Reuse canonical chat context

src/runtime.rs

@@ -19,7 +19,6 @@ pub(crate) struct GenerationService {
 pub(crate) struct ActiveGeneration {
     pub(crate) events: Receiver<GenerationEvent>,
     pub(crate) cancel: Arc<AtomicBool>,
-    pub(crate) started_at: Instant,
 }
 
 impl Drop for ActiveGeneration {
@@ -29,7 +28,11 @@ impl Drop for ActiveGeneration {
 }
 
 pub(crate) enum CheckpointTarget {
-    Local(PathBuf),
+    Local {
+        checkpoint: PathBuf,
+        /// Shared cache for the deterministic rendered system prompt.
+        bootstrap: Option<PathBuf>,
+    },
     Transient(PathBuf),
     /// Same transient KV handling as [`CheckpointTarget::Transient`], but asked
     /// for by the app itself (session titling) rather than by an HTTP client.
@@ -39,7 +42,7 @@ pub(crate) enum CheckpointTarget {
 impl CheckpointTarget {
     fn source(&self) -> WorkSource {
         match self {
-            Self::Local(_) | Self::OneShot(_) => WorkSource::LocalChat,
+            Self::Local { .. } | Self::OneShot(_) => WorkSource::LocalChat,
             Self::Transient(_) => WorkSource::Http,
         }
     }
@@ -108,7 +111,6 @@ impl GenerationService {
         checkpoint: CheckpointTarget,
         idle_timeout: Duration,
     ) -> Result<ActiveGeneration, String> {
-        let started_at = Instant::now();
         let source = checkpoint.source();
         let cancel = Arc::new(AtomicBool::new(false));
         let (events, receiver) = mpsc::channel();
@@ -133,7 +135,6 @@ impl GenerationService {
         Ok(ActiveGeneration {
             events: receiver,
             cancel,
-            started_at,
         })
     }
 
@@ -148,7 +149,6 @@ impl GenerationService {
         checkpoint: PathBuf,
         idle_timeout: Duration,
     ) -> Result<ActiveGeneration, String> {
-        let started_at = Instant::now();
         let cancel = Arc::new(AtomicBool::new(false));
         let (events, receiver) = mpsc::channel();
         self.commands
@@ -156,7 +156,10 @@ impl GenerationService {
                 engine,
                 turn,
                 messages,
-                checkpoint: CheckpointTarget::Local(checkpoint),
+                checkpoint: CheckpointTarget::Local {
+                    checkpoint,
+                    bootstrap: None,
+                },
                 operation: Operation::Compact {
                     reason: reason.to_owned(),
                     rebuild_system_prompt,
@@ -169,7 +172,6 @@ impl GenerationService {
         Ok(ActiveGeneration {
             events: receiver,
             cancel,
-            started_at,
         })
     }
 
@@ -180,7 +182,6 @@ impl GenerationService {
         messages: Vec<ChatTurn>,
         idle_timeout: Duration,
     ) -> Result<ActiveGeneration, String> {
-        let started_at = Instant::now();
         let cancel = Arc::new(AtomicBool::new(false));
         let (events, receiver) = mpsc::channel();
         self.metrics.request_queued(WorkSource::LocalChat);
@@ -199,7 +200,6 @@ impl GenerationService {
         Ok(ActiveGeneration {
             events: receiver,
             cancel,
-            started_at,
         })
     }
 }
@@ -335,7 +335,7 @@ fn run_command(
             rebuild_system_prompt,
         } = &command.operation
         {
-            let CheckpointTarget::Local(checkpoint) = &command.checkpoint else {
+            let CheckpointTarget::Local { checkpoint, .. } = &command.checkpoint else {
                 unreachable!("compaction checkpoints are local")
             };
             let result = generator.compact(
@@ -367,8 +367,12 @@ fn run_command(
             return;
         }
         let result = match command.checkpoint {
-            CheckpointTarget::Local(path) => generator.generate(
-                &path,
+            CheckpointTarget::Local {
+                checkpoint,
+                bootstrap,
+            } => generator.generate(
+                &checkpoint,
+                bootstrap.as_deref(),
                 &command.messages,
                 &command.turn,
                 &command.cancel,
@@ -440,7 +444,6 @@ mod tests {
         drop(ActiveGeneration {
             events,
             cancel: Arc::clone(&cancel),
-            started_at: Instant::now(),
         });
 
         assert!(cancel.load(Ordering::Relaxed));