2021-04-19 22:22:41 +08:00
|
|
|
use std::borrow::Cow;
|
2021-03-16 01:11:10 +08:00
|
|
|
use std::collections::{BTreeMap, HashSet};
|
|
|
|
use std::time::Instant;
|
2021-03-04 18:56:32 +08:00
|
|
|
|
|
|
|
use anyhow::bail;
|
2021-03-16 01:11:10 +08:00
|
|
|
use either::Either;
|
2021-03-04 18:56:32 +08:00
|
|
|
use heed::RoTxn;
|
|
|
|
use meilisearch_tokenizer::{Analyzer, AnalyzerConfig};
|
2021-04-19 22:22:41 +08:00
|
|
|
use milli::{facet::FacetValue, FacetCondition, FieldId, FieldsIdsMap, MatchingWords};
|
2021-03-16 01:11:10 +08:00
|
|
|
use serde::{Deserialize, Serialize};
|
|
|
|
use serde_json::{Map, Value};
|
2021-03-04 18:56:32 +08:00
|
|
|
|
|
|
|
use super::Index;
|
|
|
|
|
2021-04-19 22:22:41 +08:00
|
|
|
pub type Document = Map<String, Value>;
|
|
|
|
|
2021-03-04 18:56:32 +08:00
|
|
|
pub const DEFAULT_SEARCH_LIMIT: usize = 20;
|
|
|
|
|
|
|
|
const fn default_search_limit() -> usize {
|
|
|
|
DEFAULT_SEARCH_LIMIT
|
|
|
|
}
|
|
|
|
|
|
|
|
#[derive(Deserialize)]
|
|
|
|
#[serde(rename_all = "camelCase", deny_unknown_fields)]
|
|
|
|
pub struct SearchQuery {
|
|
|
|
pub q: Option<String>,
|
|
|
|
pub offset: Option<usize>,
|
|
|
|
#[serde(default = "default_search_limit")]
|
|
|
|
pub limit: usize,
|
2021-04-19 22:22:41 +08:00
|
|
|
pub attributes_to_retrieve: Option<HashSet<String>>,
|
|
|
|
pub attributes_to_crop: Option<HashSet<String>>,
|
2021-03-04 18:56:32 +08:00
|
|
|
pub crop_length: Option<usize>,
|
|
|
|
pub attributes_to_highlight: Option<HashSet<String>>,
|
|
|
|
pub filters: Option<String>,
|
|
|
|
pub matches: Option<bool>,
|
|
|
|
pub facet_filters: Option<Value>,
|
|
|
|
pub facet_distributions: Option<Vec<String>>,
|
|
|
|
}
|
|
|
|
|
2021-04-19 16:13:13 +08:00
|
|
|
#[derive(Debug, Clone, Serialize)]
|
|
|
|
pub struct SearchHit {
|
|
|
|
#[serde(flatten)]
|
2021-04-19 22:22:41 +08:00
|
|
|
pub document: Document,
|
2021-04-20 19:10:50 +08:00
|
|
|
#[serde(rename = "_formatted", skip_serializing_if = "Document::is_empty")]
|
2021-04-19 22:22:41 +08:00
|
|
|
pub formatted: Document,
|
2021-04-19 16:13:13 +08:00
|
|
|
}
|
|
|
|
|
2021-03-04 18:56:32 +08:00
|
|
|
#[derive(Serialize)]
|
|
|
|
#[serde(rename_all = "camelCase")]
|
|
|
|
pub struct SearchResult {
|
2021-04-19 16:13:13 +08:00
|
|
|
pub hits: Vec<SearchHit>,
|
2021-03-04 18:56:32 +08:00
|
|
|
pub nb_hits: u64,
|
2021-03-24 02:13:22 +08:00
|
|
|
pub exhaustive_nb_hits: bool,
|
2021-03-04 18:56:32 +08:00
|
|
|
pub query: String,
|
|
|
|
pub limit: usize,
|
|
|
|
pub offset: usize,
|
|
|
|
pub processing_time_ms: u128,
|
|
|
|
#[serde(skip_serializing_if = "Option::is_none")]
|
|
|
|
pub facet_distributions: Option<BTreeMap<String, BTreeMap<FacetValue, u64>>>,
|
|
|
|
}
|
|
|
|
|
|
|
|
impl Index {
|
|
|
|
pub fn perform_search(&self, query: SearchQuery) -> anyhow::Result<SearchResult> {
|
|
|
|
