Fast startup: mmap backward scan instead of reading full log

Uses JsonlBackwardIter (SIMD memrchr3) to scan the conversation log
newest-first without reading/parsing the whole file. Stops as soon
as the conversation budget is full. Only the kept nodes get
retokenized and pushed into context.

18MB log → only tokenize the ~50 nodes that fit in the budget.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/mind/log.rs

@@ -1,8 +1,10 @@
 use anyhow::{Context, Result};
 use std::fs::{File, OpenOptions};
-use std::io::{BufRead, BufReader, Seek, SeekFrom, Write};
+use std::io::Write;
 use std::path::{Path, PathBuf};
 use crate::agent::context::AstNode;
+use crate::hippocampus::transcript::JsonlBackwardIter;
+use memmap2::Mmap;
 
 pub struct ConversationLog {
     path: PathBuf,
@@ -33,32 +35,19 @@ impl ConversationLog {
         Ok(())
     }
 
-    pub fn read_nodes(&self, max_bytes: u64) -> Result<Vec<AstNode>> {
+    /// Read nodes from the tail of the log, newest first.
+    /// Caller decides when to stop (budget, count, etc).
+    pub fn read_tail(&self) -> Result<TailNodes> {
         if !self.path.exists() {
-            return Ok(Vec::new());
+            anyhow::bail!("log does not exist");
         }
         let file = File::open(&self.path)
             .with_context(|| format!("opening log {}", self.path.display()))?;
-        let file_len = file.metadata()?.len();
-        let mut reader = BufReader::new(file);
-
-        if file_len > max_bytes {
-            reader.seek(SeekFrom::Start(file_len - max_bytes))?;
-            let mut discard = String::new();
-            reader.read_line(&mut discard)?;
+        if file.metadata()?.len() == 0 {
+            anyhow::bail!("log is empty");
         }
-
-        let mut nodes = Vec::new();
-        for line in reader.lines() {
-            let line = line.context("reading log tail")?;
-            let line = line.trim();
-            if line.is_empty() { continue; }
-            if let Ok(node) = serde_json::from_str::<AstNode>(line) {
-                nodes.push(node);
-            }
-            // Old format entries silently skipped — journal has the context
-        }
-        Ok(nodes)
+        let mmap = unsafe { Mmap::map(&file)? };
+        Ok(TailNodes { _file: file, mmap })
     }
 
     pub fn path(&self) -> &Path {
@@ -66,12 +55,13 @@ impl ConversationLog {
     }
 
     pub fn oldest_timestamp(&self) -> Option<chrono::DateTime<chrono::Utc>> {
+        // Read forward from the start to find first timestamp
         let file = File::open(&self.path).ok()?;
-        let reader = BufReader::new(file);
-        for line in reader.lines().flatten() {
-            let line = line.trim().to_string();
+        let mmap = unsafe { Mmap::map(&file).ok()? };
+        // Find first { ... } and parse
+        for line in mmap.split(|&b| b == b'\n') {
             if line.is_empty() { continue; }
-            if let Ok(node) = serde_json::from_str::<AstNode>(&line) {
+            if let Ok(node) = serde_json::from_slice::<AstNode>(line) {
                 if let Some(leaf) = node.leaf() {
                     if let Some(ts) = leaf.timestamp() {
                         return Some(ts);
@@ -82,3 +72,16 @@ impl ConversationLog {
         None
     }
 }
+
+/// Iterates over conversation log nodes newest-first, using mmap + backward scan.
+pub struct TailNodes {
+    _file: File,
+    mmap: Mmap,
+}
+
+impl TailNodes {
+    pub fn iter(&self) -> impl Iterator<Item = AstNode> + '_ {
+        JsonlBackwardIter::new(&self.mmap)
+            .filter_map(|bytes| serde_json::from_slice::<AstNode>(bytes).ok())
+    }
+}