You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
559 lines
25 KiB
559 lines
25 KiB
// ReferenceContainer.java
|
|
// (C) 2006 by Michael Peter Christen; mc@yacy.net, Frankfurt a. M., Germany
|
|
// first published 04.07.2006 on http://yacy.net
|
|
//
|
|
// This is a part of YaCy, a peer-to-peer based web search engine
|
|
//
|
|
// $LastChangedDate$
|
|
// $LastChangedRevision$
|
|
// $LastChangedBy$
|
|
//
|
|
// LICENSE
|
|
//
|
|
// This program is free software; you can redistribute it and/or modify
|
|
// it under the terms of the GNU General Public License as published by
|
|
// the Free Software Foundation; either version 2 of the License, or
|
|
// (at your option) any later version.
|
|
//
|
|
// This program is distributed in the hope that it will be useful,
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
// GNU General Public License for more details.
|
|
//
|
|
// You should have received a copy of the GNU General Public License
|
|
// along with this program; if not, write to the Free Software
|
|
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
|
|
package de.anomic.kelondro.text;
|
|
|
|
import java.lang.reflect.Method;
|
|
import java.util.Collection;
|
|
import java.util.ConcurrentModificationException;
|
|
import java.util.Iterator;
|
|
import java.util.Map;
|
|
import java.util.Set;
|
|
import java.util.TreeMap;
|
|
|
|
import de.anomic.kelondro.index.Row;
|
|
import de.anomic.kelondro.index.RowSet;
|
|
import de.anomic.kelondro.order.Base64Order;
|
|
import de.anomic.kelondro.text.referencePrototype.WordReferenceRow;
|
|
import de.anomic.kelondro.util.ByteBuffer;
|
|
|
|
/**
|
|
* A ReferenceContainer is a set of ReferenceRows entries for a specific term.
|
|
* Since ReferenceRow entries are special Row entries, a collection of ReferenceRows
|
|
* can be contained in a RowSet.
|
|
* This class extends the RowSet with methods for the handling of
|
|
* special ReferenceRow Row entry objects.
|
|
*/
|
|
public class ReferenceContainer<ReferenceType extends Reference> extends RowSet {
|
|
|
|
private String termHash;
|
|
private ReferenceFactory<ReferenceType> factory;
|
|
|
|
public ReferenceContainer(final ReferenceFactory<ReferenceType> factory, final String termHash, final RowSet collection) {
|
|
super(collection);
|
|
this.factory = factory;
|
|
this.termHash = termHash;
|
|
}
|
|
|
|
public ReferenceContainer(final ReferenceFactory<ReferenceType> factory, final String termHash, final Row rowdef, final int objectCount) {
|
|
super(rowdef, objectCount);
|
|
this.termHash = termHash;
|
|
this.factory = factory;
|
|
this.lastTimeWrote = 0;
|
|
}
|
|
|
|
public ReferenceContainer<ReferenceType> topLevelClone() {
|
|
final ReferenceContainer<ReferenceType> newContainer = new ReferenceContainer<ReferenceType>(this.factory, this.termHash, this.rowdef, this.size());
|
|
newContainer.addAllUnique(this);
|
|
return newContainer;
|
|
}
|
|
|
|
public static <RT extends Reference> ReferenceContainer<RT> emptyContainer(final ReferenceFactory<RT> factory, final String wordHash, final int elementCount) {
|
|
return new ReferenceContainer<RT>(factory, wordHash, WordReferenceRow.urlEntryRow, elementCount);
|
|
}
|
|
|
|
public void setWordHash(final String newWordHash) {
|
|
this.termHash = newWordHash;
|
|
}
|
|
|
|
public long updated() {
|
|
return super.lastWrote();
|
|
}
|
|
|
|
public String getTermHash() {
|
|
return termHash;
|
|
}
|
|
|
|
public void add(final Reference entry) {
|
|
// add without double-occurrence test
|
|
assert entry.toKelondroEntry().objectsize() == super.rowdef.objectsize;
|
|
this.addUnique(entry.toKelondroEntry());
|
|
}
|
|
|
|
public ReferenceContainer<ReferenceType> merge(final ReferenceContainer<ReferenceType> c) {
|
|
return new ReferenceContainer<ReferenceType>(this.factory, this.termHash, super.merge(c));
|
|
}
|
|
|
|
public Reference put(final Reference entry) {
