'
|
||||
cargo run --manifest-path .\epiphany-core\Cargo.toml --bin epiphany-phase6-freshness-smoke
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_invalidation_smoke.py'
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_context_smoke.py'
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_graph_query_smoke.py'
|
||||
|
|
|
|||
|
|
@ -63,7 +63,7 @@ The landed machine now has:
|
|||
- read-only Phase 6 job/progress reflection through `thread/epiphany/jobs`, with durable `jobBindings` plus live runtime `agent_jobs` overlay
|
||||
- live Phase 6 jobs app-server smoke coverage in `tools/epiphany_phase6_jobs_smoke.py`
|
||||
- read-only Phase 6 retrieval/graph freshness reflection plus watcher-backed invalidation inputs through `thread/epiphany/freshness`
|
||||
- live Phase 6 freshness app-server smoke coverage in `tools/epiphany_phase6_freshness_smoke.py`
|
||||
- live Phase 6 freshness app-server smoke coverage through native `epiphany-phase6-freshness-smoke`
|
||||
- live Phase 6 watcher-backed invalidation smoke coverage in `tools/epiphany_phase6_invalidation_smoke.py`
|
||||
- read-only Phase 6 targeted state-shard reflection through `thread/epiphany/context`
|
||||
- live Phase 6 context app-server smoke coverage in `tools/epiphany_phase6_context_smoke.py`
|
||||
|
|
@ -441,7 +441,7 @@ Before modifying jobs reflection behavior, run:
|
|||
Before modifying freshness reflection behavior, run:
|
||||
|
||||
```powershell
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_freshness_smoke.py'
|
||||
cargo run --manifest-path .\epiphany-core\Cargo.toml --bin epiphany-phase6-freshness-smoke
|
||||
```
|
||||
|
||||
Before modifying watcher-backed invalidation behavior inside freshness reflection, run:
|
||||
|
|
|
|||
|
|
@ -145,7 +145,7 @@ The current spine, blessed but not yet finished:
|
|||
- first GUI graph dashboard through the adjacent `@epiphanygraph/epiphany-graph-viewer` package over typed Epiphany graph state
|
||||
- live scene app-server smoke through `tools/epiphany_phase6_scene_smoke.py`
|
||||
- live jobs app-server smoke through `tools/epiphany_phase6_jobs_smoke.py`
|
||||
- live freshness app-server smoke through `tools/epiphany_phase6_freshness_smoke.py`
|
||||
- live freshness app-server smoke through native `epiphany-phase6-freshness-smoke`
|
||||
- live watcher-backed invalidation smoke through `tools/epiphany_phase6_invalidation_smoke.py`
|
||||
- live context app-server smoke through `tools/epiphany_phase6_context_smoke.py`
|
||||
- live graph traversal app-server smoke through `tools/epiphany_phase6_graph_query_smoke.py`
|
||||
|
|
@ -216,7 +216,7 @@ For launch/interrupt authority changes over the thin job seam, run:
|
|||
For freshness reflection behavior changes, run:
|
||||
|
||||
```powershell
|
||||
& 'C:\Users\Meta\.cache\codex-runtimes\codex-primary-runtime\dependencies\python\python.exe' '.\tools\epiphany_phase6_freshness_smoke.py'
|
||||
cargo run --manifest-path .\epiphany-core\Cargo.toml --bin epiphany-phase6-freshness-smoke
|
||||
```
|
||||
|
||||
For watcher-backed invalidation behavior inside freshness reflection, run:
|
||||
|
|
|
|||
Binary file not shown.
|
|
@ -201,7 +201,7 @@ landed_baseline:
|
|||
- Use `tools/epiphany_phase6_scene_smoke.py` before modifying scene projection behavior.
|
||||
- Use `tools/epiphany_phase6_jobs_smoke.py` before modifying job/progress reflection behavior.
|
||||
- Use `tools/epiphany_phase6_job_control_smoke.py` before modifying launch/interrupt authority over the thin job seam.
|
||||
- Use `tools/epiphany_phase6_freshness_smoke.py` before modifying freshness reflection behavior.
|
||||
- Use `cargo run --manifest-path .\epiphany-core\Cargo.toml --bin epiphany-phase6-freshness-smoke` before modifying freshness reflection behavior.
|
||||
- Use `tools/epiphany_phase6_invalidation_smoke.py` before modifying watcher-backed invalidation behavior inside freshness reflection.
|
||||
- Use `tools/epiphany_phase6_context_smoke.py` before modifying targeted context-shard behavior.
|
||||
- Use `tools/epiphany_phase6_graph_query_smoke.py` before modifying graph traversal behavior.
|
||||
|
|
|
|||
|
|
@ -1,240 +0,0 @@
|
|||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import json
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from epiphany_phase5_smoke import AppServerClient
|
||||
from epiphany_phase5_smoke import DEFAULT_APP_SERVER
|
||||
from epiphany_phase5_smoke import ROOT
|
||||
from epiphany_phase5_smoke import require
|
||||
from epiphany_phase5_smoke import reset_smoke_paths
|
||||
|
||||
|
||||
DEFAULT_CODEX_HOME = ROOT / ".epiphany-smoke" / "phase6-freshness-codex-home"
|
||||
DEFAULT_RESULT = ROOT / ".epiphany-smoke" / "phase6-freshness-smoke-result.json"
|
||||
DEFAULT_TRANSCRIPT = ROOT / ".epiphany-smoke" / "phase6-freshness-smoke-transcript.jsonl"
|
||||
DEFAULT_STDERR = ROOT / ".epiphany-smoke" / "phase6-freshness-smoke-server.stderr.log"
|
||||
|
||||
FRESHNESS_CODE_REF = {
|
||||
"path": "app-server/src/codex_message_processor.rs",
|
||||
"start_line": 4200,
|
||||
"end_line": 4340,
|
||||
"symbol": "thread_epiphany_freshness",
|
||||
}
|
||||
|
||||
|
||||
def freshness_patch() -> dict[str, Any]:
|
||||
return {
|
||||
"objective": "Expose read-only Epiphany freshness reflection without inventing watcher-driven invalidation.",
|
||||
"activeSubgoalId": "phase6-freshness-smoke",
|
||||
"subgoals": [
|
||||
{
|
||||
"id": "phase6-freshness-smoke",
|
||||
"title": "Live-smoke freshness reflection",
|
||||
"status": "active",
|
||||
"summary": "The freshness surface should reflect graph staleness and live retrieval state without mutation.",
|
||||
}
|
||||
],
|
||||
"graphs": {
|
||||
"architecture": {
|
||||
"nodes": [
|
||||
{
|
||||
"id": "freshness-surface",
|
||||
"title": "Freshness reflection surface",
|
||||
"purpose": "Expose exact retrieval and graph freshness pressure without becoming a scheduler.",
|
||||
"code_refs": [FRESHNESS_CODE_REF],
|
||||
}
|
||||
]
|
||||
},
|
||||
"dataflow": {"nodes": []},
|
||||
"links": [],
|
||||
},
|
||||
"graphFrontier": {
|
||||
"active_node_ids": ["freshness-surface"],
|
||||
"dirty_paths": ["app-server/src/codex_message_processor.rs"],
|
||||
"open_question_ids": ["q-freshness-gap"],
|
||||
},
|
||||
"graphCheckpoint": {
|
||||
"checkpoint_id": "ck-freshness-1",
|
||||
"graph_revision": 1,
|
||||
"summary": "Freshness smoke checkpoint",
|
||||
"frontier_node_ids": ["freshness-surface"],
|
||||
"open_question_ids": ["q-freshness-gap"],
|
||||
},
|
||||
"churn": {
|
||||
"understanding_status": "ready",
|
||||
"diff_pressure": "low",
|
||||
"graph_freshness": "stale",
|
||||
"unexplained_writes": 0,
|
||||
},
|
||||
}
|
||||
|
||||
|
||||
def assert_missing_state_freshness(response: dict[str, Any]) -> None:
|
||||
require(
|
||||
response["source"] == "live",
|
||||
"freshness should report live source for a loaded thread",
