Optimization for objectbuffer about kryo HOT 1 CLOSED

From [email protected] on March 08, 2011 16:37:04

Kryo serializers write to a ByteBuffer directly. When the buffer is overrun, it throws an exception which unrolls the stack all the way back to the ObjectBuffer, where the buffer size is doubled. At this point, there is no way to resume serialization where the overrun occurred, so serialization has to start from the beginning. The only thing that will really help is to use a larger initial buffer size.

The solution to this is to have Kryo not use ByteBuffer directly, but to use a Kryo specific buffer class. This would allow the buffer to grow without copying and without interrupting serialization (by keeping a list of buffers and allocating new ones as needed). It would also allow more flexibility when dealing with streams. I've collected a number of ideas for a rewrite of Kryo, but don't currently have a timeline.

Status: WontFix

from kryo.