|
|
assert entry.toKelondroEntry().objectsize() == super.rowdef.objectsize;
|
|
final Row.Entry r = super.replace(entry.toKelondroEntry());
|
|
if (r == null) return null;
|
|
return new WordReferenceRow(r);
|
|
}
|
|
|
|
public boolean putRecent(final Reference entry) {
|
|
assert entry.toKelondroEntry().objectsize() == super.rowdef.objectsize;
|
|
// returns true if the new entry was added, false if it already existed
|
|
final Row.Entry oldEntryRow = this.replace(entry.toKelondroEntry());
|
|
if (oldEntryRow == null) {
|
|
return true;
|
|
}
|
|
final Reference oldEntry = new WordReferenceRow(oldEntryRow);
|
|
if (entry.isOlder(oldEntry)) { // A more recent Entry is already in this container
|
|
this.replace(oldEntry.toKelondroEntry()); // put it back
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
public int putAllRecent(final ReferenceContainer<ReferenceType> c) {
|
|
// adds all entries in c and checks every entry for double-occurrence
|
|
// returns the number of new elements
|
|
if (c == null) return 0;
|
|
int x = 0;
|
|
synchronized (c) {
|
|
final Iterator<ReferenceType> i = c.entries();
|
|
while (i.hasNext()) {
|
|
try {
|
|
if (putRecent(i.next())) x++;
|
|
} catch (final ConcurrentModificationException e) {
|
|
e.printStackTrace();
|
|
}
|
|
}
|
|
}
|
|
this.lastTimeWrote = java.lang.Math.max(this.lastTimeWrote, c.updated());
|
|
return x;
|
|
}
|
|
|
|
public ReferenceType get(final String urlHash) {
|
|
final Row.Entry entry = this.get(urlHash.getBytes());
|
|
if (entry == null) return null;
|
|
return this.factory.produceSlow(entry);
|
|
}
|
|
|
|
/**
|
|
* remove a url reference from the container.
|
|
* if the url hash was found, return the entry, but delete the entry from the container
|
|
* if the entry was not found, return null.
|
|
*/
|
|
public ReferenceType remove(final String urlHash) {
|
|
final Row.Entry entry = remove(urlHash.getBytes());
|
|
if (entry == null) return null;
|
|
return this.factory.produceSlow(entry);
|
|
}
|
|
|
|
public int removeEntries(final Set<String> urlHashes) {
|
|
int count = 0;
|
|
final Iterator<String> i = urlHashes.iterator();
|
|
while (i.hasNext()) count += (remove(i.next()) == null) ? 0 : 1;
|
|
return count;
|
|
}
|
|
|
|
public Iterator<ReferenceType> entries() {
|
|
// returns an iterator of indexRWIEntry objects
|
|
return new entryIterator();
|
|
}
|
|
|
|
public class entryIterator implements Iterator<ReferenceType> {
|
|
|
|
Iterator<Row.Entry> rowEntryIterator;
|
|
|
|
public entryIterator() {
|
|
rowEntryIterator = iterator();
|
|
}
|
|
|
|
public boolean hasNext() {
|
|
return rowEntryIterator.hasNext();
|
|
}
|
|
|
|
public ReferenceType next() {
|
|
final Row.Entry rentry = rowEntryIterator.next();
|
|
if (rentry == null) return null;
|
|
return factory.produceSlow(rentry);
|
|
}
|
|
|
|
public void remove() {
|
|
rowEntryIterator.remove();
|
|
}
|
|
|
|
}
|
|
|
|
public static Object mergeUnique(final Object a, final Object b) {
|
|
final ReferenceContainer<?> c = (ReferenceContainer<?>) a;
|
|
c.addAllUnique((ReferenceContainer<?>) b);
|
|
return c;
|
|
}
|
|
|
|
public static final Method containerMergeMethod;
|
|
static {
|
|
Method meth = null;
|
|
try {
|
|
final Class<?> c = Class.forName("de.anomic.kelondro.text.ReferenceContainer");
|
|
meth = c.getMethod("mergeUnique", new Class[]{Object.class, Object.class});
|
|
} catch (final SecurityException e) {
|
|
System.out.println("Error while initializing containerMerge.SecurityException: " + e.getMessage());
|
|
meth = null;
|
|
} catch (final ClassNotFoundException e) {
|
|
System.out.println("Error while initializing containerMerge.ClassNotFoundException: " + e.getMessage());
|
|
meth = null;
|
|
} catch (final NoSuchMethodException e) {
|
|
System.out.println("Error while initializing containerMerge.NoSuchMethodException: " + e.getMessage());
|
|
meth = null;
|
|
}
|
|
