feat: Introduce the Criterion trait

examples/serve-console.rs

@@ -3,11 +3,10 @@ use std::io::{self, Write};
 use structopt::StructOpt;
 use std::path::PathBuf;
 
-use fst::Streamer;
 use elapsed::measure_time;
 use rocksdb::{DB, DBOptions, IngestExternalFileOptions};
 use raptor::{automaton, Metadata, CommonWords};
-use raptor::rank;
+use raptor::rank::{criterion, RankedStreamBuilder};
 
 #[derive(Debug, StructOpt)]
 pub struct CommandConsole {
@@ -70,15 +69,13 @@ fn search(metadata: &Metadata, database: &DB, common_words: &CommonWords, query:
         automatons.push(lev);
     }
 
-    let config = rank::Config {
-        criteria: rank::criterion::default(),
-        metadata: &metadata,
-        automatons: automatons,
-        limit: 20,
-    };
+    let mut builder = RankedStreamBuilder::new(metadata, automatons);
+    builder.criteria(criterion::default());
 
-    let mut stream = rank::RankedStream::new(config);
-    while let Some(document) = stream.next() {
+    let mut stream = builder.build();
+    let documents = stream.retrieve_documents(20);
+
+    for document in documents {
         let id_key = format!("{}-id", document.id);
         let id = database.get(id_key.as_bytes()).unwrap().unwrap();
         let id = unsafe { from_utf8_unchecked(&id) };

examples/serve-http.rs

@@ -7,10 +7,9 @@ use std::path::PathBuf;
 use std::error::Error;
 use std::sync::Arc;
 
-use raptor::rank;
+use raptor::rank::{criterion, RankedStreamBuilder};
 use raptor::{automaton, Metadata, CommonWords};
 use rocksdb::{DB, DBOptions, IngestExternalFileOptions};
-use fst::Streamer;
 use warp::Filter;
 
 use structopt::StructOpt;
@@ -100,19 +99,17 @@ where M: AsRef<Metadata>,
         automatons.push(lev);
     }
 
-    let config = rank::Config {
-        criteria: rank::criterion::default(),
-        metadata: metadata.as_ref(),
-        automatons: automatons,
-        limit: 20,
-    };
+    let mut builder = RankedStreamBuilder::new(metadata.as_ref(), automatons);
+    builder.criteria(criterion::default());
+
+    let mut stream = builder.build();
+    let documents = stream.retrieve_documents(20);
 
-    let mut stream = rank::RankedStream::new(config);
     let mut body = Vec::new();
     write!(&mut body, "[")?;
 
     let mut first = true;
-    while let Some(document) = stream.next() {
+    for document in documents {
         let title_key = format!("{}-title", document.id);
         let title = database.as_ref().get(title_key.as_bytes()).unwrap().unwrap();
         let title = unsafe { from_utf8_unchecked(&title) };

src/rank/criterion/exact.rs

@@ -2,6 +2,7 @@ use std::cmp::Ordering;
 use group_by::GroupBy;
 use crate::Match;
 use crate::rank::{match_query_index, Document};
+use crate::rank::criterion::Criterion;
 
 #[inline]
 fn contains_exact(matches: &[Match]) -> bool {
@@ -13,10 +14,14 @@ fn number_exact_matches(matches: &[Match]) -> usize {
     GroupBy::new(matches, match_query_index).map(contains_exact).count()
 }
 
-#[inline]
-pub fn exact(lhs: &Document, rhs: &Document) -> Ordering {
+#[derive(Debug, Clone, Copy)]
+pub struct Exact;
+
+impl Criterion for Exact {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
         let lhs = number_exact_matches(&lhs.matches);
         let rhs = number_exact_matches(&rhs.matches);
 
         lhs.cmp(&rhs).reverse()
+    }
 }

src/rank/criterion/mod.rs

@@ -7,65 +7,64 @@ mod exact;
 
 use std::vec;
 use std::cmp::Ordering;
+use std::ops::Deref;
 use crate::rank::Document;
 
 pub use self::{
-    sum_of_typos::sum_of_typos,
-    number_of_words::number_of_words,
-    words_proximity::words_proximity,
-    sum_of_words_attribute::sum_of_words_attribute,
-    sum_of_words_position::sum_of_words_position,
-    exact::exact,
+    sum_of_typos::SumOfTypos,
+    number_of_words::NumberOfWords,
+    words_proximity::WordsProximity,
+    sum_of_words_attribute::SumOfWordsAttribute,
+    sum_of_words_position::SumOfWordsPosition,
+    exact::Exact,
 };
 
