Update session tree design to use UUIDs instead of indices

This commit is contained in:
Mario Zechner 2025-12-25 21:19:02 +01:00
parent 351faef604
commit 64e7c80c7e

View file

@ -16,65 +16,71 @@ Context is built by scanning linearly, applying compaction ranges.
## Proposed Format (Tree)
Each entry has a `parent` field pointing to its parent entry's index (-1 for root):
Each entry has a `uuid` and `parentUuid` field (null for root):
```jsonl
{"type":"session","id":"...","parent":-1} // 0: root
{"type":"message","parent":0,"message":{"role":"user",...}} // 1
{"type":"message","parent":1,"message":{"role":"assistant",...}} // 2
{"type":"message","parent":2,"message":{"role":"user",...}} // 3
{"type":"message","parent":3,"message":{"role":"assistant",...}} // 4
{"type":"head","parent":4,"name":"main"} // 5: current head
{"type":"session","uuid":"a1b2c3","parentUuid":null,"id":"...","cwd":"..."}
{"type":"message","uuid":"d4e5f6","parentUuid":"a1b2c3","message":{"role":"user",...}}
{"type":"message","uuid":"g7h8i9","parentUuid":"d4e5f6","message":{"role":"assistant",...}}
{"type":"message","uuid":"j0k1l2","parentUuid":"g7h8i9","message":{"role":"user",...}}
{"type":"message","uuid":"m3n4o5","parentUuid":"j0k1l2","message":{"role":"assistant",...}}
```
The **last entry** is always the current leaf. Context = walk from leaf to root.
The **last entry** is always the current leaf. Context = walk from leaf to root via `parentUuid`.
Using UUIDs (like Claude Code does) instead of indices because:
- No remapping needed when branching to new file
- Robust to entry deletion/reordering
- Orphan references are detectable
- ~30 extra bytes per entry is negligible for text-heavy sessions
### Branching
Branch from entry 2 (after first assistant response):
Branch from entry `g7h8i9` (after first assistant response):
```jsonl
... entries 0-5 unchanged ...
{"type":"message","parent":2,"message":{"role":"user",...}} // 6: new branch
{"type":"message","parent":6,"message":{"role":"assistant",...}} // 7
{"type":"head","parent":7,"name":"main"} // 8: new head
... entries unchanged ...
{"type":"message","uuid":"p6q7r8","parentUuid":"g7h8i9","message":{"role":"user",...}}
{"type":"message","uuid":"s9t0u1","parentUuid":"p6q7r8","message":{"role":"assistant",...}}
```
Now entry 8 is the leaf. Walking 8→7→6→2→1→0 gives the branched context.
Walking s9t0u1→p6q7r8→g7h8i9→d4e5f6→a1b2c3 gives the branched context.
The old path (entries 3-5) remains in the file but is not in the current context.
The old path (j0k1l2, m3n4o5) remains in the file but is not in the current context.
### Visual
```
[0:session]
[a1b2:session]
[1:user "hello"]
[d4e5:user "hello"]
[2:assistant "hi"]
[g7h8:assistant "hi"]
┌────┴────┐
│ │
[3:user A] [6:user B] ← branch point at 2
│ │
[4:asst A] [7:asst B]
(old) [8:head] ← current leaf
[j0k1:user A] [p6q7:user B] ← branch point
[m3n4:asst A] [s9t0:asst B] ← current leaf
(old path)
```
## Context Building
```typescript
function buildContext(entries: SessionEntry[]): AppMessage[] {
// Find current head (last entry, or last head entry)
let current = entries.length - 1;
// Build UUID -> entry map
const byUuid = new Map(entries.map(e => [e.uuid, e]));
// Start from last entry (current leaf)
let current: SessionEntry | undefined = entries[entries.length - 1];
// Walk to root, collecting messages
const path: SessionEntry[] = [];
while (current >= 0) {
const entry = entries[current];
path.unshift(entry);
current = entry.parent;
while (current) {
path.unshift(current);
current = current.parentUuid ? byUuid.get(current.parentUuid) : undefined;
}
// Extract messages, apply compaction summaries
@ -82,7 +88,7 @@ function buildContext(entries: SessionEntry[]): AppMessage[] {
}
```
Complexity: O(depth) instead of O(n) for linear scan.