let before_search = Instant::now();
|
|
|
|
let rtxn = self.read_txn()?;
|
|
|
|
|
|
|
|
let mut search = self.search(&rtxn);
|
|
|
|
|
|
|
|
if let Some(ref query) = query.q {
|
|
|
|
search.query(query);
|
|
|
|
}
|
|
|
|
|
|
|
|
search.limit(query.limit);
|
|
|
|
search.offset(query.offset.unwrap_or_default());
|
|
|
|
|
|
|
|
if let Some(ref facets) = query.facet_filters {
|
|
|
|
if let Some(facets) = parse_facets(facets, self, &rtxn)? {
|
|
|
|
search.facet_condition(facets);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
let milli::SearchResult {
|
|
|
|
documents_ids,
|
2021-03-12 02:40:18 +08:00
|
|
|
matching_words,
|
2021-03-04 18:56:32 +08:00
|
|
|
candidates,
|
|
|
|
..
|
|
|
|
} = search.execute()?;
|
|
|
|
let mut documents = Vec::new();
|
|
|
|
let fields_ids_map = self.fields_ids_map(&rtxn).unwrap();
|
|
|
|
|
2021-04-20 19:10:50 +08:00
|
|
|
let displayed_ids: HashSet<FieldId> = fields_ids_map.iter().map(|(id, _)| id).collect();
|
|
|
|
|
2021-04-19 22:22:41 +08:00
|
|
|
let fids = |attrs: &HashSet<String>| {
|
2021-04-20 19:10:50 +08:00
|
|
|
let mut ids = HashSet::new();
|
|
|
|
for attr in attrs {
|
|
|
|
if attr == "*" {
|
|
|
|
ids = displayed_ids.clone();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if let Some(id) = fields_ids_map.id(attr) {
|
|
|
|
ids.insert(id);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ids
|
2021-04-19 22:22:41 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
let to_retrieve_ids = query
|
|
|
|
.attributes_to_retrieve
|
|
|
|
.as_ref()
|
|
|
|
.map(fids)
|
|
|
|
.unwrap_or_else(|| displayed_ids.clone());
|
|
|
|
|
|
|
|
let to_highlight_ids = query
|
|
|
|
.attributes_to_highlight
|
|
|
|
.as_ref()
|
|
|
|
.map(fids)
|
|
|
|
.unwrap_or_default();
|
|
|
|
|
|
|
|
let to_crop_ids = query
|
|
|
|
.attributes_to_crop
|
|
|
|
.as_ref()
|
|
|
|
.map(fids)
|
|
|
|
.unwrap_or_default();
|
|
|
|
|
|
|
|
// The attributes to retrieve are:
|
|
|
|
// - the ones explicitly marked as to retrieve that are also in the displayed attributes
|
|
|
|
let all_attributes: Vec<_> = to_retrieve_ids
|
|
|
|
.intersection(&displayed_ids)
|
|
|
|
.cloned()
|
|
|
|
.collect();
|
|
|
|
|
|
|
|
// The formatted attributes are:
|
|
|
|
// - The one in either highlighted attributes or cropped attributes if there are attributes
|
|
|
|
// to retrieve
|
|
|
|
// - All the attributes to retrieve if there are either highlighted or cropped attributes
|
|
|
|
// the request specified that all attributes are to retrieve (i.e attributes to retrieve is
|
|
|
|
// empty in the query)
|
|
|
|
let all_formatted = if query.attributes_to_retrieve.is_none() {
|
|
|
|
if query.attributes_to_highlight.is_some() || query.attributes_to_crop.is_some() {
|
|
|
|
Cow::Borrowed(&all_attributes)
|
|
|
|
} else {
|
|
|
|
Cow::Owned(Vec::new())
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