-#[inline]
-pub fn document_id(lhs: &Document, rhs: &Document) -> Ordering {
+pub trait Criterion {
+    #[inline]
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering;
+
+    #[inline]
+    fn eq(&self, lhs: &Document, rhs: &Document) -> bool {
+        self.evaluate(lhs, rhs) == Ordering::Equal
+    }
+}
+
+impl<'a, T: Criterion + ?Sized> Criterion for &'a T {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
+        self.deref().evaluate(lhs, rhs)
+    }
+
+    fn eq(&self, lhs: &Document, rhs: &Document) -> bool {
+        self.deref().eq(lhs, rhs)
+    }
+}
+
+impl<T: Criterion + ?Sized> Criterion for Box<T> {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
+        self.deref().evaluate(lhs, rhs)
+    }
+
+    fn eq(&self, lhs: &Document, rhs: &Document) -> bool {
+        self.deref().eq(lhs, rhs)
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+pub struct DocumentId;
+
+impl Criterion for DocumentId {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
         lhs.id.cmp(&rhs.id)
-}
-
-#[derive(Debug)]
-pub struct Criteria<F>(Vec<F>);
-
-impl<F> Criteria<F> {
-    pub fn new() -> Self {
-        Criteria(Vec::new())
-    }
-
-    pub fn with_capacity(cap: usize) -> Self {
-        Criteria(Vec::with_capacity(cap))
-    }
-
-    pub fn push(&mut self, criterion: F) {
-        self.0.push(criterion)
-    }
-
-    pub fn add(mut self, criterion: F) -> Self {
-        self.push(criterion);
-        self
     }
 }
 
-impl<F> IntoIterator for Criteria<F> {
-    type Item = F;
-    type IntoIter = vec::IntoIter<Self::Item>;
-
-    fn into_iter(self) -> Self::IntoIter {
-        self.0.into_iter()
-    }
-}
-
-pub fn default() -> Criteria<impl Fn(&Document, &Document) -> Ordering + Copy> {
-    let functions = &[
-        sum_of_typos,
-        number_of_words,
-        words_proximity,
-        sum_of_words_attribute,
-        sum_of_words_position,
-        exact,
-        document_id,
-    ];
-
-    let mut criteria = Criteria::with_capacity(functions.len());
-    for f in functions { criteria.push(f) }
-    criteria
+pub fn default() -> Vec<Box<dyn Criterion>> {
+    vec![
+        Box::new(SumOfTypos),
+        Box::new(NumberOfWords),
+        Box::new(WordsProximity),
+        Box::new(SumOfWordsAttribute),
+        Box::new(SumOfWordsPosition),
+        Box::new(Exact),
+    ]
 }

src/rank/criterion/number_of_words.rs

@@ -2,16 +2,21 @@ use std::cmp::Ordering;
 use group_by::GroupBy;
 use crate::Match;
 use crate::rank::{match_query_index, Document};
+use crate::rank::criterion::Criterion;
 
 #[inline]
 fn number_of_query_words(matches: &[Match]) -> usize {
     GroupBy::new(matches, match_query_index).count()
 }
 
-#[inline]
-pub fn number_of_words(lhs: &Document, rhs: &Document) -> Ordering {
+#[derive(Debug, Clone, Copy)]
+pub struct NumberOfWords;
+
+impl Criterion for NumberOfWords {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
         let lhs = number_of_query_words(&lhs.matches);
         let rhs = number_of_query_words(&rhs.matches);
 
         lhs.cmp(&rhs).reverse()
+    }
 }

src/rank/criterion/sum_of_typos.rs

@@ -2,6 +2,7 @@ use std::cmp::Ordering;
 use group_by::GroupBy;
 use crate::Match;
 use crate::rank::{match_query_index, Document};
+use crate::rank::criterion::Criterion;
 
 #[inline]
 fn sum_matches_typos(matches: &[Match]) -> i8 {
@@ -18,14 +19,19 @@ fn sum_matches_typos(matches: &[Match]) -> i8 {
     sum_typos - number_words
 }
 