Complexity: O(n) to build map, O(depth) to walk. Total O(n), but walk is fast.
## Consequences for Stacking
@ -106,31 +112,29 @@ Context building requires tracking ranges, IDs, "later wins" logic.
Stacking becomes trivial branching:
```jsonl
... conversation entries 0-10 ...
{"type":"stack_summary","parent":3,"summary":"Work done in entries 4-10"} // 11
{"type":"head","parent":11} // 12
... conversation entries ...
{"type":"stack_summary","uuid":"x1y2z3","parentUuid":"g7h8i9","summary":"Work done after this point"}
```
To "pop" to entry 3:
1. Generate summary of entries 4-10
2. Append summary entry with `parent:3`
3. Append new head
To "pop" to entry `g7h8i9`:
1. Generate summary of entries after `g7h8i9`
2. Append summary entry with `parentUuid: "g7h8i9"`
Context walk: 12→11→3→2→1→0. Entries 4-10 are not traversed.
Context walk follows parentUuid chain. Abandoned entries are not traversed.
**No range tracking. No overlap rules. No "later wins" logic.**
### Multiple Pops
```
[0]─[1]─[2]─[3]─[4]─[5]─[6]─[7]─[8]
[a]─[b]─[c]─[d]─[e]─[f]─[g]─[h]
└─[9:summary of 4-8]─[10]─[11]─[12]
└─[13:summary of 10-12]─[14:head]
└─[i:summary]─[j]─[k]─[l]
└─[m:summary]─[n:current]
```
Each pop just creates a new branch. Context: 14→13→9→3→2→1→0.
Each pop just creates a new branch. Context: n→m→k→j→i→c→b→a.
## Consequences for Compaction
@ -143,27 +147,27 @@ Compaction stores `firstKeptEntryIndex` and requires careful handling when stack
Compaction creates a summary node:
```jsonl
{"type":"compaction","parent":0,"summary":"...","summarizedPath":[1,2,3,4,5]} // 6
{"type":"message","parent":6,"message":{"role":"user",...}} // 7
{"type":"compaction","uuid":"c1","parentUuid":"a1b2c3","summary":"...","summarizedUuids":["d4e5f6","g7h8i9"]}
{"type":"message","uuid":"m1","parentUuid":"c1","message":{"role":"user",...}}
```
The compaction node's `parent:0` means it attaches to root. Walking from 7: 7→6→0.
The compaction node's `parentUuid` points to root. Walking from m1: m1→c1→a1b2c3.
Entries 1-5 are still in the file (for export, debugging) but not in context.
Summarized entries are still in the file (for export, debugging) but not in context.
### Compaction + Stacking
No special handling needed. They're both just branches:
```
[0]─[1]─[2]─[3]─[4]─[5]
└─[6:compaction summary]─[7]─[8]─[9]─[10]
└─[11:stack summary]─[12:head]
[root]─[msg1]─[msg2]─[msg3]─[msg4]─[msg5]
└─[compaction]─[msg6]─[msg7]─[msg8]
└─[stack_summary]─[msg9:current]
```
Context: 12→11→7→6→0. Clean.
Context: msg9→stack_summary→msg6→compaction→root. Clean.
## Consequences for API
@ -172,32 +176,36 @@ Context: 12→11→7→6→0. Clean.