let attrs = (&to_crop_ids | &to_highlight_ids)
|
|
|
|
.intersection(&displayed_ids)
|
|
|
|
.cloned()
|
|
|
|
.collect::<Vec<_>>();
|
|
|
|
Cow::Owned(attrs)
|
|
|
|
};
|
2021-03-04 18:56:32 +08:00
|
|
|
|
|
|
|
let stop_words = fst::Set::default();
|
2021-04-19 22:22:41 +08:00
|
|
|
let highlighter = Highlighter::new(
|
|
|
|
&stop_words,
|
|
|
|
(String::from("<mark>"), String::from("</mark>")),
|
|
|
|
);
|
2021-03-04 18:56:32 +08:00
|
|
|
|
|
|
|
for (_id, obkv) in self.documents(&rtxn, documents_ids)? {
|
2021-04-19 22:22:41 +08:00
|
|
|
let document = milli::obkv_to_json(&all_attributes, &fields_ids_map, obkv.clone())?;
|
|
|
|
let formatted = compute_formatted(
|
|
|
|
&fields_ids_map,
|
|
|
|
obkv,
|
|
|
|
&highlighter,
|
|
|
|
&matching_words,
|
|
|
|
all_formatted.as_ref().as_slice(),
|
|
|
|
&to_highlight_ids,
|
|
|
|
)?;
|
2021-04-19 16:13:13 +08:00
|
|
|
let hit = SearchHit {
|
2021-04-19 22:22:41 +08:00
|
|
|
document,
|
|
|
|
formatted,
|
2021-04-19 16:13:13 +08:00
|
|
|
};
|
|
|
|
documents.push(hit);
|
2021-03-04 18:56:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
let nb_hits = candidates.len();
|
|
|
|
|
|
|
|
let facet_distributions = match query.facet_distributions {
|
|
|
|
Some(ref fields) => {
|
|
|
|
let mut facet_distribution = self.facets_distribution(&rtxn);
|
|
|
|
if fields.iter().all(|f| f != "*") {
|
|
|
|
facet_distribution.facets(fields);
|
|
|
|
}
|
|
|
|
Some(facet_distribution.candidates(candidates).execute()?)
|
|
|
|
}
|
|
|
|
None => None,
|
|
|
|
};
|
|
|
|
|
|
|
|
let result = SearchResult {
|
2021-03-23 17:47:19 +08:00
|
|
|
exhaustive_nb_hits: false, // not implemented yet
|
2021-03-04 18:56:32 +08:00
|
|
|
hits: documents,
|
|
|
|
nb_hits,
|
|
|
|
query: query.q.clone().unwrap_or_default(),
|
|
|
|
limit: query.limit,
|
|
|
|
offset: query.offset.unwrap_or_default(),
|
|
|
|
processing_time_ms: before_search.elapsed().as_millis(),
|
|
|
|
facet_distributions,
|
|
|
|
};
|
|
|
|
Ok(result)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-04-19 22:22:41 +08:00
|
|
|
fn compute_formatted<A: AsRef<[u8]>>(
|
|
|
|
field_ids_map: &FieldsIdsMap,
|
|
|
|
obkv: obkv::KvReader,
|
|
|
|
highlighter: &Highlighter<A>,
|
2021-04-20 01:03:53 +08:00
|
|
|
matching_words: &impl Matcher,
|
2021-04-19 22:22:41 +08:00
|
|
|
all_formatted: &[FieldId],
|
|
|
|
to_highlight_ids: &HashSet<FieldId>,
|
|
|
|
) -> anyhow::Result<Document> {
|
|
|
|
let mut document = Document::new();
|
|
|
|
|
|
|
|
for field in all_formatted {
|
|
|
|
if let Some(value) = obkv.get(*field) {
|
|
|
|
let mut value: Value = serde_json::from_slice(value)?;
|
|
|
|
|
|
|
|
if to_highlight_ids.contains(field) {
|
|
|
|
value = highlighter.highlight_value(value, matching_words);
|
|
|
|
}
|
|
|
|
|
|
|
|
// This unwrap must be safe since we got the ids from the fields_ids_map just
|
|
|
|
// before.
