merge(opencode): suite de tests fiable en exécution parallèle — #253 (QA verte)

.ideai/tickets/253/carnet.md

@@ -1,6 +1,14 @@
 ---
 issueRef: "#253"
-version: 1
+version: 2
-updatedBy: {"kind":"agent","agent_id":"a6c6ea12-bfc6-4bdc-8031-324102dfa34d"}
+updatedBy: {"kind":"agent","agent_id":"1ff94b51-3e17-4a39-8543-7715d1ff0f80"}
-updatedAt: 1787551681631
+updatedAt: 1787551753848
 ---
+# Git — décision de branche (2026-08-24)
+
+- Branche de travail : `fix/253-opencode-tests-parallel-flakiness` (créée et checkout, base = tip develop `f303cc1`, inclut le merge #252 `4553e5d` — lignage exact où le flakiness a été observé).
+- DevBackend : corrections sur cette branche (timeouts, fixtures `sleep`, isolation des tests à timing dans `crates/infrastructure/src/session/opencode.rs`). Commits : `test(opencode): …` (ou `fix(opencode): …` si du code produit est touché) `— #253`. Sans dégrader la couverture : les scénarios restent, seule la fiabilité change.
+- QA : preuve = sérialisé toujours vert (`--test-threads=1`) **ET** exécution **parallèle** verte sur ≥5 runs consécutifs de `cargo test -p infrastructure --lib session::opencode` (le mode par défaut, celui qui flake). Consigner runs et sortie réelle dans ce carnet.
+- Interdits : aucun push distant ; pas de merge/rebase vers `develop` sans l'agent Git. Fusion locale décidée par Git après QA verte.
+- `scripts/restart-idea-over-ssh.sh` reste non suivi — ne pas le committer.
+- Contexte : symptômes initiaux documentés dans le carnet #252 (section « Signalement QA » du 2026-08-24).

.ideai/tickets/253/issue.md

@@ -9,9 +9,9 @@ links: []
 agentRefs: []
 attachments: []
 createdBy: {"kind":"agent","agent_id":"a6c6ea12-bfc6-4bdc-8031-324102dfa34d"}
-updatedBy: {"kind":"agent","agent_id":"a6c6ea12-bfc6-4bdc-8031-324102dfa34d"}
+updatedBy: {"kind":"agent","agent_id":"1ff94b51-3e17-4a39-8543-7715d1ff0f80"}
 createdAt: 1787551681631
-updatedAt: 1787551681631
+updatedAt: 1787551753848
-version: 1
+version: 2
 ---
 Des tests anciens de `session::opencode` échouent de façon intermittente en exécution parallèle, alors qu'ils passent en série (`--test-threads=1`). Symptômes remontés pendant le ticket #252: tests différents à chaque run, concernés notamment `seeded_session_ignores_and_does_not_capture_other_session_events`, `seeded_or_captured_session_id_is_passed_as_opencode_session_arg`, etc. Objectif: supprimer la flakiness sans dégrader la couverture, obtenir une exécution verte en parallèle sur la suite ciblée, et documenter la cause racine ainsi que la stratégie retenue.

.ideai/tickets/index.json

@@ -3419,7 +3419,7 @@
         "kind": "agent",
         "agent_id": "a6c6ea12-bfc6-4bdc-8031-324102dfa34d"
       },
-      "updatedAt": 1787551681631
+      "updatedAt": 1787551753848
     }
   ]
 }

crates/infrastructure/src/session/opencode.rs

@@ -783,6 +783,9 @@ mod tests {
 
     use super::*;
 
+    #[cfg(unix)]
+    const LIVE_EVENT_TIMEOUT: Duration = Duration::from_secs(5);
+
     fn without_progress(events: Vec<ReplyEvent>) -> Vec<ReplyEvent> {
         events
             .into_iter()
@@ -1034,18 +1037,56 @@ mod tests {
 
     #[cfg(unix)]
     fn write_executable_script(path: &Path, body: impl AsRef<str>) {
-        fs::write(path, body.as_ref()).unwrap();
+        let tmp = path.with_extension(format!("tmp-{}", uuid::Uuid::new_v4()));
+        {
+            use std::io::Write as _;
+
+            let mut file = fs::File::create(&tmp).unwrap();
+            file.write_all(body.as_ref().as_bytes()).unwrap();
+            file.sync_all().unwrap();
+        }
+        fs::rename(&tmp, path).unwrap();
         let mut permissions = fs::metadata(path).unwrap().permissions();
         permissions.set_mode(0o755);
         fs::set_permissions(path, permissions).unwrap();
     }
 
