c21 commented on a change in pull request #29342:
URL: https://github.com/apache/spark/pull/29342#discussion_r470169165
##########
File path: core/src/main/java/org/apache/spark/unsafe/map/BytesToBytesMap.java
##########
@@ -428,6 +428,62 @@ public MapIterator destructiveIterator() {
return new MapIterator(numValues, new Location(), true);
}
+ /**
+ * Iterator for the entries of this map. This is to first iterate over key
index array
+ * `longArray` then accessing values in `dataPages`. NOTE: this is different
from `MapIterator`
+ * in the sense that key index is preserved here
+ * (See `UnsafeHashedRelation` for example of usage).
+ */
+ public final class MapIteratorWithKeyIndex implements Iterator<Location> {
+
+ private int keyIndex = 0;
+ private int numRecords;
+ private final Location loc;
+
+ private MapIteratorWithKeyIndex(int numRecords, Location loc) {
+ this.numRecords = numRecords;
+ this.loc = loc;
+ }
+
+ @Override
+ public boolean hasNext() {
+ return numRecords > 0;
+ }
+
+ @Override
+ public Location next() {
+ if (!loc.isDefined() || !loc.nextValue()) {
+ while (longArray.get(keyIndex * 2) == 0) {
+ keyIndex++;
+ }
+ loc.with(keyIndex, (int) longArray.get(keyIndex * 2 + 1), true);
+ keyIndex++;
Review comment:
> Basically keyIndex can grow beyond the longArray.size() if numRecords
is sufficiently big ?
@agrawaldevesh - no. If that happens, then there's a bug in
`BytesToBytesMap` and we need to fail loudly anyway. I intentionally avoid
bound checking for every key probing to avoid doing extra unnecessary work for
saving CPU. But if others also think we should add that, I can add bound check
too.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