-#[inline]
-pub fn sum_of_typos(lhs: &Document, rhs: &Document) -> Ordering {
+#[derive(Debug, Clone, Copy)]
+pub struct SumOfTypos;
+
+impl Criterion for SumOfTypos {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
         let lhs = sum_matches_typos(&lhs.matches);
         let rhs = sum_matches_typos(&rhs.matches);
 
         lhs.cmp(&rhs)
+    }
 }
 
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -42,7 +48,7 @@ mod tests {
                 Match { query_index: 1, distance: 0, attribute: 0, attribute_index: 2, is_exact: false },
             ];
             Document {
-                document_id: 0,
+                id: 0,
                 matches: matches,
             }
         };
@@ -53,12 +59,12 @@ mod tests {
                 Match { query_index: 1, distance: 0, attribute: 0, attribute_index: 2, is_exact: false },
             ];
             Document {
-                document_id: 1,
+                id: 1,
                 matches: matches,
             }
         };
 
-        assert_eq!(sum_of_typos(&doc0, &doc1), Ordering::Less);
+        assert_eq!(SumOfTypos.evaluate(&doc0, &doc1), Ordering::Less);
     }
 
     // typing: "bouton manchette"
@@ -73,7 +79,7 @@ mod tests {
                 Match { query_index: 1, distance: 0, attribute: 0, attribute_index: 1, is_exact: false },
             ];
             Document {
-                document_id: 0,
+                id: 0,
                 matches: matches,
             }
         };
@@ -83,12 +89,12 @@ mod tests {
                 Match { query_index: 0, distance: 0, attribute: 0, attribute_index: 0, is_exact: false },
             ];
             Document {
-                document_id: 1,
+                id: 1,
                 matches: matches,
             }
         };
 
-        assert_eq!(sum_of_typos(&doc0, &doc1), Ordering::Less);
+        assert_eq!(SumOfTypos.evaluate(&doc0, &doc1), Ordering::Less);
     }
 
     // typing: "bouton manchztte"
@@ -103,7 +109,7 @@ mod tests {
                 Match { query_index: 1, distance: 1, attribute: 0, attribute_index: 1, is_exact: false },
             ];
             Document {
-                document_id: 0,
+                id: 0,
                 matches: matches,
             }
         };
@@ -113,11 +119,11 @@ mod tests {
                 Match { query_index: 0, distance: 0, attribute: 0, attribute_index: 0, is_exact: false },
             ];
             Document {
-                document_id: 1,
+                id: 1,
                 matches: matches,
             }
         };
 
-        assert_eq!(sum_of_typos(&doc0, &doc1), Ordering::Equal);
+        assert_eq!(SumOfTypos.evaluate(&doc0, &doc1), Ordering::Equal);
     }
 }

src/rank/criterion/sum_of_words_attribute.rs

@@ -2,6 +2,7 @@ use std::cmp::Ordering;
 use group_by::GroupBy;
 use crate::Match;
 use crate::rank::{match_query_index, Document};
+use crate::rank::criterion::Criterion;
 
 #[inline]
 fn sum_matches_attributes(matches: &[Match]) -> u8 {
@@ -12,10 +13,14 @@ fn sum_matches_attributes(matches: &[Match]) -> u8 {
     }).sum()
 }
 
-#[inline]
-pub fn sum_of_words_attribute(lhs: &Document, rhs: &Document) -> Ordering {
+#[derive(Debug, Clone, Copy)]
+pub struct SumOfWordsAttribute;
+
+impl Criterion for SumOfWordsAttribute {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
         let lhs = sum_matches_attributes(&lhs.matches);
         let rhs = sum_matches_attributes(&rhs.matches);
 
         lhs.cmp(&rhs)
+    }
 }

src/rank/criterion/sum_of_words_position.rs

@@ -2,6 +2,7 @@ use std::cmp::Ordering;
 use group_by::GroupBy;
 use crate::Match;
 use crate::rank::{match_query_index, Document};
+use crate::rank::criterion::Criterion;
 
 #[inline]
 fn sum_matches_attribute_index(matches: &[Match]) -> u32 {
@@ -12,10 +13,14 @@ fn sum_matches_attribute_index(matches: &[Match]) -> u32 {
     }).sum()
 }
 
-#[inline]
-pub fn sum_of_words_position(lhs: &Document, rhs: &Document) -> Ordering {
+#[derive(Debug, Clone, Copy)]
+pub struct SumOfWordsPosition;
+
+impl Criterion for SumOfWordsPosition {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
         let lhs = sum_matches_attribute_index(&lhs.matches);
         let rhs = sum_matches_attribute_index(&rhs.matches);
 
         lhs.cmp(&rhs)
+    }
 }

src/rank/criterion/words_proximity.rs

@@ -2,6 +2,7 @@ use std::cmp::{self, Ordering};
 use group_by::GroupBy;
 use crate::Match;
 use crate::rank::{match_query_index, Document};
+use crate::rank::criterion::Criterion;
 
 const MAX_DISTANCE: u32 = 8;
 