|
|
|
|
let key = field_ids_map
|
|
|
|
.name(*field)
|
|
|
|
.expect("Missing field name")
|
|
|
|
.to_string();
|
|
|
|
|
|
|
|
document.insert(key, value);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
Ok(document)
|
|
|
|
}
|
|
|
|
|
2021-04-20 01:03:53 +08:00
|
|
|
/// trait to allow unit testing of `compute_formated`
|
|
|
|
trait Matcher {
|
|
|
|
fn matches(&self, w: &str) -> bool;
|
|
|
|
}
|
|
|
|
|
|
|
|
#[cfg(test)]
|
|
|
|
impl Matcher for HashSet<String> {
|
|
|
|
fn matches(&self, w: &str) -> bool {
|
|
|
|
self.contains(w)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
impl Matcher for MatchingWords {
|
|
|
|
fn matches(&self, w: &str) -> bool {
|
|
|
|
self.matches(w)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-04 18:56:32 +08:00
|
|
|
fn parse_facets_array(
|
|
|
|
txn: &RoTxn,
|
|
|
|
index: &Index,
|
2021-03-15 23:52:05 +08:00
|
|
|
arr: &[Value],
|
2021-03-04 18:56:32 +08:00
|
|
|
) -> anyhow::Result<Option<FacetCondition>> {
|
|
|
|
let mut ands = Vec::new();
|
|
|
|
for value in arr {
|
|
|
|
match value {
|
|
|
|
Value::String(s) => ands.push(Either::Right(s.clone())),
|
|
|
|
Value::Array(arr) => {
|
|
|
|
let mut ors = Vec::new();
|
|
|
|
for value in arr {
|
|
|
|
match value {
|
|
|
|
Value::String(s) => ors.push(s.clone()),
|
|
|
|
v => bail!("Invalid facet expression, expected String, found: {:?}", v),
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ands.push(Either::Left(ors));
|
|
|
|
}
|
|
|
|
v => bail!(
|
|
|
|
"Invalid facet expression, expected String or [String], found: {:?}",
|
|
|
|
v
|
|
|
|
),
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
FacetCondition::from_array(txn, &index.0, ands)
|
|
|
|
}
|
|
|
|
|
2021-04-20 01:03:53 +08:00
|
|
|
struct Highlighter<'a, A> {
|
2021-03-04 18:56:32 +08:00
|
|
|
analyzer: Analyzer<'a, A>,
|
2021-04-19 22:22:41 +08:00
|
|
|
marks: (String, String),
|
2021-03-04 18:56:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
impl<'a, A: AsRef<[u8]>> Highlighter<'a, A> {
|
2021-04-19 22:22:41 +08:00
|
|
|
pub fn new(stop_words: &'a fst::Set<A>, marks: (String, String)) -> Self {
|
2021-04-01 22:49:11 +08:00
|
|
|
let mut config = AnalyzerConfig::default();
|
|
|
|
config.stop_words(stop_words);
|
|
|
|
|
|
|
|
let analyzer = Analyzer::new(config);
|
2021-03-04 18:56:32 +08:00
|
|
|
|
2021-04-19 22:22:41 +08:00
|
|
|
Self { analyzer, marks }
|
2021-03-04 18:56:32 +08:00
|
|
|
}
|
|
|
|
|
2021-04-20 01:03:53 +08:00
|
|
|
fn highlight_value(&self, value: Value, words_to_highlight: &impl Matcher) -> Value {
|
2021-03-04 18:56:32 +08:00
|
|
|
match value {
|
|
|
|
Value::Null => Value::Null,
|
|
|