|
||||
)
|
||||
require(
|
||||
"stateRevision" not in response,
|
||||
"missing-state freshness should not invent a revision",
|
||||
)
|
||||
retrieval = response["retrieval"]
|
||||
require(retrieval["status"] == "ready", "live retrieval freshness should be available")
|
||||
require(
|
||||
retrieval["note"] == "Retrieval catalog is ready.",
|
||||
"fresh live retrieval should report ready note",
|
||||
)
|
||||
graph = response["graph"]
|
||||
require(graph["status"] == "missing", "missing Epiphany state should block graph freshness")
|
||||
require(
|
||||
graph["note"] == "Epiphany state is missing, so graph freshness cannot be assessed.",
|
||||
"missing graph freshness should explain itself",
|
||||
)
|
||||
|
||||
|
||||
def assert_ready_freshness(response: dict[str, Any]) -> None:
|
||||
require(response["source"] == "live", "ready freshness should report live source")
|
||||
require(response["stateRevision"] == 1, "freshness should preserve state revision identity")
|
||||
|
||||
retrieval = response["retrieval"]
|
||||
require(
|
||||
retrieval["status"] == "ready",
|
||||
"live retrieval freshness should stay ready for the smoke workspace",
|
||||
)
|
||||
require(
|
||||
retrieval["semanticAvailable"] is True,
|
||||
"live retrieval freshness should report semantic availability",
|
||||
)
|
||||
|
||||
graph = response["graph"]
|
||||
require(graph["status"] == "stale", "dirty graph frontier should report stale freshness")
|
||||
require(
|
||||
graph["graphFreshness"] == "stale",
|
||||
"graph freshness should expose the churn hint",
|
||||
)
|
||||
require(
|
||||
graph["checkpointId"] == "ck-freshness-1",
|
||||
"graph freshness should expose the checkpoint id",
|
||||
)
|
||||
require(graph["dirtyPathCount"] == 1, "graph freshness should count dirty paths")
|
||||
require(graph["dirtyPaths"] == ["app-server/src/codex_message_processor.rs"], "graph freshness should expose dirty paths")
|
||||
require(graph["openQuestionCount"] == 1, "graph freshness should count open questions")
|
||||
require(graph["openGapCount"] == 0, "graph freshness should count open gaps")
|
||||
|
||||
|
||||
def run_smoke(args: argparse.Namespace) -> dict[str, Any]:
|
||||
app_server = args.app_server.resolve()
|
||||
if not app_server.exists():
|
||||
raise FileNotFoundError(f"codex app-server binary not found: {app_server}")
|
||||
|
||||
codex_home = args.codex_home.resolve()
|
||||
result_path = args.result.resolve()
|
||||
transcript_path = args.transcript.resolve()
|
||||
stderr_path = args.stderr.resolve()
|
||||
reset_smoke_paths(codex_home, result_path, transcript_path, stderr_path)
|
||||
|
||||
with AppServerClient(app_server, codex_home, transcript_path, stderr_path) as client:
|
||||
client.send(
|
||||
"initialize",
|
||||
{
|
||||
"clientInfo": {
|
||||
"name": "epiphany-phase6-freshness-smoke",
|
||||
"title": "Epiphany Phase 6 Freshness Smoke",
|
||||
"version": "0.1.0",
|
||||
},
|
||||
"capabilities": {"experimentalApi": True},
|
||||
},
|
||||
)
|
||||
client.send("initialized", expect_response=False)
|
||||
started = client.send(
|
||||
"thread/start",
|
||||
{"cwd": str(ROOT / "epiphany-core"), "ephemeral": True},
|
||||
)
|
||||
assert started is not None
|
||||
thread_id = started["thread"]["id"]
|
||||
|
||||
missing_notification_start = len(client.notifications)