@@ -42,10 +43,19 @@ fn matches_proximity(matches: &[Match]) -> u32 {
     proximity
 }
 
-pub fn words_proximity(lhs: &Document, rhs: &Document) -> Ordering {
-    matches_proximity(&lhs.matches).cmp(&matches_proximity(&rhs.matches))
+#[derive(Debug, Clone, Copy)]
+pub struct WordsProximity;
+
+impl Criterion for WordsProximity {
+    fn evaluate(&self, lhs: &Document, rhs: &Document) -> Ordering {
+        let lhs = matches_proximity(&lhs.matches);
+        let rhs = matches_proximity(&rhs.matches);
+
+        lhs.cmp(&rhs)
+    }
 }
 
+
 #[cfg(test)]
 mod tests {
     use super::*;

src/rank/mod.rs

@@ -3,7 +3,7 @@ mod ranked_stream;
 
 use crate::{Match, DocumentId};
 
-pub use self::ranked_stream::{RankedStream, Config};
+pub use self::ranked_stream::{RankedStreamBuilder, RankedStream};
 
 #[inline]
 fn match_query_index(a: &Match, b: &Match) -> bool {
@@ -18,10 +18,10 @@ pub struct Document {
 
 impl Document {
     pub fn new(doc: DocumentId, match_: Match) -> Self {
-        Self::from_sorted_matches(doc, vec![match_])
+        unsafe { Self::from_sorted_matches(doc, vec![match_]) }
     }
 
-    pub fn from_sorted_matches(id: DocumentId, matches: Vec<Match>) -> Self {
+    pub unsafe fn from_sorted_matches(id: DocumentId, matches: Vec<Match>) -> Self {
         Self { id, matches }
     }
 }

src/rank/ranked_stream.rs

@@ -1,4 +1,3 @@
-use std::cmp::Ordering;
 use std::rc::Rc;
 use std::{mem, vec};
 
@@ -8,76 +7,60 @@ use group_by::GroupByMut;
 
 use crate::automaton::{DfaExt, AutomatonExt};
 use crate::metadata::Metadata;
-use crate::metadata::ops::{OpBuilder, Union};
-use crate::rank::criterion::Criteria;
+use crate::metadata::ops::OpBuilder;
+use crate::rank::criterion::Criterion;
 use crate::rank::Document;
-use crate::{Match, DocumentId};
+use crate::Match;
 
-pub struct Config<'m, F> {
-    pub criteria: Criteria<F>,
-    pub metadata: &'m Metadata,
-    pub automatons: Vec<DfaExt>,
-    pub limit: usize,
-}
-
-pub struct RankedStream<'m, F>(RankedStreamInner<'m, F>);
-
-impl<'m, F> RankedStream<'m, F> {
-    pub fn new(config: Config<'m, F>) -> Self {
-        let automatons: Vec<_> = config.automatons.into_iter().map(Rc::new).collect();
-        let mut builder = OpBuilder::with_automatons(automatons.clone());
-        builder.push(config.metadata);
-
-        let inner = RankedStreamInner::Fed {
-            inner: builder.union(),
-            automatons: automatons,
-            criteria: config.criteria,
-            limit: config.limit,
-            matches: FnvHashMap::default(),
-        };
-
-        RankedStream(inner)
-    }
-}
-
-impl<'m, 'a, F> fst::Streamer<'a> for RankedStream<'m, F>
-where F: Fn(&Document, &Document) -> Ordering + Copy,
-{
-    type Item = Document;
-
-    fn next(&'a mut self) -> Option<Self::Item> {
-        self.0.next()
-    }
-}
-
-enum RankedStreamInner<'m, F> {
-    Fed {
-        inner: Union<'m>,
+#[derive(Clone)]
+pub struct RankedStreamBuilder<'m, C> {
+    metadata: &'m Metadata,
     automatons: Vec<Rc<DfaExt>>,
-        criteria: Criteria<F>,
-        limit: usize,
-        matches: FnvHashMap<DocumentId, Vec<Match>>,
-    },
-    Pours {
-        inner: vec::IntoIter<Document>,
-    },
+    criteria: Vec<C>,
 }
 
