Issue/822 TopDocs sorted by i64, and date fastfield (in addition to u64) (#890)

* Unsatisfactory implementation.

The fastfield are hit. But for performance, we want the comparison to happen on u64,
and the conversion to the FastType to be done only on the selected TopK
elements.

For i64, the current approach might be ok.
For DateTime, it is most likely catastrophic.

Closes #822

* Decoupled SegmentCollector Fruit from Collector Fruit.

Deferred conversion from u64 to the proper FastField type to after the overall collection.
(tantivy guarantees that u64 encoding is consistent with the original
ordering of the fastfield)

Closes #882
This commit is contained in:
Paul Masurel
2020-09-30 17:51:11 +09:00
committed by GitHub
parent 838c476733
commit 674cae8ee2
7 changed files with 340 additions and 53 deletions

View File

@@ -117,6 +117,12 @@ impl<Guard> From<PoisonError<Guard>> for TantivyError {
}
}
impl From<chrono::ParseError> for TantivyError {
fn from(err: chrono::ParseError) -> TantivyError {
TantivyError::InvalidArgument(err.to_string())
}
}
impl From<schema::DocParsingError> for TantivyError {
fn from(error: schema::DocParsingError) -> TantivyError {
TantivyError::InvalidArgument(format!("Failed to parse document {:?}", error))