jecsand838 commented on code in PR #8349:
URL: https://github.com/apache/arrow-rs/pull/8349#discussion_r2370460438
##########
arrow-avro/src/reader/record.rs:
##########
@@ -1518,19 +1104,340 @@ impl Decoder {
.map_err(|e| ArrowError::ParseError(e.to_string()))?;
Arc::new(vals)
}
- Self::Union(fields, type_ids, offsets, encodings, _, None) => {
- flush_union!(fields, type_ids, offsets, encodings)
- }
- Self::Union(fields, type_ids, offsets, encodings, _,
Some(union_resolution)) => {
- match &mut union_resolution.kind {
- UnionResolvedKind::Both { .. } |
UnionResolvedKind::FromSingle { .. } => {
- flush_union!(fields, type_ids, offsets, encodings)
- }
- UnionResolvedKind::ToSingle { target } =>
target.flush(nulls)?,
+ Self::Union(u) => u.flush(nulls)?,
+ })
+ }
+}
+
+#[derive(Debug)]
+struct DispatchLut {
+ to_reader: Box<[i16]>,
+ promotion: Box<[Promotion]>,
+}
+
+impl DispatchLut {
+ fn from_writer_to_reader(promotion_map: &[Option<(usize, Promotion)>]) ->
Self {
+ let mut to_reader = Vec::with_capacity(promotion_map.len());
+ let mut promotion = Vec::with_capacity(promotion_map.len());
+ for map in promotion_map {
+ match *map {
+ Some((idx, promo)) => {
+ debug_assert!(idx <= i16::MAX as usize);
+ to_reader.push(idx as i16);
Review Comment:
> Seems like, if we care to squeeze space we should squeeze it all the way
(i8)? If we don't care, then why squeeze at all?
That's a fair call-out. Based on how I see this crate being used in a
program (many `Decoder` instances each with potentially multiple
`RecordDecoder`s), I figured small memory wins like this would add up at the
`RecordDecoder` level.
> Just double checking -- UnionFields forbids negative type ids, so we
actually can use them as sentinel values?
In Arrow, union type IDs are 8‑bit signed integers and the implementation
expects them to be positive.
https://docs.rs/arrow/latest/arrow/array/struct.UnionArray.html#safety
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