+    #[cfg(unix)]
+    fn shell_wait_for_file(path: &Path) -> String {
+        format!("while [ ! -f '{}' ]; do sleep 0.05; done", path.display())
+    }
+
+    #[cfg(unix)]
+    fn shell_script_command(path: &Path) -> (&'static str, Vec<String>) {
+        ("sh", vec![path.to_string_lossy().into_owned()])
+    }
+
+    #[cfg(unix)]
+    fn spawn_send_with_tap(
+        session: Arc<OpenCodeSession>,
+        prompt: &'static str,
+        tx: std::sync::mpsc::Sender<ReplyEvent>,
+    ) -> std::thread::JoinHandle<Result<Vec<ReplyEvent>, AgentSessionError>> {
+        std::thread::spawn(move || {
+            tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()
+                .unwrap()
+                .block_on(async move {
+                    session
+                        .send_with_tap(prompt, tx)
+                        .await
+                        .map(|stream| stream.collect::<Vec<_>>())
+                })
+        })
+    }
+
     #[cfg(unix)]
     #[tokio::test]
     async fn send_prefers_parseable_stdout_over_nonzero_exit_status() {
         let tmp = TempDir::new("exit-one");
         let script = tmp.path().join("opencode-fixture.sh");
-        fs::write(
+        write_executable_script(
             &script,
             r#"#!/bin/sh
 printf '%s\n' '{"type":"step_start"}'
@@ -1054,15 +1095,12 @@ printf '%s\n' '{"type":"step_finish"}'
 printf '%s\n' 'stderr générique' >&2
 exit 1
 "#,
-        )
+        );
-        .unwrap();
+        let (command, args) = shell_script_command(&script);
-        let mut permissions = fs::metadata(&script).unwrap().permissions();
-        permissions.set_mode(0o755);
-        fs::set_permissions(&script, permissions).unwrap();
         let session = OpenCodeSession::new(
             SessionId::new_random(),
-            script.to_string_lossy(),
+            command,
-            Vec::new(),
+            args,
             tmp.path().to_string_lossy(),
             None,
             Vec::new(),
@@ -1089,25 +1127,28 @@ exit 1
     }
 
     #[cfg(unix)]
-    #[tokio::test]
+    #[test]
-    async fn send_with_tap_emits_progress_live_before_final_and_tags_idea_tool_as_mcp() {
+    fn send_with_tap_emits_progress_live_before_final_and_tags_idea_tool_as_mcp() {
         let tmp = TempDir::new("tap-live");
         let script = tmp.path().join("opencode-fixture.sh");
+        let release = tmp.path().join("release");
         write_executable_script(
             &script,
             r#"#!/bin/sh
 printf '%s\n' '{"type":"step_start","sessionID":"fresh-engine","part":{"messageID":"m1","type":"step-start"}}'
-sleep 1
+{{WAIT_FOR_RELEASE}}
 printf '%s\n' '{"type":"tool_use","sessionID":"fresh-engine","part":{"messageID":"m1","type":"tool","tool":"idea_ask_agent"}}'
 printf '%s\n' '{"type":"text","sessionID":"fresh-engine","part":{"messageID":"m1","type":"text","text":"réponse live"}}'
 printf '%s\n' '{"type":"step_finish","sessionID":"fresh-engine","part":{"messageID":"m1","type":"step-finish"}}'
-"#,
+"#
+            .replace("{{WAIT_FOR_RELEASE}}", &shell_wait_for_file(&release)),
         );
+        let (command, args) = shell_script_command(&script);
         let session = Arc::new(
             OpenCodeSession::new(
                 SessionId::new_random(),
-                script.to_string_lossy(),
+                command,
-                Vec::new(),
+                args,
                 tmp.path().to_string_lossy(),
                 None,
                 Vec::new(),
@@ -1117,24 +1158,12 @@ printf '%s\n' '{"type":"step_finish","sessionID":"fresh-engine","part":{"message
             .unwrap(),
         );
         let (tx, rx) = std::sync::mpsc::channel();
-        let worker = {
+        let worker = spawn_send_with_tap(Arc::clone(&session), "prompt", tx);
-            let session = Arc::clone(&session);
-            tokio::spawn(async move {
-                session
-                    .send_with_tap("prompt", tx)
-                    .await
-                    .map(|stream| stream.collect::<Vec<_>>())
-            })
-        };
 
-        let (first_live, rx) = tokio::task::spawn_blocking(move || {
+        let first_live = rx
-            let first_live = rx
+            .recv_timeout(LIVE_EVENT_TIMEOUT)
-                .recv_timeout(Duration::from_millis(500))
+            .expect("un événement de progress doit atteindre le tap avant le Final");
-                .expect("un événement de progress doit atteindre le tap avant le Final");
+        fs::write(&release, "").unwrap();
-            (first_live, rx)
-        })
-        .await
-        .unwrap();
         assert!(
             matches!(
                 first_live,
@@ -1144,7 +1173,7 @@ printf '%s\n' '{"type":"step_finish","sessionID":"fresh-engine","part":{"message
             "le premier événement live doit être le step_start provider: {first_live:?}"
         );
 
-        let returned = worker.await.unwrap().unwrap();
+        let returned = worker.join().unwrap().unwrap();
         let live: Vec<_> = std::iter::once(first_live).chain(rx.into_iter()).collect();
 
         assert!(
@@ -1187,10 +1216,11 @@ printf '%s\n' '{"type":"text","sessionID":"seeded-engine","part":{"messageID":"n
 printf '%s\n' '{"type":"step_finish","sessionID":"seeded-engine","part":{"messageID":"new","type":"step-finish"}}'
 "#,
         );
+        let (command, args) = shell_script_command(&script);
         let session = OpenCodeSession::new(
             SessionId::new_random(),
-            script.to_string_lossy(),
+            command,
-            Vec::new(),
+            args,
             tmp.path().to_string_lossy(),
             Some("seeded-engine".to_owned()),
             Vec::new(),
@@ -1230,25 +1260,28 @@ printf '%s\n' '{"type":"step_finish","sessionID":"seeded-engine","part":{"messag
     }
 
     #[cfg(unix)]
-    #[tokio::test]
+    #[test]
-    async fn send_with_tap_streams_progress_for_seeded_session_with_single_current_message_id() {
+    fn send_with_tap_streams_progress_for_seeded_session_with_single_current_message_id() {
         let tmp = TempDir::new("tap-seeded-single-message");
         let script = tmp.path().join("opencode-fixture.sh");
+        let release = tmp.path().join("release");
         write_executable_script(
             &script,
             r#"#!/bin/sh
 printf '%s\n' '{"type":"step_start","sessionID":"seeded-engine","part":{"messageID":"current","type":"step-start"}}'
-sleep 1
+{{WAIT_FOR_RELEASE}}
 printf '%s\n' '{"type":"tool_use","sessionID":"seeded-engine","part":{"messageID":"current","type":"tool","tool":"idea_ask_agent"}}'
 printf '%s\n' '{"type":"text","sessionID":"seeded-engine","part":{"messageID":"current","type":"text","text":"réponse courante"}}'
 printf '%s\n' '{"type":"step_finish","sessionID":"seeded-engine","part":{"messageID":"current","type":"step-finish"}}'
-"#,
+"#
+            .replace("{{WAIT_FOR_RELEASE}}", &shell_wait_for_file(&release)),
         );
+        let (command, args) = shell_script_command(&script);
         let session = Arc::new(
             OpenCodeSession::new(
                 SessionId::new_random(),
-                script.to_string_lossy(),
+                command,
-                Vec::new(),
+                args,
                 tmp.path().to_string_lossy(),
                 Some("seeded-engine".to_owned()),
                 Vec::new(),
@@ -1258,24 +1291,12 @@ printf '%s\n' '{"type":"step_finish","sessionID":"seeded-engine","part":{"messag
             .unwrap(),
         );
         let (tx, rx) = std::sync::mpsc::channel();
-        let worker = {
+        let worker = spawn_send_with_tap(Arc::clone(&session), "prompt", tx);
-            let session = Arc::clone(&session);
-            tokio::spawn(async move {
-                session
-                    .send_with_tap("prompt", tx)
-                    .await
-                    .map(|stream| stream.collect::<Vec<_>>())
-            })
-        };
 