```typescript
interface SessionEntry {
type: string;
parent: number; // NEW: -1 for root, otherwise index of parent
uuid: string; // NEW: unique identifier
parentUuid: string | null; // NEW: null for root
timestamp?: string;
// ... type-specific fields
}
class SessionManager {
// NEW: Get current head index
getCurrentHead(): number;
// NEW: Get current leaf entry
getCurrentLeaf(): SessionEntry;
// NEW: Walk from entry to root
getPath(fromIndex?: number): SessionEntry[];
getPath(fromUuid?: string): SessionEntry[];
// NEW: Get entry by UUID
getEntry(uuid: string): SessionEntry | undefined;
// CHANGED: Uses tree walk instead of linear scan
buildSessionContext(): SessionContext;
// NEW: Create branch point, returns new head index
branch(parentIndex: number): number;
// NEW: Create branch point
branch(parentUuid: string): string; // returns new entry's uuid
// NEW: Create branch with summary of abandoned subtree
branchWithSummary(parentIndex: number, summary: string): number;
branchWithSummary(parentUuid: string, summary: string): string;
// CHANGED: Simpler, just creates summary node
saveCompaction(entry: CompactionEntry): void;
// UNCHANGED
saveMessage(message: AppMessage): void;
// CHANGED: Now requires parentUuid (uses current leaf if omitted)
saveMessage(message: AppMessage, parentUuid?: string): void;
saveEntry(entry: SessionEntry): void;
}
```
@ -207,10 +215,10 @@ class SessionManager {
```typescript
class AgentSession {
// CHANGED: Uses tree-based branching
async branch(entryIndex: number): Promise<BranchResult>;
async branch(entryUuid: string): Promise<BranchResult>;
// NEW: Branch in current session (no new file)
async branchInPlace(entryIndex: number, options?: {
async branchInPlace(entryUuid: string, options?: {
summarize?: boolean; // Generate summary of abandoned subtree
}): Promise<void>;
@ -235,10 +243,10 @@ interface BranchResult {
interface HookEventContext {
// NEW: Tree-aware entry access
entries: readonly SessionEntry[];
currentPath: readonly SessionEntry[]; // Entries from root to current head
currentPath: readonly SessionEntry[]; // Entries from root to current leaf
// NEW: Branch without creating new file
branchInPlace(parentIndex: number, summary?: string): Promise<void>;
branchInPlace(parentUuid: string, summary?: string): Promise<void>;
// Existing
saveEntry(entry: SessionEntry): Promise<void>;
@ -263,7 +271,7 @@ Use case: Quick "let me try something else" without file proliferation.
```
/branches → List all branches in current session
/switch 3 → Switch to branch starting at entry 3
/switch <uuid> → Switch to branch at entry
```
The session file contains full history. UI can visualize the tree.
@ -274,7 +282,7 @@ No hooks needed for basic stacking:
```
/pop → Branch to previous user message with auto-summary
/pop 5 → Branch to entry 5 with auto-summary
/pop <uuid> → Branch to specific entry with auto-summary
```
Core functionality, not hook-dependent.
@ -282,32 +290,37 @@ Core functionality, not hook-dependent.
### 4. Subtree Export
```
/export-branch 3 → Export just the subtree from entry 3
/export-branch <uuid> → Export just the subtree from entry
```
Useful for sharing specific conversation paths.
Useful for sharing specific conversation paths. No index remapping needed since UUIDs are stable.
### 5. Merge/Cherry-pick (Future)
With tree structure, could support:
```
/cherry-pick 7 → Copy entry 7's message to current branch
/merge 3 → Merge branch at entry 3 into current
/cherry-pick <uuid> → Copy entry's message to current branch
/merge <uuid> → Merge branch into current
```
## Migration
### File Format
Add `parent` field to all entries. Existing sessions get `parent: previousIndex`:
Add `uuid` and `parentUuid` fields to all entries. Existing sessions get generated UUIDs with linear parentage:
```typescript
function migrateSession(content: string): string {
const lines = content.trim().split('\n');
const uuids: string[] = [];
return lines.map((line, i) => {
const entry = JSON.parse(line);
entry.parent = i === 0 ? -1 : i - 1; // Linear chain
const uuid = generateUuid();
uuids.push(uuid);
entry.uuid = uuid;
entry.parentUuid = i === 0 ? null : uuids[i - 1];
return JSON.stringify(entry);
}).join('\n');
}
@ -318,7 +331,7 @@ Migrated sessions work exactly as before (linear path).
### API Compatibility
- `buildSessionContext()` returns same structure
- `branch()` still works, just more options
- `branch()` still works, just uses UUIDs
- Existing hooks continue to work
## Complexity Analysis
@ -326,36 +339,35 @@ Migrated sessions work exactly as before (linear path).
| Operation | Linear | Tree |
|-----------|--------|------|
| Append message | O(1) | O(1) |
| Build context | O(n) | O(depth) |
| Branch | O(n) copy | O(1) append |
| Find entry | O(n) | O(n) |
| Build context | O(n) | O(n) map + O(depth) walk |
| Branch to new file | O(n) copy | O(path) copy, no remapping |
| Find entry by UUID | O(n) | O(1) with map |
| Compaction | O(n) | O(depth) |
Tree is better or equal for all operations except "find arbitrary entry" (same).
