feat: Separate highlights from matches to make the code easier to follow

This commit is contained in:
Clément Renault 2019-07-15 19:34:53 +02:00
parent 9959f2e952
commit 89df496f0c
3 changed files with 44 additions and 28 deletions

View File

@ -13,7 +13,7 @@ pub mod criterion;
use serde::{Serialize, Deserialize}; use serde::{Serialize, Deserialize};
use zerocopy::{AsBytes, FromBytes}; use zerocopy::{AsBytes, FromBytes};
use self::raw_document::raw_documents_from_matches; use self::raw_document::raw_documents_from;
pub use self::query_builder::{QueryBuilder, DistinctQueryBuilder, normalize_str}; pub use self::query_builder::{QueryBuilder, DistinctQueryBuilder, normalize_str};
pub use self::raw_document::RawDocument; pub use self::raw_document::RawDocument;

View File

@ -17,7 +17,7 @@ use crate::automaton::{build_dfa, build_prefix_dfa};
use crate::criterion::Criteria; use crate::criterion::Criteria;
use crate::distinct_map::{DistinctMap, BufferedDistinctMap}; use crate::distinct_map::{DistinctMap, BufferedDistinctMap};
use crate::query_enhancer::{QueryEnhancerBuilder, QueryEnhancer}; use crate::query_enhancer::{QueryEnhancerBuilder, QueryEnhancer};
use crate::raw_documents_from_matches; use crate::raw_documents_from;
use crate::reordered_attrs::ReorderedAttrs; use crate::reordered_attrs::ReorderedAttrs;
use crate::{TmpMatch, Highlight, DocumentId, Store, RawDocument, Document}; use crate::{TmpMatch, Highlight, DocumentId, Store, RawDocument, Document};
@ -215,6 +215,7 @@ where S: Store,
}; };
let mut matches = Vec::new(); let mut matches = Vec::new();
let mut highlights = Vec::new();
while let Some((input, indexed_values)) = stream.next() { while let Some((input, indexed_values)) = stream.next() {
for iv in indexed_values { for iv in indexed_values {
@ -240,23 +241,21 @@ where S: Store,
is_exact, is_exact,
}; };
// TODO do not store in the same matches vec
let highlight = Highlight { let highlight = Highlight {
attribute: di.attribute, attribute: di.attribute,
char_index: di.char_index, char_index: di.char_index,
char_length: di.char_length, char_length: di.char_length,
}; };
matches.push((di.document_id, match_, highlight)); matches.push((di.document_id, match_));
highlights.push((di.document_id, highlight));
} }
} }
} }
} }
// we sort the matches to make them rewritable // we sort the matches to make them rewritable
matches.par_sort_unstable_by_key(|(id, match_, _)| { matches.par_sort_unstable_by_key(|(id, match_)| (*id, match_.attribute, match_.word_index));
(*id, match_.attribute, match_.word_index) // query_id ???
});
let mut padded_matches = Vec::with_capacity(matches.len()); let mut padded_matches = Vec::with_capacity(matches.len());
for same_document in matches.linear_group_by(|a, b| a.0 == b.0) { for same_document in matches.linear_group_by(|a, b| a.0 == b.0) {
@ -268,7 +267,7 @@ where S: Store,
while let Some(same_word_index) = iter.next() { while let Some(same_word_index) = iter.next() {
let mut biggest = 0; let mut biggest = 0;
for (id, match_, highlight) in same_word_index { for (id, match_) in same_word_index {
let mut replacement = query_enhancer.replacement(match_.query_index); let mut replacement = query_enhancer.replacement(match_.query_index);
let replacement_len = replacement.len() - 1; let replacement_len = replacement.len() - 1;
@ -280,7 +279,7 @@ where S: Store,
word_index: match_.word_index + padding as u16, word_index: match_.word_index + padding as u16,
..match_.clone() ..match_.clone()
}; };
padded_matches.push((*id, match_, *highlight)); padded_matches.push((*id, match_));
} }
let mut found = false; let mut found = false;
@ -296,7 +295,7 @@ where S: Store,
..match_.clone() ..match_.clone()
}; };
for (_, nmatch_, _) in next_group { for (_, nmatch_) in next_group {
let mut rep = query_enhancer.replacement(nmatch_.query_index); let mut rep = query_enhancer.replacement(nmatch_.query_index);
let query_index = rep.next().unwrap(); let query_index = rep.next().unwrap();
let nmatch_ = TmpMatch { query_index, ..nmatch_.clone() }; let nmatch_ = TmpMatch { query_index, ..nmatch_.clone() };
@ -311,12 +310,12 @@ where S: Store,
word_index: match_.word_index + padding as u16 + (i + 1) as u16, word_index: match_.word_index + padding as u16 + (i + 1) as u16,
..match_.clone() ..match_.clone()
}; };
padded_matches.push((*id, match_, *highlight)); padded_matches.push((*id, match_));
biggest = biggest.max(i + 1); biggest = biggest.max(i + 1);
} }
} }
padded_matches.push((*id, padmatch_, *highlight)); padded_matches.push((*id, padmatch_));
found = true; found = true;
continue 'padding; continue 'padding;
} }
@ -337,7 +336,7 @@ where S: Store,
word_index: match_.word_index + padding as u16 + (i + 1) as u16, word_index: match_.word_index + padding as u16 + (i + 1) as u16,
..match_.clone() ..match_.clone()
}; };
padded_matches.push((*id, match_, *highlight)); padded_matches.push((*id, match_));
} }
biggest = biggest.max(replacement_len); biggest = biggest.max(replacement_len);
@ -350,11 +349,19 @@ where S: Store,
} }
let total_matches = padded_matches.len();
padded_matches.par_sort_unstable();
let padded_matches = SetBuf::new_unchecked(padded_matches);
let raw_documents = raw_documents_from_matches(padded_matches); let matches = {
padded_matches.par_sort_unstable();
SetBuf::new_unchecked(padded_matches)
};
let highlights = {
highlights.par_sort_unstable_by_key(|(id, _)| *id);
SetBuf::new_unchecked(highlights)
};
let total_matches = matches.len();
let raw_documents = raw_documents_from(matches, highlights);
info!("{} total documents to classify", raw_documents.len()); info!("{} total documents to classify", raw_documents.len());
info!("{} total matches to classify", total_matches); info!("{} total matches to classify", total_matches);

