public final class CharsToNameCanonicalizer
extends java.lang.Object
For optimal performance, usage pattern should be one where matches
should be very common (especially after "warm-up"), and as with most hash-based
maps/sets, that hash codes are uniformly distributed. Also, collisions
are slightly more expensive than with HashMap or HashSet, since hash codes
are not used in resolving collisions; that is, equals() comparison is
done with all symbols in same bucket index.
Finally, rehashing is also more expensive, as hash codes are not
stored; rehashing requires all entries' hash codes to be recalculated.
Reason for not storing hash codes is reduced memory usage, hoping
for better memory locality.
Usual usage pattern is to create a single "master" instance, and either use that instance in sequential fashion, or to create derived "child" instances, which after use, are asked to return possible symbol additions to master instance. In either case benefit is that symbol table gets initialized so that further uses are more efficient, as eventually all symbols needed will already be in symbol table. At that point no more Symbol String allocations are needed, nor changes to symbol table itself.
Note that while individual SymbolTable instances are NOT thread-safe (much like generic collection classes), concurrently used "child" instances can be freely used without synchronization. However, using master table concurrently with child instances can only be done if access to master instance is read-only (i.e. no modifications done).
Modifier and Type | Class and Description |
---|---|
(package private) static class |
CharsToNameCanonicalizer.Bucket
This class is a symbol table entry.
|
Modifier and Type | Field and Description |
---|---|
protected CharsToNameCanonicalizer.Bucket[] |
_buckets
Overflow buckets; if primary doesn't match, lookup is done
from here.
|
protected boolean |
_canonicalize
Whether any canonicalization should be attempted (whether using
intern or not)
|
protected boolean |
_dirty
Flag that indicates if any changes have been made to the data;
used to both determine if bucket array needs to be copied when
(first) change is made, and potentially if updated bucket list
is to be resync'ed back to master instance.
|
protected int |
_flags |
private int |
_hashSeed
Seed value we use as the base to make hash codes non-static between
different runs, but still stable for lifetime of a single symbol table
instance.
|
protected int |
_indexMask
Mask used to get index from hash values; equal to
_buckets.length - 1 , when _buckets.length is
a power of two. |
protected int |
_longestCollisionList
We need to keep track of the longest collision list; this is needed
both to indicate problems with attacks and to allow flushing for
other cases.
|
protected java.util.BitSet |
_overflows
Lazily constructed structure that is used to keep track of
collision buckets that have overflowed once: this is used
to detect likely attempts at denial-of-service attacks that
uses hash collisions.
|
protected CharsToNameCanonicalizer |
_parent
Sharing of learnt symbols is done by optional linking of symbol
table instances with their parents.
|
protected int |
_size
Current size (number of entries); needed to know if and when
rehash.
|
protected int |
_sizeThreshold
Limit that indicates maximum size this instance can hold before
it needs to be expanded and rehashed.
|
protected java.lang.String[] |
_symbols
Primary matching symbols; it's expected most match occur from
here.
|
protected static int |
DEFAULT_T_SIZE
Default initial table size.
|
static int |
HASH_MULT |
(package private) static int |
MAX_COLL_CHAIN_LENGTH
Also: to thwart attacks based on hash collisions (which may or may not
be cheap to calculate), we will need to detect "too long"
collision chains.
|
(package private) static int |
MAX_ENTRIES_FOR_REUSE
Let's only share reasonably sized symbol tables.
|
protected static int |
MAX_T_SIZE
Let's not expand symbol tables past some maximum size;
this should protected against OOMEs caused by large documents
with unique (~= random) names.
|
(package private) static CharsToNameCanonicalizer |
sBootstrapSymbolTable |
Modifier | Constructor and Description |
---|---|
private |
CharsToNameCanonicalizer()
Main method for constructing a master symbol table instance.
|
private |
CharsToNameCanonicalizer(CharsToNameCanonicalizer parent,
int flags,
java.lang.String[] symbols,
CharsToNameCanonicalizer.Bucket[] buckets,
int size,
int hashSeed,
int longestColl)
Internal constructor used when creating child instances.
|
Modifier and Type | Method and Description |
---|---|
private java.lang.String |
_addSymbol(char[] buffer,
int start,
int len,
int h,
int index) |
private java.lang.String |
_findSymbol2(char[] buffer,
int start,
int len,
CharsToNameCanonicalizer.Bucket b) |
private void |
_handleSpillOverflow(int bindex,
CharsToNameCanonicalizer.Bucket newBucket) |
int |
_hashToIndex(int rawHash)
Helper method that takes in a "raw" hash value, shuffles it as necessary,
and truncates to be used as the index.
