gcc/libjava/gnu/gcj/runtime/PersistentByteMap.java
Robin Green 832a0afe9c re PR libgcj/22189 (Table Full in gcj-dbtool if -m option used with smallest possible input)
2005-06-28  Robin Green  <greenrd@greenrd.org>

	PR java/22189
	* gnu/gcj/runtime/PersistentByteMap.java (init): Fix rounding
	error.

From-SVN: r101382
2005-06-28 14:31:22 +00:00

620 lines
16 KiB
Java

/* Copyright (C) 2004, 2005 Free Software Foundation
This file is part of libgcj.
This software is copyrighted work licensed under the terms of the
Libgcj License. Please consult the file "LIBGCJ_LICENSE" for
details. */
/* A PersistentByteMap maps a byte array to another byte array. It
uses a file that does not need to be serialized but may be
memory-mapped and read in-place. So, even if there are many instances
of gcj applications running, they can share PersistentByteMaps.
The idea is to make searches as fast as possible: opening a
PersistentByteMap is cheap and search time doesn't grow with the
number of entries in the table. On the other hand, enumerating the
map is slow, but that is a relatively uncommon operation.
The main use of this class is to provide a way to map the
MessageDigest of a class file to the location of a DSO that contains
the compiled version of that class. It is up the the installer of an
application to keep the DSO up to date with the jar.
USAGE:
MessageDigest md = MessageDigest.getInstance("MD5");
digest = md.digest(bytes);
PersistentByteMap map
= new PersistentByteMap
(fileName, PersistentByteMap.AccessMode.READ_ONLY);
byte[] soName = map.get(digest);
if (soName)
{
String SharedLibraryName = new String(soName);
BUGS/FEATURES:
remove() isn't written yet.
capacity is fixed once the map has been created.
We use linear probing to resolve collisions. It might be
better to use a scheme that results in fewer probes to
determine that an item isn't found. However, even when the
table is half full there are only on average 1.5 probes for a
successful search and 2.5 probes for an unsuccessful one.
We don't do any locking at all: adding to a PersistentByteMap
at runtime is possible, but it requires filesystem locks
around get(), put(), and remove().
*/
package gnu.gcj.runtime;
import java.io.*;
import java.nio.*;
import java.nio.channels.*;
import java.util.*;
import java.security.MessageDigest;
import java.math.BigInteger;
public class PersistentByteMap
{
private MappedByteBuffer buf;
static private final int MAGIC = 0;
static private final int VERSION = 4;
static private final int CAPACITY = 8;
static private final int TABLE_BASE = 12;
static private final int STRING_BASE = 16;
static private final int STRING_SIZE = 20;
static private final int FILE_SIZE = 24;
static private final int ELEMENTS = 28;
static private final int INT_SIZE = 4;
static private final int TABLE_ENTRY_SIZE = 2 * INT_SIZE;
private int capacity; // number of entries
private int table_base; // offset from start of file, in bytes
private int string_base; // offset from start of file, in bytes
private int string_size; // size of string table, in bytes
private int file_size; // size of file, in bytes;
private int elements; // number of elements in table
private long length; // the length of the underlying file
private final File name; // The name of the underlying file
static private final int UNUSED_ENTRY = -1;
static public final int KEYS = 0;
static public final int VALUES = 1;
static public final int ENTRIES = 2;
private HashMap values; // A map of strings in the string table.
FileChannel fc; // The underlying file channel.
static final public class AccessMode
{
private final FileChannel.MapMode mapMode;
static
{
READ_ONLY = new AccessMode(FileChannel.MapMode.READ_ONLY);
READ_WRITE = new AccessMode(FileChannel.MapMode.READ_WRITE);
PRIVATE = new AccessMode(FileChannel.MapMode.PRIVATE);
}
public static final AccessMode READ_ONLY;
public static final AccessMode READ_WRITE;
public static final AccessMode PRIVATE;
private AccessMode(FileChannel.MapMode mode)
{
this.mapMode = mode;
}
}
private PersistentByteMap(File name)
{
this.name = name;
}
public PersistentByteMap(String filename, AccessMode mode)
throws IOException
{
this(new File(filename), mode);
}
public PersistentByteMap(File f, AccessMode mode)
throws IOException
{
name = f;
if (mode == AccessMode.READ_ONLY)
{
FileInputStream fis = new FileInputStream(f);
fc = fis.getChannel();
}
else
{
RandomAccessFile fos = new RandomAccessFile(f, "rw");
fc = fos.getChannel();
}
length = fc.size();
buf = fc.map(mode.mapMode, 0, length);
int magic = getWord (MAGIC);
if (magic != 0x67636a64) /* "gcjd" */
throw new IllegalArgumentException(f.getName());
table_base = getWord (TABLE_BASE);
capacity = getWord (CAPACITY);
string_base = getWord (STRING_BASE);
string_size = getWord (STRING_SIZE);
file_size = getWord (FILE_SIZE);
elements = getWord (ELEMENTS);
// FIXME: Insert a bunch of sanity checks here
}
private void init (PersistentByteMap m, File f, int capacity, int strtabSize)
throws IOException
{
f.createNewFile();
RandomAccessFile raf = new RandomAccessFile(f, "rw");
{
// The user has explicitly provided a size for the table.
// We're going to make that size prime. This isn't
// strictly necessary but it can't hurt.
//
// We expand the size by 3/2 and round the result because the
// hash table is intolerably slow when more than 2/3 full.
BigInteger size = new BigInteger(Integer.toString(((capacity*3)+1)/2));
BigInteger two = BigInteger.ONE.add(BigInteger.ONE);
if (size.getLowestSetBit() != 0) // A hard way to say isEven()
size = size.add(BigInteger.ONE);
while (! size.isProbablePrime(10))
size = size.add(two);
this.capacity = capacity = size.intValue();
}
table_base = 64;
string_base = table_base + capacity * TABLE_ENTRY_SIZE;
string_size = 0;
file_size = string_base;
elements = 0;
int totalFileSize = string_base + strtabSize;
// Create the file; this rounds up the size of the file to a fixed
// number of 4k pages.
byte[] _4k = new byte[4096];
for (long i = 0; i < totalFileSize; i+= 4096)
raf.write(_4k);
fc = raf.getChannel();
buf = fc.map(FileChannel.MapMode.READ_WRITE, 0, raf.length());
for (int i = 0; i < capacity; i++)
putKeyPos(UNUSED_ENTRY, i);
putWord(0x67636a64, MAGIC);
putWord(0x01, VERSION);
putWord(capacity, CAPACITY);
putWord(table_base, TABLE_BASE);
putWord(string_base, STRING_BASE);
putWord(file_size, FILE_SIZE);
putWord(elements, ELEMENTS);
buf.force();
length = fc.size();
string_size = 0;
}
static public PersistentByteMap
emptyPersistentByteMap(File name, int capacity, int strtabSize)
throws IOException
{
PersistentByteMap m = new PersistentByteMap(name);
m.init(m, name, capacity, strtabSize);
return m;
}
private int getWord (int index)
{
buf.position(index);
byte[] wordBuf = new byte[4];
buf.get(wordBuf);
int result = (int)wordBuf[0]&0xff;
result += ((int)wordBuf[1]&0xff) << 8;
result += ((int)wordBuf[2]&0xff) << 16;
result += ((int)wordBuf[3]&0xff) << 24;
return result;
}
private void putWord (int word, int index)
{
buf.position(index);
byte[] wordBuf = new byte[4];
wordBuf[0] = (byte)(word);
wordBuf[1] = (byte)(word >>> 8);
wordBuf[2] = (byte)(word >>> 16);
wordBuf[3] = (byte)(word >>> 24);
buf.put(wordBuf);
}
public Set entrySet()
{
return null;
}
private int getBucket(int n)
{
return table_base + (2*n * INT_SIZE);
}
private int getKeyPos(int n)
{
return getWord(getBucket(n));
}
private int getValuePos(int n)
{
return getWord(getBucket(n) + INT_SIZE);
}
private void putKeyPos(int index, int n)
{
putWord(index, getBucket(n));
}
private void putValuePos(int index, int n)
{
putWord(index, getBucket(n) + INT_SIZE);
}
private byte[] getBytes(int n)
{
int len = getWord (string_base + n);
int base = string_base + n + INT_SIZE;
byte[] key = new byte[len];
buf.position(base);
buf.get(key, 0, len);
return key;
}
private int hash (byte[] b)
{
// We assume that the message digest is evenly distributed, so we
// only need to use a few bytes of it as the hash function.
long hashIndex
= ((b[0]&0xffL)
+ ((b[1]&0xffL)<<8)
+ ((b[2]&0xffL)<<16)
+ ((b[3]&0xffL)<<24));
long result = hashIndex % (long)capacity;
return (int)result;
}
public byte[] get(byte[] digest)
{
int hashIndex = hash(digest);
do
{
int k = getKeyPos(hashIndex);
if (k == UNUSED_ENTRY)
return null;
if (Arrays.equals ((byte[])digest, getBytes(k)))
return getBytes(getValuePos(hashIndex));
// Use linear probing to resolve hash collisions. This may
// not be theoretically as good as open addressing, but it has
// good cache behviour.
hashIndex++;
hashIndex %= capacity;
}
while (true);
}
public void put(byte[] digest, byte[] value)
throws IllegalAccessException
{
int hashIndex = hash(digest);
if (elements >= capacity())
throw new IllegalAccessException("Table Full: " + elements);
do
{
int k = getKeyPos(hashIndex);
if (k == UNUSED_ENTRY)
{
int newKey = addBytes(digest);
putKeyPos(newKey, hashIndex);
int newValue = addBytes(value);
putValuePos(newValue, hashIndex);
elements++;
putWord(elements, ELEMENTS);
return;
}
else if (Arrays.equals (digest, getBytes(k)))
{
int newValue = addBytes((byte[])value);
putValuePos(newValue, hashIndex);
return;
}
hashIndex++;
hashIndex %= capacity;
}
while (true);
}
private int addBytes (byte[] data)
throws IllegalAccessException
{
if (data.length > 16)
{
// Keep track of long strings in the hope that we will be able
// to re-use them.
