package com.fasterxml.jackson.databind.util;
Base class for specialized primitive array builders.
/**
* Base class for specialized primitive array builders.
*/
public abstract class PrimitiveArrayBuilder<T>
{
Let's start with small chunks; typical usage is for small arrays anyway.
/**
* Let's start with small chunks; typical usage is for small arrays anyway.
*/
final static int INITIAL_CHUNK_SIZE = 12;
Also: let's expand by doubling up until 64k chunks (which is 16k entries for
32-bit machines)
/**
* Also: let's expand by doubling up until 64k chunks (which is 16k entries for
* 32-bit machines)
*/
final static int SMALL_CHUNK_SIZE = (1 << 14);
Let's limit maximum size of chunks we use; helps avoid excessive allocation
overhead for huge data sets.
For now, let's limit to quarter million entries, 1 meg chunks for 32-bit
machines.
/**
* Let's limit maximum size of chunks we use; helps avoid excessive allocation
* overhead for huge data sets.
* For now, let's limit to quarter million entries, 1 meg chunks for 32-bit
* machines.
*/
final static int MAX_CHUNK_SIZE = (1 << 18);
// // // Data storage
protected T _freeBuffer;
protected Node<T> _bufferHead;
protected Node<T> _bufferTail;
Number of total buffered entries in this buffer, counting all instances within linked list formed by following PrimitiveArrayBuilder<T>._bufferHead
. /**
* Number of total buffered entries in this buffer, counting all instances
* within linked list formed by following {@link #_bufferHead}.
*/
protected int _bufferedEntryCount;
// // // Recycled instances of sub-classes
// // // Life-cycle
protected PrimitiveArrayBuilder() { }
/*
/**********************************************************
/* Public API
/**********************************************************
*/
public int bufferedSize() { return _bufferedEntryCount; }
public T resetAndStart()
{
_reset();
return (_freeBuffer == null) ?
_constructArray(INITIAL_CHUNK_SIZE) : _freeBuffer;
}
Returns: Length of the next chunk to allocate
/**
* @return Length of the next chunk to allocate
*/
public final T appendCompletedChunk(T fullChunk, int fullChunkLength)
{
Node<T> next = new Node<T>(fullChunk, fullChunkLength);
if (_bufferHead == null) { // first chunk
_bufferHead = _bufferTail = next;
} else { // have something already
_bufferTail.linkNext(next);
_bufferTail = next;
}
_bufferedEntryCount += fullChunkLength;
int nextLen = fullChunkLength; // start with last chunk size
// double the size for small chunks
if (nextLen < SMALL_CHUNK_SIZE) {
nextLen += nextLen;
} else { // but by +25% for larger (to limit overhead)
nextLen += (nextLen >> 2);
}
return _constructArray(nextLen);
}
public T completeAndClearBuffer(T lastChunk, int lastChunkEntries)
{
int totalSize = lastChunkEntries + _bufferedEntryCount;
T resultArray = _constructArray(totalSize);
int ptr = 0;
for (Node<T> n = _bufferHead; n != null; n = n.next()) {
ptr = n.copyData(resultArray, ptr);
}
System.arraycopy(lastChunk, 0, resultArray, ptr, lastChunkEntries);
ptr += lastChunkEntries;
// sanity check (could have failed earlier due to out-of-bounds, too)
if (ptr != totalSize) {
throw new IllegalStateException("Should have gotten "+totalSize+" entries, got "+ptr);
}
return resultArray;
}
/*
/**********************************************************
/* Abstract methods for sub-classes to implement
/**********************************************************
*/
protected abstract T _constructArray(int len);
/*
/**********************************************************
/* Internal methods
/**********************************************************
*/
protected void _reset()
{
// can we reuse the last (and thereby biggest) array for next time?
if (_bufferTail != null) {
_freeBuffer = _bufferTail.getData();
}
// either way, must discard current contents
_bufferHead = _bufferTail = null;
_bufferedEntryCount = 0;
}
/*
/**********************************************************
/* Helper classes
/**********************************************************
*/
For actual buffering beyond the current buffer, we can actually use shared class which only deals with opaque "untyped" chunks. This works because System.arraycopy
does not take type; hence we can implement some aspects of primitive data handling in generic fashion. /**
* For actual buffering beyond the current buffer, we can actually
* use shared class which only deals with opaque "untyped" chunks.
* This works because {@link java.lang.System#arraycopy} does not
* take type; hence we can implement some aspects of primitive data
* handling in generic fashion.
*/
final static class Node<T>
{
Data stored in this node.
/**
* Data stored in this node.
*/
final T _data;
Number entries in the (untyped) array. Offset is assumed to be 0.
/**
* Number entries in the (untyped) array. Offset is assumed to be 0.
*/
final int _dataLength;
Node<T> _next;
public Node(T data, int dataLen)
{
_data = data;
_dataLength = dataLen;
}
public T getData() { return _data; }
public int copyData(T dst, int ptr)
{
System.arraycopy(_data, 0, dst, ptr, _dataLength);
ptr += _dataLength;
return ptr;
}
public Node<T> next() { return _next; }
public void linkNext(Node<T> next)
{
if (_next != null) { // sanity check
throw new IllegalStateException();
}
_next = next;
}
}
}