Tree with UUIDs is comparable or better. The UUID map can be cached.
## File Size
Tree format adds ~15 bytes per entry (`"parent":123,`). For 1000-entry session: ~15KB overhead. Negligible.
Tree format adds ~50 bytes per entry (`"uuid":"...","parentUuid":"..."`, 36 chars each). For 1000-entry session: ~50KB overhead. Negligible for text-heavy sessions.
Abandoned branches remain in file but don't affect context building performance.
## Example: Full Session with Branching
```jsonl
{"type":"session","id":"abc","parent":-1,"cwd":"/project"}
{"type":"message","parent":0,"message":{"role":"user","content":"Build a CLI"}}
{"type":"message","parent":1,"message":{"role":"assistant","content":"I'll create..."}}
{"type":"message","parent":2,"message":{"role":"user","content":"Add --verbose flag"}}
{"type":"message","parent":3,"message":{"role":"assistant","content":"Here's the flag..."}}
{"type":"message","parent":4,"message":{"role":"user","content":"Actually use Python"}}
{"type":"message","parent":5,"message":{"role":"assistant","content":"Converting to Python..."}}
{"type":"branch_summary","parent":2,"summary":"Attempted Node.js CLI with --verbose flag"}
{"type":"message","parent":7,"message":{"role":"user","content":"Use Rust instead"}}
{"type":"message","parent":8,"message":{"role":"assistant","content":"Creating Rust CLI..."}}
{"type":"head","parent":9}
{"type":"session","uuid":"ses1","parentUuid":null,"id":"abc","cwd":"/project"}
{"type":"message","uuid":"m1","parentUuid":"ses1","message":{"role":"user","content":"Build a CLI"}}
{"type":"message","uuid":"m2","parentUuid":"m1","message":{"role":"assistant","content":"I'll create..."}}
{"type":"message","uuid":"m3","parentUuid":"m2","message":{"role":"user","content":"Add --verbose flag"}}
{"type":"message","uuid":"m4","parentUuid":"m3","message":{"role":"assistant","content":"Here's the flag..."}}
{"type":"message","uuid":"m5","parentUuid":"m4","message":{"role":"user","content":"Actually use Python"}}
{"type":"message","uuid":"m6","parentUuid":"m5","message":{"role":"assistant","content":"Converting to Python..."}}
{"type":"branch_summary","uuid":"bs1","parentUuid":"m2","summary":"Attempted Node.js CLI with --verbose flag"}
{"type":"message","uuid":"m7","parentUuid":"bs1","message":{"role":"user","content":"Use Rust instead"}}
{"type":"message","uuid":"m8","parentUuid":"m7","message":{"role":"assistant","content":"Creating Rust CLI..."}}
```
Context path: 10→9→8→7→2→1→0
Context path: m8→m7→bs1→m2→m1→ses1
Result:
1. User: "Build a CLI"
@ -364,16 +376,25 @@ Result:
4. User: "Use Rust instead"
5. Assistant: "Creating Rust CLI..."
Entries 3-6 (the Node.js path) are preserved but not in context.
Entries m3-m6 (the Node.js/Python path) are preserved but not in context.
## Prior Art
Claude Code uses the same approach:
- `uuid` field on each entry
- `parentUuid` links to parent (null for root)
- `leafUuid` in summary entries to track conversation endpoints
- Separate files for sidechains (`isSidechain: true`)
## Recommendation
The tree format:
The tree format with UUIDs:
- Simplifies stacking (no range overlap logic)
- Simplifies compaction (no boundary crossing)
- Enables in-place branching
- Enables branch visualization/navigation
- No index remapping on branch-to-file
- Maintains backward compatibility
- Has negligible overhead
- Validated by Claude Code's implementation
**Recommend implementing for v2 of hooks/session system.**