|
private static int |
_thresholdSize(int hashAreaSize) |
int |
bucketCount()
Method for checking number of primary hash buckets this symbol
table uses.
|
int |
calcHash(char[] buffer,
int start,
int len)
Implementation of a hashing method for variable length
Strings.
|
int |
calcHash(java.lang.String key) |
int |
collisionCount()
Method mostly needed by unit tests; calculates number of
entries that are in collision list.
|
private void |
copyArrays()
Method called when copy-on-write is needed; generally when first
change is made to a derived symbol table.
|
static CharsToNameCanonicalizer |
createRoot()
Method called to create root canonicalizer for a
JsonFactory
instance. |
protected static CharsToNameCanonicalizer |
createRoot(int hashSeed) |
java.lang.String |
findSymbol(char[] buffer,
int start,
int len,
int h) |
int |
hashSeed() |
private void |
initTables(int initialSize) |
CharsToNameCanonicalizer |
makeChild(int flags)
"Factory" method; will create a new child instance of this symbol
table.
|
private CharsToNameCanonicalizer |
makeOrphan(int seed) |
int |
maxCollisionLength()
Method mostly needed by unit tests; calculates length of the
longest collision chain.
|
boolean |
maybeDirty() |
private void |
mergeChild(CharsToNameCanonicalizer child)
Method that allows contents of child table to potentially be
"merged in" with contents of this symbol table.
|
private void |
rehash()
Method called when size (number of entries) of symbol table grows
so big that load factor is exceeded.
|
void |
release() |
protected void |
reportTooManyCollisions(int maxLen) |
int |
size() |
public static final int HASH_MULT
protected static final int DEFAULT_T_SIZE
protected static final int MAX_T_SIZE
static final int MAX_ENTRIES_FOR_REUSE
static final int MAX_COLL_CHAIN_LENGTH
Note: longest chain we have been able to produce without malicious intent has been 38 (with "com.fasterxml.jackson.core.main.TestWithTonsaSymbols"); our setting should be reasonable here.
Also note that value was lowered from 255 (2.3 and earlier) to 100 for 2.4
static final CharsToNameCanonicalizer sBootstrapSymbolTable
protected CharsToNameCanonicalizer _parent
release
),
parent's shared tables may be updated from the child instance.private final int _hashSeed
protected final int _flags
protected boolean _canonicalize
protected java.lang.String[] _symbols
protected CharsToNameCanonicalizer.Bucket[] _buckets
Note: Number of buckets is half of number of symbol entries, on assumption there's less need for buckets.
protected int _size
protected int _sizeThreshold
protected int _indexMask
_buckets.length - 1
, when _buckets.length is
a power of two.protected int _longestCollisionList
protected boolean _dirty
protected java.util.BitSet _overflows
private CharsToNameCanonicalizer()
private CharsToNameCanonicalizer(CharsToNameCanonicalizer parent, int flags, java.lang.String[] symbols, CharsToNameCanonicalizer.Bucket[] buckets, int size, int hashSeed, int longestColl)
public static CharsToNameCanonicalizer createRoot()
JsonFactory
instance. Root instance is never used directly; its main use is for
storing and sharing underlying symbol arrays as needed.protected static CharsToNameCanonicalizer createRoot(int hashSeed)
private void initTables(int initialSize)
private static int _thresholdSize(int hashAreaSize)
public CharsToNameCanonicalizer makeChild(int flags)
Note: while this method is synchronized, it is generally not safe to both use makeChild/mergeChild, AND to use instance actively. Instead, a separate 'root' instance should be used on which only makeChild/mergeChild are called, but instance itself is not used as a symbol table.
private CharsToNameCanonicalizer makeOrphan(int seed)
private void mergeChild(CharsToNameCanonicalizer child)
Note that caller has to make sure symbol table passed in is really a child or sibling of this symbol table.
public void release()
public int size()
public int bucketCount()
public boolean maybeDirty()
public int hashSeed()
public int collisionCount()
size()
- 1), but should usually be much lower, ideally 0.public int maxCollisionLength()
size()
- 1 in the pathological casepublic java.lang.String findSymbol(char[] buffer, int start, int len, int h)
private java.lang.String _findSymbol2(char[] buffer, int start, int len, CharsToNameCanonicalizer.Bucket b)
private java.lang.String _addSymbol(char[] buffer, int start, int len, int h, int index)
private void _handleSpillOverflow(int bindex, CharsToNameCanonicalizer.Bucket newBucket)
public int _hashToIndex(int rawHash)
public int calcHash(char[] buffer, int start, int len)
len
- Length of String; has to be at least 1 (caller guarantees
this pre-condition)public int calcHash(java.lang.String key)
private void copyArrays()
private void rehash()
protected void reportTooManyCollisions(int maxLen)