MeiliSearch/src/lib.rs

123 lines
4.4 KiB
Rust
Raw Normal View History

2020-08-12 10:43:02 +02:00
mod criterion;
mod query_tokens;
2020-08-13 14:15:05 +02:00
mod search;
2020-08-28 14:16:37 +02:00
pub mod heed_codec;
pub mod proximity;
2020-08-30 21:50:30 +02:00
pub mod tokenizer;
2020-06-04 20:25:51 +02:00
2020-08-13 14:15:05 +02:00
use std::collections::HashMap;
2020-05-31 16:09:34 +02:00
use std::hash::BuildHasherDefault;
use anyhow::Context;
use csv::StringRecord;
use fxhash::{FxHasher32, FxHasher64};
2020-05-31 17:48:13 +02:00
use heed::types::*;
use heed::{PolyDatabase, Database};
2020-06-04 20:25:51 +02:00
2020-08-13 14:15:05 +02:00
pub use self::search::{Search, SearchResult};
pub use self::criterion::{Criterion, default_criteria};
pub use self::heed_codec::{
RoaringBitmapCodec, BEU32StrCodec, CsvStringRecordCodec,
ByteorderXRoaringBitmapCodec,
};
2020-05-31 16:09:34 +02:00
pub type FastMap4<K, V> = HashMap<K, V, BuildHasherDefault<FxHasher32>>;
pub type FastMap8<K, V> = HashMap<K, V, BuildHasherDefault<FxHasher64>>;
2020-05-31 16:09:34 +02:00
pub type SmallString32 = smallstr::SmallString<[u8; 32]>;
2020-06-11 11:55:03 +02:00
pub type SmallVec32<T> = smallvec::SmallVec<[T; 32]>;
pub type SmallVec16<T> = smallvec::SmallVec<[T; 16]>;
2020-05-31 16:09:34 +02:00
pub type BEU32 = heed::zerocopy::U32<heed::byteorder::BE>;
pub type DocumentId = u32;
pub type Attribute = u32;
pub type Position = u32;
2020-05-31 16:09:34 +02:00
const WORDS_FST_KEY: &str = "words-fst";
const HEADERS_KEY: &str = "headers";
const DOCUMENTS_IDS_KEY: &str = "documents-ids";
2020-05-31 17:48:13 +02:00
#[derive(Clone)]
pub struct Index {
/// Contains many different types (e.g. the documents CSV headers).
2020-05-31 17:48:13 +02:00
pub main: PolyDatabase,
/// A word and all the documents ids containing the word.
pub word_docids: Database<Str, RoaringBitmapCodec>,
/// Maps a word and a document id (u32) to all the positions where the given word appears.
pub docid_word_positions: Database<BEU32StrCodec, ByteorderXRoaringBitmapCodec>,
/// Maps the document id to the document as a CSV line.
pub documents: Database<OwnedType<BEU32>, ByteSlice>,
2020-05-31 17:48:13 +02:00
}
impl Index {
pub fn new(env: &heed::Env) -> anyhow::Result<Index> {
Ok(Index {
main: env.create_poly_database(None)?,
word_docids: env.create_database(Some("word-docids"))?,
docid_word_positions: env.create_database(Some("docid-word-positions"))?,
documents: env.create_database(Some("documents"))?,
})
2020-05-31 17:48:13 +02:00
}
pub fn put_headers(&self, wtxn: &mut heed::RwTxn, headers: &StringRecord) -> heed::Result<()> {
self.main.put::<_, Str, CsvStringRecordCodec>(wtxn, HEADERS_KEY, headers)
2020-06-29 13:54:47 +02:00
}
pub fn headers(&self, rtxn: &heed::RoTxn) -> heed::Result<Option<StringRecord>> {
self.main.get::<_, Str, CsvStringRecordCodec>(rtxn, HEADERS_KEY)
2020-05-31 17:48:13 +02:00
}
pub fn number_of_attributes(&self, rtxn: &heed::RoTxn) -> anyhow::Result<Option<usize>> {
match self.headers(rtxn)? {
Some(headers) => Ok(Some(headers.len())),
None => Ok(None),
}
}
2020-06-29 13:54:47 +02:00
pub fn put_fst<A: AsRef<[u8]>>(&self, wtxn: &mut heed::RwTxn, fst: &fst::Set<A>) -> anyhow::Result<()> {
Ok(self.main.put::<_, Str, ByteSlice>(wtxn, WORDS_FST_KEY, fst.as_fst().as_bytes())?)
2020-06-29 13:54:47 +02:00
}
pub fn fst<'t>(&self, rtxn: &'t heed::RoTxn) -> anyhow::Result<Option<fst::Set<&'t [u8]>>> {
match self.main.get::<_, Str, ByteSlice>(rtxn, WORDS_FST_KEY)? {
Some(bytes) => Ok(Some(fst::Set::new(bytes)?)),
None => Ok(None),
}
}
/// Returns a [`Vec`] of the requested documents. Returns an error if a document is missing.
pub fn documents<'t>(
&self,
rtxn: &'t heed::RoTxn,
iter: impl IntoIterator<Item=DocumentId>,
) -> anyhow::Result<Vec<(DocumentId, StringRecord)>>
{
let ids: Vec<_> = iter.into_iter().collect();
let mut content = Vec::new();
for id in ids.iter().cloned() {
let document_content = self.documents.get(rtxn, &BEU32::new(id))?
.with_context(|| format!("Could not find document {}", id))?;
content.extend_from_slice(document_content);
}
let mut rdr = csv::ReaderBuilder::new().has_headers(false).from_reader(&content[..]);
let mut documents = Vec::with_capacity(ids.len());
for (id, result) in ids.into_iter().zip(rdr.records()) {
documents.push((id, result?));
}
Ok(documents)
}
/// Returns the number of documents indexed in the database.
pub fn number_of_documents<'t>(&self, rtxn: &'t heed::RoTxn) -> anyhow::Result<usize> {
let docids = self.main.get::<_, Str, RoaringBitmapCodec>(rtxn, DOCUMENTS_IDS_KEY)?
.with_context(|| format!("Could not find the list of documents ids"))?;
Ok(docids.len() as usize)
}
2020-08-13 14:15:05 +02:00
pub fn search<'a>(&'a self, rtxn: &'a heed::RoTxn) -> Search<'a> {
Search::new(rtxn, self)
2020-05-31 17:48:13 +02:00
}
}