containerMergeMethod = meth;
|
|
}
|
|
|
|
public static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> joinExcludeContainers(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
final Collection<ReferenceContainer<ReferenceType>> includeContainers,
|
|
final Collection<ReferenceContainer<ReferenceType>> excludeContainers,
|
|
final int maxDistance) {
|
|
// join a search result and return the joincount (number of pages after join)
|
|
|
|
// since this is a conjunction we return an empty entity if any word is not known
|
|
if (includeContainers == null) return ReferenceContainer.emptyContainer(factory, null, 0);
|
|
|
|
// join the result
|
|
final ReferenceContainer<ReferenceType> rcLocal = ReferenceContainer.joinContainers(factory, includeContainers, maxDistance);
|
|
if (rcLocal == null) return ReferenceContainer.emptyContainer(factory, null, 0);
|
|
excludeContainers(factory, rcLocal, excludeContainers);
|
|
|
|
return rcLocal;
|
|
}
|
|
|
|
public static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> joinContainers(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
final Collection<ReferenceContainer<ReferenceType>> containers,
|
|
final int maxDistance) {
|
|
|
|
// order entities by their size
|
|
final TreeMap<Long, ReferenceContainer<ReferenceType>> map = new TreeMap<Long, ReferenceContainer<ReferenceType>>();
|
|
ReferenceContainer<ReferenceType> singleContainer;
|
|
final Iterator<ReferenceContainer<ReferenceType>> i = containers.iterator();
|
|
int count = 0;
|
|
while (i.hasNext()) {
|
|
// get next entity:
|
|
singleContainer = i.next();
|
|
|
|
// check result
|
|
if ((singleContainer == null) || (singleContainer.size() == 0)) return null; // as this is a cunjunction of searches, we have no result if any word is not known
|
|
|
|
// store result in order of result size
|
|
map.put(Long.valueOf(singleContainer.size() * 1000 + count), singleContainer);
|
|
count++;
|
|
}
|
|
|
|
// check if there is any result
|
|
if (map.size() == 0) return null; // no result, nothing found
|
|
|
|
// the map now holds the search results in order of number of hits per word
|
|
// we now must pairwise build up a conjunction of these sets
|
|
Long k = map.firstKey(); // the smallest, which means, the one with the least entries
|
|
ReferenceContainer<ReferenceType> searchA, searchB, searchResult = map.remove(k);
|
|
while ((map.size() > 0) && (searchResult.size() > 0)) {
|
|
// take the first element of map which is a result and combine it with result
|
|
k = map.firstKey(); // the next smallest...
|
|
searchA = searchResult;
|
|
searchB = map.remove(k);
|
|
searchResult = ReferenceContainer.joinConstructive(factory, searchA, searchB, maxDistance);
|
|
// free resources
|
|
searchA = null;
|
|
searchB = null;
|
|
}
|
|
|
|
// in 'searchResult' is now the combined search result
|
|
if (searchResult.size() == 0) return null;
|
|
return searchResult;
|
|
}
|
|
|
|
public static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> excludeContainers(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
ReferenceContainer<ReferenceType> pivot,
|
|
final Collection<ReferenceContainer<ReferenceType>> containers) {
|
|
|
|
// check if there is any result
|
|
if ((containers == null) || (containers.size() == 0)) return pivot; // no result, nothing found
|
|
|
|
final Iterator<ReferenceContainer<ReferenceType>> i = containers.iterator();
|
|
while (i.hasNext()) {
|
|
pivot = excludeDestructive(factory, pivot, i.next());
|
|
if ((pivot == null) || (pivot.size() == 0)) return null;
|
|
}
|
|
|
|
return pivot;
|
|
}
|
|
|
|
// join methods
|
|
private static int log2(int x) {
|
|
int l = 0;
|
|
while (x > 0) {x = x >> 1; l++;}
|
|
return l;
|
|
}
|
|
|
|
public static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> joinConstructive(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
final ReferenceContainer<ReferenceType> i1,
|
|
final ReferenceContainer<ReferenceType> i2,
|
|
final int maxDistance) {
|
|
if ((i1 == null) || (i2 == null)) return null;
|
|
if ((i1.size() == 0) || (i2.size() == 0)) return null;
|
|
|
|
// decide which method to use
|
|
final int high = ((i1.size() > i2.size()) ? i1.size() : i2.size());
|
|
final int low = ((i1.size() > i2.size()) ? i2.size() : i1.size());
|
|
final int stepsEnum = 10 * (high + low - 1);
|
|
final int stepsTest = 12 * log2(high) * low;
|
|
|
|
// start most efficient method
|
|
if (stepsEnum > stepsTest) {
|
|
if (i1.size() < i2.size())
|
|
return joinConstructiveByTest(factory, i1, i2, maxDistance);
|
|
else
|
|
return joinConstructiveByTest(factory, i2, i1, maxDistance);
|
|
}
|
|
return joinConstructiveByEnumeration(factory, i1, i2, maxDistance);
|
|
}
|
|
|
|
private static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> joinConstructiveByTest(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
final ReferenceContainer<ReferenceType> small,
|
|
final ReferenceContainer<ReferenceType> large,
|
|
final int maxDistance) {
|
|
System.out.println("DEBUG: JOIN METHOD BY TEST, maxdistance = " + maxDistance);
|
|
assert small.rowdef.equals(large.rowdef) : "small = " + small.rowdef.toString() + "; large = " + large.rowdef.toString();
|
|
final int keylength = small.rowdef.width(0);
|
|
assert (keylength == large.rowdef.width(0));
|
|
final ReferenceContainer<ReferenceType> conj = new ReferenceContainer<ReferenceType>(factory, null, small.rowdef, 0); // start with empty search result
|
|
final Iterator<ReferenceType> se = small.entries();
|
|
ReferenceType ie1;
|
|
ReferenceType ie2;
|
|
while (se.hasNext()) {
|
|
ie1 = se.next();
|
|
ie2 = large.get(ie1.metadataHash());
|
|
if ((ie1 != null) && (ie2 != null)) {
|
|
assert (ie1.metadataHash().length() == keylength) : "ie0.urlHash() = " + ie1.metadataHash();
|
|
assert (ie2.metadataHash().length() == keylength) : "ie1.urlHash() = " + ie2.metadataHash();
|
|
// this is a hit. Calculate word distance:
|
|
|
|
ie1 = factory.produceFast(ie2);
|
|
ie1.join(ie2);
|
|
if (ie1.distance() <= maxDistance) conj.add(ie1);
|
|
}
|
|
}
|
|
return conj;
|
|
}
|
|
|
|
private static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> joinConstructiveByEnumeration(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
final ReferenceContainer<ReferenceType> i1,
|
|
final ReferenceContainer<ReferenceType> i2,
|
|
final int maxDistance) {
|
|
System.out.println("DEBUG: JOIN METHOD BY ENUMERATION, maxdistance = " + maxDistance);
|
|
assert i1.rowdef.equals(i2.rowdef) : "i1 = " + i1.rowdef.toString() + "; i2 = " + i2.rowdef.toString();
|
|
final int keylength = i1.rowdef.width(0);
|
|
assert (keylength == i2.rowdef.width(0));
|
|
final ReferenceContainer<ReferenceType> conj = new ReferenceContainer<ReferenceType>(factory, null, i1.rowdef, 0); // start with empty search result
|
|
if (!((i1.rowdef.getOrdering().signature().equals(i2.rowdef.getOrdering().signature())))) return conj; // ordering must be equal
|
|
final Iterator<ReferenceType> e1 = i1.entries();
|
|
final Iterator<ReferenceType> e2 = i2.entries();
|
|
int c;
|
|
if ((e1.hasNext()) && (e2.hasNext())) {
|
|
ReferenceType ie1;
|
|
ReferenceType ie2;
|
|
ie1 = e1.next();
|
|
ie2 = e2.next();
|
|
|
|
while (true) {
|
|
assert (ie1.metadataHash().length() == keylength) : "ie1.urlHash() = " + ie1.metadataHash();
|
|
assert (ie2.metadataHash().length() == keylength) : "ie2.urlHash() = " + ie2.metadataHash();
|
|
c = i1.rowdef.getOrdering().compare(ie1.metadataHash().getBytes(), ie2.metadataHash().getBytes());
|
|
//System.out.println("** '" + ie1.getUrlHash() + "'.compareTo('" + ie2.getUrlHash() + "')="+c);
|
|
if (c < 0) {
|
|
if (e1.hasNext()) ie1 = e1.next(); else break;
|
|
} else if (c > 0) {
|
|
if (e2.hasNext()) ie2 = e2.next(); else break;
|
|
} else {
|
|
// we have found the same urls in different searches!
|
|
ie1 = factory.produceFast(ie1);
|
|
ie1.join(ie2);
|
|
if (ie1.distance() <= maxDistance) conj.add(ie1);
|
|
if (e1.hasNext()) ie1 = e1.next(); else break;
|
|
if (e2.hasNext()) ie2 = e2.next(); else break;
|
|
}
|
|
}
|
|
}
|
|
return conj;
|
|
}
|
|
|
|
public static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> excludeDestructive(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
final ReferenceContainer<ReferenceType> pivot,
|
|
final ReferenceContainer<ReferenceType> excl) {
|
|
if (pivot == null) return null;
|
|
if (excl == null) return pivot;
|
|
if (pivot.size() == 0) return null;
|
|
if (excl.size() == 0) return pivot;
|
|
|
|
// decide which method to use
|
|
final int high = ((pivot.size() > excl.size()) ? pivot.size() : excl.size());
|
|
final int low = ((pivot.size() > excl.size()) ? excl.size() : pivot.size());
|
|
final int stepsEnum = 10 * (high + low - 1);
|
|
final int stepsTest = 12 * log2(high) * low;
|
|
|
|
// start most efficient method
|
|
if (stepsEnum > stepsTest) {
|
|
return excludeDestructiveByTest(pivot, excl);
|
|
}
|
|
return excludeDestructiveByEnumeration(factory, pivot, excl);
|
|
}
|
|
|
|
private static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> excludeDestructiveByTest(
|
|
final ReferenceContainer<ReferenceType> pivot,
|
|
final ReferenceContainer<ReferenceType> excl) {
|
|
assert pivot.rowdef.equals(excl.rowdef) : "small = " + pivot.rowdef.toString() + "; large = " + excl.rowdef.toString();
|
|
final int keylength = pivot.rowdef.width(0);
|
|
assert (keylength == excl.rowdef.width(0));
|
|
final boolean iterate_pivot = pivot.size() < excl.size();
|
|
final Iterator<ReferenceType> se = (iterate_pivot) ? pivot.entries() : excl.entries();
|
|
Reference ie0, ie1;
|
|
while (se.hasNext()) {
|
|
ie0 = se.next();
|
|
ie1 = excl.get(ie0.metadataHash());
|
|
if ((ie0 != null) && (ie1 != null)) {
|
|
assert (ie0.metadataHash().length() == keylength) : "ie0.urlHash() = " + ie0.metadataHash();
|
|
assert (ie1.metadataHash().length() == keylength) : "ie1.urlHash() = " + ie1.metadataHash();
|
|
if (iterate_pivot) se.remove(); pivot.remove(ie0.metadataHash().getBytes());
|
|
}
|
|
}
|
|
return pivot;
|
|
}
|
|
|
|
private static <ReferenceType extends Reference> ReferenceContainer<ReferenceType> excludeDestructiveByEnumeration(
|
|
final ReferenceFactory<ReferenceType> factory,
|
|
final ReferenceContainer<ReferenceType> pivot,
|
|
final ReferenceContainer<ReferenceType> excl) {
|
|
assert pivot.rowdef.equals(excl.rowdef) : "i1 = " + pivot.rowdef.toString() + "; i2 = " + excl.rowdef.toString();
|
|
final int keylength = pivot.rowdef.width(0);
|
|
assert (keylength == excl.rowdef.width(0));
|
|
if (!((pivot.rowdef.getOrdering().signature().equals(excl.rowdef.getOrdering().signature())))) return pivot; // ordering must be equal
|
|
final Iterator<ReferenceType> e1 = pivot.entries();
|
|
final Iterator<ReferenceType> e2 = excl.entries();
|
|
int c;
|
|
if ((e1.hasNext()) && (e2.hasNext())) {
|
|
ReferenceType ie1;
|
|
ReferenceType ie2;
|
|
ie1 = e1.next();
|
|
ie2 = e2.next();
|
|
|
|
while (true) {
|
|
assert (ie1.metadataHash().length() == keylength) : "ie1.urlHash() = " + ie1.metadataHash();
|
|
assert (ie2.metadataHash().length() == keylength) : "ie2.urlHash() = " + ie2.metadataHash();
|
|
c = pivot.rowdef.getOrdering().compare(ie1.metadataHash().getBytes(), ie2.metadataHash().getBytes());
|
|
//System.out.println("** '" + ie1.getUrlHash() + "'.compareTo('" + ie2.getUrlHash() + "')="+c);
|
|
if (c < 0) {
|
|
if (e1.hasNext()) ie1 = e1.next(); else break;
|
|
} else if (c > 0) {
|
|
if (e2.hasNext()) ie2 = e2.next(); else break;
|
|
} else {
|
|
// we have found the same urls in different searches!
|
|
ie1 = factory.produceFast(ie1);
|
|
ie1.join(ie2);
|
|
e1.remove();
|
|
if (e1.hasNext()) ie1 = e1.next(); else break;
|
|
if (e2.hasNext()) ie2 = e2.next(); else break;
|
|
}
|
|
}
|
|
}
|
|
return pivot;
|
|
}
|
|
|
|
public String toString() {
|
|
return "C[" + termHash + "] has " + this.size() + " entries";
|
|
}
|
|
|
|
public int hashCode() {
|
|
return (int) Base64Order.enhancedCoder.decodeLong(this.termHash.substring(0, 4));
|
|
}
|
|
|
|
|
|
public static final <ReferenceType extends Reference> ByteBuffer compressIndex(final ReferenceContainer<ReferenceType> inputContainer, final ReferenceContainer<ReferenceType> excludeContainer, final long maxtime) {
|
|
// collect references according to domains
|
|
final long timeout = (maxtime < 0) ? Long.MAX_VALUE : System.currentTimeMillis() + maxtime;
|
|
final TreeMap<String, String> doms = new TreeMap<String, String>();
|
|
synchronized (inputContainer) {
|
|
final Iterator<ReferenceType> i = inputContainer.entries();
|
|
Reference iEntry;
|
|
String dom, paths;
|
|
while (i.hasNext()) {
|
|
iEntry = i.next();
|
|
if ((excludeContainer != null) && (excludeContainer.get(iEntry.metadataHash()) != null)) continue; // do not include urls that are in excludeContainer
|
|
dom = iEntry.metadataHash().substring(6);
|
|
if ((paths = doms.get(dom)) == null) {
|
|
doms.put(dom, iEntry.metadataHash().substring(0, 6));
|
|
} else {
|
|
doms.put(dom, paths + iEntry.metadataHash().substring(0, 6));
|
|
}
|
|
if (System.currentTimeMillis() > timeout)
|
|
break;
|
|
}
|
|
}
|
|
// construct a result string
|
|
final ByteBuffer bb = new ByteBuffer(inputContainer.size() * 6);
|
|
bb.append('{');
|
|
final Iterator<Map.Entry<String, String>> i = doms.entrySet().iterator();
|
|
Map.Entry<String, String> entry;
|
|
while (i.hasNext()) {
|
|
entry = i.next();
|
|
bb.append(entry.getKey());
|
|
bb.append(':');
|
|
bb.append(entry.getValue());
|
|
if (System.currentTimeMillis() > timeout)
|
|
break;
|
|
if (i.hasNext())
|
|
bb.append(',');
|
|
}
|
|
bb.append('}');
|
|
return bb;
|
|
}
|
|
|
|
public static final <ReferenceType extends Reference> void decompressIndex(final TreeMap<String, String> target, ByteBuffer ci, final String peerhash) {
|
|
// target is a mapping from url-hashes to a string of peer-hashes
|
|
if ((ci.byteAt(0) == '{') && (ci.byteAt(ci.length() - 1) == '}')) {
|
|
//System.out.println("DEBUG-DECOMPRESS: input is " + ci.toString());
|
|
ci = ci.trim(1, ci.length() - 2);
|
|
String dom, url, peers;
|
|
while ((ci.length() >= 13) && (ci.byteAt(6) == ':')) {
|
|
assert ci.length() >= 6 : "ci.length() = " + ci.length();
|
|
dom = ci.toString(0, 6);
|
|
ci.trim(7);
|
|
while ((ci.length() > 0) && (ci.byteAt(0) != ',')) {
|
|
assert ci.length() >= 6 : "ci.length() = " + ci.length();
|
|
url = ci.toString(0, 6) + dom;
|
|
ci.trim(6);
|
|
peers = target.get(url);
|
|
if (peers == null) {
|
|
target.put(url, peerhash);
|
|
} else {
|
|
target.put(url, peers + peerhash);
|
|
}
|
|
//System.out.println("DEBUG-DECOMPRESS: " + url + ":" + target.get(url));
|
|
}
|
|
if (ci.byteAt(0) == ',') ci.trim(1);
|
|
}
|
|
}
|
|
}
|
|
}
|