View File

@ -66,25 +66,34 @@ impl fmt::Debug for RawDocument {
} }
} }
pub fn raw_documents_from_matches(matches: SetBuf<(DocumentId, TmpMatch, Highlight)>) -> Vec<RawDocument> { pub fn raw_documents_from(
matches: SetBuf<(DocumentId, TmpMatch)>,
highlights: SetBuf<(DocumentId, Highlight)>,
) -> Vec<RawDocument>
{
let mut docs_ranges: Vec<(_, Range, _)> = Vec::new(); let mut docs_ranges: Vec<(_, Range, _)> = Vec::new();
let mut matches2 = Matches::with_capacity(matches.len()); let mut matches2 = Matches::with_capacity(matches.len());
for group in matches.linear_group_by(|(a, _, _), (b, _, _)| a == b) { let matches = matches.linear_group_by(|(a, _), (b, _)| a == b);
let document_id = group[0].0; let highlights = highlights.linear_group_by(|(a, _), (b, _)| a == b);
let start = docs_ranges.last().map(|(_, r, _)| r.end).unwrap_or(0);
let end = start + group.len();
let highlights = group.iter().map(|(_, _, h)| *h).collect(); for (mgroup, hgroup) in matches.zip(highlights) {
debug_assert_eq!(mgroup[0].0, hgroup[0].0);
let document_id = mgroup[0].0;
let start = docs_ranges.last().map(|(_, r, _)| r.end).unwrap_or(0);
let end = start + mgroup.len();
let highlights = hgroup.iter().map(|(_, h)| *h).collect();
docs_ranges.push((document_id, Range { start, end }, highlights)); docs_ranges.push((document_id, Range { start, end }, highlights));
matches2.extend_from_slice(group); matches2.extend_from_slice(mgroup);
} }
let matches = Arc::new(matches2); let matches = Arc::new(matches2);
docs_ranges.into_iter().map(|(i, range, highlights)| { docs_ranges.into_iter().map(|(id, range, highlights)| {
let matches = SharedMatches { range, matches: matches.clone() }; let matches = SharedMatches { range, matches: matches.clone() };
RawDocument::new(i, matches, highlights) RawDocument::new(id, matches, highlights)
}).collect() }).collect()
} }
@ -120,8 +129,8 @@ impl Matches {
} }
} }
fn extend_from_slice(&mut self, matches: &[(DocumentId, TmpMatch, Highlight)]) { fn extend_from_slice(&mut self, matches: &[(DocumentId, TmpMatch)]) {
for (_, match_, _) in matches { for (_, match_) in matches {
self.query_index.push(match_.query_index); self.query_index.push(match_.query_index);
self.distance.push(match_.distance); self.distance.push(match_.distance);
self.attribute.push(match_.attribute); self.attribute.push(match_.attribute);