|
Value::Bool(boolean) => Value::Bool(boolean),
|
|
|
|
Value::Number(number) => Value::Number(number),
|
|
|
|
Value::String(old_string) => {
|
|
|
|
let mut string = String::new();
|
|
|
|
let analyzed = self.analyzer.analyze(&old_string);
|
|
|
|
for (word, token) in analyzed.reconstruct() {
|
|
|
|
if token.is_word() {
|
2021-03-12 02:40:18 +08:00
|
|
|
let to_highlight = words_to_highlight.matches(token.text());
|
2021-03-04 18:56:32 +08:00
|
|
|
if to_highlight {
|
2021-04-19 22:22:41 +08:00
|
|
|
string.push_str(&self.marks.0)
|
2021-03-04 18:56:32 +08:00
|
|
|
}
|
|
|
|
string.push_str(word);
|
|
|
|
if to_highlight {
|
2021-04-19 22:22:41 +08:00
|
|
|
string.push_str(&self.marks.1)
|
2021-03-04 18:56:32 +08:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
string.push_str(word);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
Value::String(string)
|
|
|
|
}
|
|
|
|
Value::Array(values) => Value::Array(
|
|
|
|
values
|
2021-03-16 01:11:10 +08:00
|
|
|
.into_iter()
|
|
|
|
.map(|v| self.highlight_value(v, words_to_highlight))
|
|
|
|
.collect(),
|
2021-03-04 18:56:32 +08:00
|
|
|
),
|
|
|
|
Value::Object(object) => Value::Object(
|
|
|
|
object
|
2021-03-16 01:11:10 +08:00
|
|
|
.into_iter()
|
|
|
|
.map(|(k, v)| (k, self.highlight_value(v, words_to_highlight)))
|
|
|
|
.collect(),
|
2021-03-04 18:56:32 +08:00
|
|
|
),
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn parse_facets(
|
|
|
|
facets: &Value,
|
|
|
|
index: &Index,
|
|
|
|
txn: &RoTxn,
|
|
|
|
) -> anyhow::Result<Option<FacetCondition>> {
|
|
|
|
match facets {
|
|
|
|
// Disabled for now
|
|
|
|
//Value::String(expr) => Ok(Some(FacetCondition::from_str(txn, index, expr)?)),
|
|
|
|
Value::Array(arr) => parse_facets_array(txn, index, arr),
|
2021-03-16 01:11:10 +08:00
|
|
|
v => bail!("Invalid facet expression, expected Array, found: {:?}", v),
|
2021-03-04 18:56:32 +08:00
|
|
|
}
|
|
|
|
}
|
2021-04-20 01:03:53 +08:00
|
|
|
|
|
|
|
#[cfg(test)]
|
|
|
|
mod test {
|
|
|
|
use std::iter::FromIterator;
|
|
|
|
use serde_json::json;
|
|
|
|
|
|
|
|
use super::*;
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn no_formatted() {
|
|
|
|
let stop_words = fst::Set::default();
|
|
|
|
let highlighter = Highlighter::new(
|
|
|
|
&stop_words,
|
|
|
|
(String::from("<mark>"), String::from("</mark>")),
|
|
|
|
);
|
|
|
|
|
|
|
|
let mut fields = FieldsIdsMap::new();
|
|
|
|
let id = fields.insert("test").unwrap();
|
|
|
|
|
|
|
|
let mut buf = Vec::new();
|
|
|
|
let mut obkv = obkv::KvWriter::new(&mut buf);
|
|
|
|
obkv.insert(id, Value::String("hello".into()).to_string().as_bytes()).unwrap();
|
|
|
|
obkv.finish().unwrap();