-impl<'m, 'a, F> fst::Streamer<'a> for RankedStreamInner<'m, F>
-where F: Fn(&Document, &Document) -> Ordering + Copy,
-{
-    type Item = Document;
+impl<'m, C> RankedStreamBuilder<'m, C> {
+    pub fn new(metadata: &'m Metadata, automatons: Vec<DfaExt>) -> Self {
+        RankedStreamBuilder {
+            metadata: metadata,
+            automatons: automatons.into_iter().map(Rc::new).collect(),
+            criteria: Vec::new(), // hummm...  prefer the criterion::default() ones !
+        }
+    }
 
-    fn next(&'a mut self) -> Option<Self::Item> {
-        loop {
-            match self {
-                RankedStreamInner::Fed { inner, automatons, criteria, limit, matches } => {
-                    match inner.next() {
-                        Some((string, indexed_values)) => {
+    pub fn criteria(&mut self, criteria: Vec<C>) {
+        self.criteria = criteria;
+    }
+
+    pub fn build(&self) -> RankedStream<C> {
+        let mut builder = OpBuilder::with_automatons(self.automatons.clone());
+        builder.push(self.metadata);
+
+        RankedStream {
+            stream: builder.union(),
+            automatons: &self.automatons,
+            criteria: &self.criteria,
+        }
+    }
+}
+
+pub struct RankedStream<'a, 'm, C> {
+    stream: crate::metadata::ops::Union<'m>,
+    automatons: &'a [Rc<DfaExt>],
+    criteria: &'a [C],
+}
+
+impl<'a, 'm, C> RankedStream<'a, 'm, C> {
+    pub fn retrieve_documents(&mut self, limit: usize) -> Vec<Document>
+    where C: Criterion
+    {
+        let mut matches = FnvHashMap::default();
+
+        while let Some((string, indexed_values)) = self.stream.next() {
             for iv in indexed_values {
-                                let automaton = &automatons[iv.index];
+                let automaton = &self.automatons[iv.index];
                 let distance = automaton.eval(string).to_u8();
-                                let same_length = string.len() == automaton.query_len();
+                let is_exact = distance == 0 && string.len() == automaton.query_len();
 
                 for di in iv.doc_indexes.as_slice() {
                     let match_ = Match {
@@ -85,50 +68,28 @@ where F: Fn(&Document, &Document) -> Ordering + Copy,
                         distance: distance,
                         attribute: di.attribute,
                         attribute_index: di.attribute_index,
-                                        is_exact: distance == 0 && same_length,
+                        is_exact: is_exact,
                     };
-                                    matches.entry(di.document)
-                                           .or_insert_with(Vec::new)
-                                           .push(match_);
-                                }
-                            }
-                        },
-                        None => {
-                            let matches = mem::replace(matches, FnvHashMap::default());
-                            let criteria = mem::replace(criteria, Criteria::new());
-                            *self = RankedStreamInner::Pours {
-                                inner: matches_into_iter(matches, criteria, *limit).into_iter()
-                            };
-                        },
-                    }
-                },
-                RankedStreamInner::Pours { inner } => {
-                    return inner.next()
-                },
+                    matches.entry(di.document).or_insert_with(Vec::new).push(match_);
                 }
             }
         }
-}
 
-fn matches_into_iter<F>(matches: FnvHashMap<DocumentId, Vec<Match>>,
-                        criteria: Criteria<F>,
-                        limit: usize) -> vec::IntoIter<Document>
-where F: Fn(&Document, &Document) -> Ordering + Copy,
-{
+        // collect matches from an HashMap into a Vec
         let mut documents: Vec<_> = matches.into_iter().map(|(id, mut matches)| {
             matches.sort_unstable();
-        Document::from_sorted_matches(id, matches)
+            unsafe { Document::from_sorted_matches(id, matches) }
         }).collect();
 
         let mut groups = vec![documents.as_mut_slice()];
 
-    for sort in criteria {
+        for criterion in self.criteria {
             let temp = mem::replace(&mut groups, Vec::new());
             let mut computed = 0;
 
             'grp: for group in temp {
-            group.sort_unstable_by(sort);
-            for group in GroupByMut::new(group, |a, b| sort(a, b) == Ordering::Equal) {
+                group.sort_unstable_by(|a, b| criterion.evaluate(a, b));
+                for group in GroupByMut::new(group, |a, b| criterion.eq(a, b)) {
                     computed += group.len();
                     groups.push(group);
                     if computed >= limit { break 'grp }
@@ -137,5 +98,6 @@ where F: Fn(&Document, &Document) -> Ordering + Copy,
         }
 
         documents.truncate(limit);
-    documents.into_iter()
+        documents
+    }
 }