if (values == null)
{
values = new HashMap();
for (int i = 0; i < capacity; i++)
if (getKeyPos(i) != UNUSED_ENTRY)
{
int pos = getValuePos(i);
ByteWrapper bytes = new ByteWrapper(getBytes(pos));
values.put(bytes, new Integer(pos));
}
}
{
Object result = values.get(new ByteWrapper(data));
if (result != null)
{
// We already have this value in the string table
return ((Integer)result).intValue();
}
}
}
if (data.length + INT_SIZE >= this.length)
throw new IllegalAccessException("String table Full");
int extent = string_base+string_size;
int top = extent;
putWord(data.length, extent);
extent += INT_SIZE;
buf.position(extent);
buf.put(data, 0, data.length);
extent += data.length;
extent += INT_SIZE-1;
extent &= ~(INT_SIZE-1); // align
string_size = extent - string_base;
file_size = extent;
putWord (string_size, STRING_SIZE);
putWord (file_size, FILE_SIZE);
if (data.length > 16)
values.put(new ByteWrapper(data), new Integer(top - string_base));
return top - string_base;
}
public Iterator iterator(int type)
{
return new HashIterator(type);
}
public int size()
{
return elements;
}
public int stringTableSize()
{
return string_size;
}
public int capacity()
{
// With the the table 2/3 full there will be on average 2 probes
// for a successful search and 5 probes for an unsuccessful one.
return capacity * 2/3;
}
public void force()
{
buf.force();
}
public File getFile()
{
return name;
}
// Close the map. Once this has been done, the map can no longer be
// used.
public void close() throws IOException
{
force();
fc.close();
}
public void
putAll(PersistentByteMap t)
throws IllegalAccessException
{
// We can use a fast copy if the size of a map has not changed.
if (this.elements == 0 && t.capacity == this.capacity
&& t.length == this.length)
{
this.buf.position(0);
t.buf.position(0);
this.buf.put(t.buf);
this.table_base = t.table_base;
this.string_base = t.string_base;
this.string_size = t.string_size;
this.file_size = t.file_size;
this.elements = t.elements;
if (t.values != null)
this.values = (HashMap)t.values.clone();
return;
}
// Otherwise do it the hard way.
Iterator iterator = t.iterator(PersistentByteMap.ENTRIES);
while (iterator.hasNext())
{
PersistentByteMap.MapEntry entry
= (PersistentByteMap.MapEntry)iterator.next();
this.put((byte[])entry.getKey(), (byte[])entry.getValue());
}
}
private final class HashIterator implements Iterator
{
/** Current index in the physical hash table. */
private int idx;
private int count;
private final int type;
/**
* Construct a new HashIterator with the supplied type.
* @param type {@link #KEYS}, {@link #VALUES}, or {@link #ENTRIES}
*/
HashIterator(int type)
{
this.type = type;
count = elements;
idx = 0;
}
/**
* Returns true if the Iterator has more elements.
* @return true if there are more elements
* @throws ConcurrentModificationException if the HashMap was modified
*/
public boolean hasNext()
{
return count > 0;
}
/**
* Returns the next element in the Iterator's sequential view.
* @return the next element
* @throws ConcurrentModificationException if the HashMap was modified
* @throws NoSuchElementException if there is none
*/
public Object next()
{
count--;
for (int i = idx; i < capacity; i++)
if (getKeyPos(i) != UNUSED_ENTRY)
{
idx = i+1;
if (type == VALUES)
return getBytes(getValuePos(i));
if (type == KEYS)
return getBytes(getKeyPos(i));
return new MapEntry(i,
getBytes(getKeyPos(i)),
getBytes(getValuePos(i)));
}
return null;
}
/**
* Remove from the underlying collection the last element returned
* by next (optional operation). This method can be called only
* once after each call to <code>next()</code>. It does not affect
* what will be returned by subsequent calls to next.
*
* @throws IllegalStateException if next has not yet been called
* or remove has already been called since the last call
* to next.
* @throws UnsupportedOperationException if this Iterator does not
* support the remove operation.
*/
public void remove()
{
throw new UnsupportedOperationException();
}
}
static public final class MapEntry
{
private final Object key;
private final Object value;
private final int bucket;
public MapEntry(int bucket, Object newKey, Object newValue)
{
this.key = newKey;
this.value = newValue;
this.bucket = bucket;
}
public final Object getKey()
{
return key;
}
public final Object getValue()
{
return value;
}
public final int getBucket()
{
return bucket;
}
}
// A wrapper class for a byte array that allows collections to be
// made.
private final class ByteWrapper
{
final byte[] bytes;
final int hash;
public ByteWrapper (byte[] bytes)
{
int sum = 0;
this.bytes = bytes;
for (int i = 0; i < bytes.length; i++)
sum += bytes[i];
hash = sum;
}
public int hashCode()
{
return hash;
}
public boolean equals(Object obj)
{
return Arrays.equals(bytes, ((ByteWrapper)obj).bytes);
}
}
}