|
|
|
|
|
|
|
|
let obkv = obkv::KvReader::new(&buf);
|
|
|
|
|
|
|
|
let all_formatted = Vec::new();
|
|
|
|
let to_highlight_ids = HashSet::new();
|
|
|
|
|
|
|
|
let matching_words = MatchingWords::default();
|
|
|
|
|
|
|
|
let value = compute_formatted(
|
|
|
|
&fields,
|
|
|
|
obkv,
|
|
|
|
&highlighter,
|
|
|
|
&matching_words,
|
|
|
|
&all_formatted,
|
|
|
|
&to_highlight_ids
|
|
|
|
).unwrap();
|
|
|
|
|
|
|
|
assert!(value.is_empty());
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn formatted_no_highlight() {
|
|
|
|
let stop_words = fst::Set::default();
|
|
|
|
let highlighter = Highlighter::new(
|
|
|
|
&stop_words,
|
|
|
|
(String::from("<mark>"), String::from("</mark>")),
|
|
|
|
);
|
|
|
|
|
|
|
|
let mut fields = FieldsIdsMap::new();
|
|
|
|
let id = fields.insert("test").unwrap();
|
|
|
|
|
|
|
|
let mut buf = Vec::new();
|
|
|
|
let mut obkv = obkv::KvWriter::new(&mut buf);
|
|
|
|
obkv.insert(id, Value::String("hello".into()).to_string().as_bytes()).unwrap();
|
|
|
|
obkv.finish().unwrap();
|
|
|
|
|
|
|
|
let obkv = obkv::KvReader::new(&buf);
|
|
|
|
|
|
|
|
let all_formatted = vec![id];
|
|
|
|
let to_highlight_ids = HashSet::new();
|
|
|
|
|
|
|
|
let matching_words = MatchingWords::default();
|
|
|
|
|
|
|
|
let value = compute_formatted(
|
|
|
|
&fields,
|
|
|
|
obkv,
|
|
|
|
&highlighter,
|
|
|
|
&matching_words,
|
|
|
|
&all_formatted,
|
|
|
|
&to_highlight_ids
|
|
|
|
).unwrap();
|
|
|
|
|
|
|
|
assert_eq!(Value::Object(value), json!({"test": "hello"}));
|
|
|
|
}
|
|
|
|
|
|
|
|
#[test]
|
|
|
|
fn formatted_with_highlight() {
|
|
|
|
let stop_words = fst::Set::default();
|
|
|
|
let highlighter = Highlighter::new(
|
|
|
|
&stop_words,
|
|
|
|
(String::from("<mark>"), String::from("</mark>")),
|
|
|
|
);
|
|
|
|
|
|
|
|
let mut fields = FieldsIdsMap::new();
|
|
|
|
let id = fields.insert("test").unwrap();
|
|
|
|
|
|
|
|
let mut buf = Vec::new();
|
|
|
|
let mut obkv = obkv::KvWriter::new(&mut buf);
|
|
|
|
obkv.insert(id, Value::String("hello".into()).to_string().as_bytes()).unwrap();
|
|
|
|
obkv.finish().unwrap();
|
|
|
|
|
|
|
|
let obkv = obkv::KvReader::new(&buf);
|
|
|
|
|
|
|
|
let all_formatted = vec![id];
|
|
|
|
let to_highlight_ids = HashSet::from_iter(Some(id));
|
|
|
|
|
|
|
|
let matching_words = HashSet::from_iter(Some(String::from("hello")));
|
|
|
|
|
|
|
|
let value = compute_formatted(
|
|
|
|
&fields,
|
|
|
|
obkv,
|
|
|
|
&highlighter,
|
|
|
|
&matching_words,
|
|
|
|
&all_formatted,
|
|
|
|
&to_highlight_ids
|
|
|
|
).unwrap();
|
|
|
|
|
|
|
|
assert_eq!(Value::Object(value), json!({"test": "<mark>hello</mark>"}));
|
|
|
|
}
|
|
|
|
}
|