add singleton implementation of JsonObject to optimize memory usage #2353
Conversation
b38e5be to
aa792ef
Compare
aa792ef to
2989877
Compare
|
@jnicoulaud-ledger If you're looking to enhance the performance of parsing and serialization with circe AST even more, I'd like to introduce you to the jsoniter-scala-circe project. This library can significantly reduce latency and boost throughput, especially when using the jsoniter-core API for parsing and serialization from/to byte arrays or java.nio.ByteBuffers. Additionally, it offers improved safety, particularly when handling BigDecimal numbers, as discussed here. Furthermore, the jsoniter-scala-benchmark module can be used to verify that there are no performance regressions in your pull requests. It includes real-world message samples from Google Maps, GitHub Actions, and Twitter APIs, ensuring that your changes maintain optimal performance. |
|
|
||
| fromLinkedHashMap(map) | ||
| val fieldsSeq = fields.iterator.toSeq | ||
| (fieldsSeq.size: @switch) match { |
There was a problem hiding this comment.
I think you should avoid converting the fields into a seq, just to use size, use sizeCompare or sizeIs on the Iterable and determine if its 0 or 1, or greater instead... this way you don't have to materialize the Seq, and sizeCompare is cheap in comparison.
| fromLinkedHashMap(map) | ||
| val fieldsSeq = fields.iterator.toSeq | ||
| (fieldsSeq.size: @switch) match { | ||
| case 0 => empty |
There was a problem hiding this comment.
empty is actually sub-optimial as well, maybe we could create a dedicated empty type as well instead of it being final val empty: JsonObject = new MapAndVectorJsonObject(Map.empty, Vector.empty)
|
|
||
| private[circe] final def fromMapAndVector(map: Map[String, Json], keys: Vector[String]): JsonObject = | ||
| new MapAndVectorJsonObject(map, keys) | ||
| (map.size: @switch) match { |
There was a problem hiding this comment.
Same critique on size above, use sizeCompare or sizeIs instead so a full traversal isn't used when it isn't needed.
|
|
||
| private[circe] final def fromLinkedHashMap(map: LinkedHashMap[String, Json]): JsonObject = | ||
| new LinkedHashMapJsonObject(map) | ||
| (map.size: @switch) match { |
This pull request optimizes memory usage in the edge case where there are many objects with only one field, default
LinkedHashMapcontainer has a lot of memory overhead in such cases.Notes: