use crate::memory::model::MemoryItem;
use serde::{Deserialize, Serialize};
use std::collections::HashMap;
/// 索引文件版本号。
///
/// 变更索引规则（例如关键字归一化策略）时递增，以触发旧索引自动重建。
pub const INDEX_VERSION: u32 = 1;
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct IndexItem {
pub id: String,
pub offset: u64,
pub length: u32,
pub recorded_at_ts: i64,
#[serde(skip_serializing_if = "Option::is_none")]
pub occurred_at_ts: Option<i64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub importance: Option<u8>,
pub keywords: Vec<String>,
}
impl IndexItem {
pub fn time_key_ts(&self) -> i64 {
self.occurred_at_ts.unwrap_or(self.recorded_at_ts)
}
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct IndexData {
pub version: u32,
pub namespace: String,
pub memories_file: String,
pub indexed_up_to_offset: u64,
pub items: Vec<IndexItem>,
pub keyword_postings: HashMap<String, Vec<u32>>,
pub time_sorted: Vec<u32>,
pub time_sorted_dirty: bool,
}
impl IndexData {
pub fn new(namespace: &str) -> Self {
Self {
version: INDEX_VERSION,
namespace: namespace.to_string(),
memories_file: "memories.jsonl".to_string(),
indexed_up_to_offset: 0,
items: Vec::new(),
keyword_postings: HashMap::new(),
time_sorted: Vec::new(),
time_sorted_dirty: false,
}
}
pub fn add_memory_item(
&mut self,
item: &MemoryItem,
offset: u64,
length: u32,
recorded_at_ts: i64,
occurred_at_ts: Option<i64>,
keywords: Vec<String>,
) {
let idx = self.items.len() as u32;
self.items.push(IndexItem {
id: item.id.clone(),
offset,
length,
recorded_at_ts,
occurred_at_ts,
importance: item.importance,
keywords: keywords.clone(),
});
for kw in keywords {
self.keyword_postings.entry(kw).or_default().push(idx);
}
self.time_sorted.push(idx);
self.time_sorted_dirty = true;
}
pub fn ensure_time_sorted(&mut self) {
if !self.time_sorted_dirty {
return;
}
let items = &self.items;
self.time_sorted.sort_by_key(|idx| {
let i = *idx as usize;
items.get(i).map(|x| x.time_key_ts()).unwrap_or(0)
});
self.time_sorted_dirty = false;
}
}