-        let (first_live, rx) = tokio::task::spawn_blocking(move || {
+        let first_live = rx
-            let first_live = rx
+            .recv_timeout(LIVE_EVENT_TIMEOUT)
-                .recv_timeout(Duration::from_millis(500))
+            .expect("la session existante doit streamer le progress du tour courant");
-                .expect("la session existante doit streamer le progress du tour courant");
+        fs::write(&release, "").unwrap();
-            (first_live, rx)
-        })
-        .await
-        .unwrap();
         assert!(
             matches!(
                 first_live,
@@ -1285,7 +1306,7 @@ printf '%s\n' '{"type":"step_finish","sessionID":"seeded-engine","part":{"messag
             "le premier événement live doit être le step_start du message courant: {first_live:?}"
         );
 
-        let returned = worker.await.unwrap().unwrap();
+        let returned = worker.join().unwrap().unwrap();
         let live: Vec<_> = std::iter::once(first_live).chain(rx.into_iter()).collect();
 
         assert!(
@@ -1332,22 +1353,19 @@ printf '%s\n' '{"type":"step_finish","sessionID":"seeded-engine","part":{"messag
     async fn send_captures_engine_session_id_from_jsonl() {
         let tmp = TempDir::new("capture-session");
         let script = tmp.path().join("opencode-fixture.sh");
-        fs::write(
+        write_executable_script(
             &script,
             r#"#!/bin/sh
 printf '%s\n' '{"type":"step_start","sessionID":"opencode-engine-1"}'
 printf '%s\n' '{"type":"text","sessionID":"opencode-engine-1","part":{"text":"ok"}}'
 printf '%s\n' '{"type":"step_finish","sessionID":"opencode-engine-1"}'
 "#,
-        )
+        );
-        .unwrap();
+        let (command, args) = shell_script_command(&script);
-        let mut permissions = fs::metadata(&script).unwrap().permissions();
-        permissions.set_mode(0o755);
-        fs::set_permissions(&script, permissions).unwrap();
         let session = OpenCodeSession::new(
             SessionId::new_random(),
-            script.to_string_lossy(),
+            command,
-            Vec::new(),
+            args,
             tmp.path().to_string_lossy(),
             None,
             Vec::new(),
@@ -1372,7 +1390,7 @@ printf '%s\n' '{"type":"step_finish","sessionID":"opencode-engine-1"}'
         let tmp = TempDir::new("resume-arg");
         let script = tmp.path().join("opencode-fixture.sh");
         let args_file = tmp.path().join("args.txt");
-        fs::write(
+        write_executable_script(
             &script,
             format!(
                 r#"#!/bin/sh
@@ -1383,15 +1401,12 @@ printf '%s\n' '{{"type":"step_finish","sessionID":"seeded-engine"}}'
 "#,
                 args_file.display()
             ),
-        )
+        );
-        .unwrap();
+        let (command, args) = shell_script_command(&script);
-        let mut permissions = fs::metadata(&script).unwrap().permissions();
-        permissions.set_mode(0o755);
-        fs::set_permissions(&script, permissions).unwrap();
         let session = OpenCodeSession::new(
             SessionId::new_random(),
-            script.to_string_lossy(),
+            command,
-            Vec::new(),
+            args,
             tmp.path().to_string_lossy(),
             Some("seeded-engine".to_owned()),
             Vec::new(),
@@ -1417,7 +1432,7 @@ printf '%s\n' '{{"type":"step_finish","sessionID":"seeded-engine"}}'
     async fn seeded_session_ignores_and_does_not_capture_other_session_events() {
         let tmp = TempDir::new("ignore-other-session");
         let script = tmp.path().join("opencode-fixture.sh");
-        fs::write(
+        write_executable_script(
             &script,
             r#"#!/bin/sh
 printf '%s\n' '{"type":"step_start","sessionID":"seeded-engine","part":{"messageID":"m1"}}'
@@ -1425,15 +1440,12 @@ printf '%s\n' '{"type":"text","sessionID":"seeded-engine","part":{"messageID":"m
 printf '%s\n' '{"type":"step_finish","sessionID":"seeded-engine","part":{"messageID":"m1"}}'
 printf '%s\n' '{"type":"text","sessionID":"other-engine","part":{"messageID":"m2","text":"pollution"}}'
 "#,
-        )
+        );
-        .unwrap();
+        let (command, args) = shell_script_command(&script);
-        let mut permissions = fs::metadata(&script).unwrap().permissions();
-        permissions.set_mode(0o755);
-        fs::set_permissions(&script, permissions).unwrap();
         let session = OpenCodeSession::new(
             SessionId::new_random(),
-            script.to_string_lossy(),
+            command,
-            Vec::new(),
+            args,
             tmp.path().to_string_lossy(),
             Some("seeded-engine".to_owned()),
             Vec::new(),