|
||||
missing_response = client.send("thread/epiphany/freshness", {"threadId": thread_id})
|
||||
assert missing_response is not None
|
||||
require(
|
||||
missing_response["threadId"] == thread_id,
|
||||
"freshness response should echo thread id",
|
||||
)
|
||||
assert_missing_state_freshness(missing_response)
|
||||
client.require_no_notification(
|
||||
"thread/epiphany/stateUpdated",
|
||||
start_index=missing_notification_start,
|
||||
)
|
||||
|
||||
update_notification_start = len(client.notifications)
|
||||
update = client.send(
|
||||
"thread/epiphany/update",
|
||||
{"threadId": thread_id, "expectedRevision": 0, "patch": freshness_patch()},
|
||||
)
|
||||
assert update is not None
|
||||
require(update["revision"] == 1, "freshness smoke patch should advance revision to 1")
|
||||
client.wait_for_notification(
|
||||
"thread/epiphany/stateUpdated",
|
||||
start_index=update_notification_start,
|
||||
)
|
||||
|
||||
ready_notification_start = len(client.notifications)
|
||||
ready_response = client.send("thread/epiphany/freshness", {"threadId": thread_id})
|
||||
assert ready_response is not None
|
||||
assert_ready_freshness(ready_response)
|
||||
client.require_no_notification(
|
||||
"thread/epiphany/stateUpdated",
|
||||
start_index=ready_notification_start,
|
||||
)
|
||||
|
||||
final_read = client.send("thread/read", {"threadId": thread_id, "includeTurns": False})
|
||||
assert final_read is not None
|
||||
require(
|
||||
final_read["thread"]["epiphanyState"]["revision"] == 1,
|
||||
"freshness reflection should not mutate state revision",
|
||||
)
|
||||
|
||||
result = {
|
||||
"threadId": thread_id,
|
||||
"codexHome": str(codex_home),
|
||||
"missingRetrievalStatus": missing_response["retrieval"]["status"],
|
||||
"missingGraphStatus": missing_response["graph"]["status"],
|
||||
"readyRevision": ready_response["stateRevision"],
|
||||
"readyRetrievalStatus": ready_response["retrieval"]["status"],
|
||||
"readyGraphStatus": ready_response["graph"]["status"],
|
||||
"readyDirtyPathCount": ready_response["graph"]["dirtyPathCount"],
|
||||
"freshnessNotificationCount": client.count_notifications(
|
||||
"thread/epiphany/stateUpdated",
|
||||
start_index=ready_notification_start,
|
||||
),
|
||||
"finalReadRevision": final_read["thread"]["epiphanyState"]["revision"],
|
||||
}
|
||||
|
||||
result_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
result_path.write_text(
|
||||
json.dumps(result, indent=2, ensure_ascii=False) + "\n",
|
||||
encoding="utf-8",
|
||||
)
|
||||
return result
|
||||
|
||||
|
||||
def main() -> int:
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Live-smoke the Phase 6 Epiphany freshness reflection surface."
|
||||
)
|
||||
parser.add_argument("--app-server", type=Path, default=DEFAULT_APP_SERVER)
|
||||
parser.add_argument("--codex-home", type=Path, default=DEFAULT_CODEX_HOME)
|
||||
parser.add_argument("--result", type=Path, default=DEFAULT_RESULT)
|
||||
parser.add_argument("--transcript", type=Path, default=DEFAULT_TRANSCRIPT)
|
||||
parser.add_argument("--stderr", type=Path, default=DEFAULT_STDERR)
|
||||
args = parser.parse_args()
|
||||
result = run_smoke(args)
|
||||
print(json.dumps(result, indent=2, ensure_ascii=False))
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
Reference in a new issue