/**
* Returns pre values.
*
* @param ids unique node ids
* @param off start offset
* @param len number of ids
* @return sorted pre values
*/
public final int[] pre(final int[] ids, final int off, final int len) {
if (meta.updindex) return idmap.pre(ids, off, len);
final IntList p = new IntList(ids.length);
for (int i = off; i < len; ++i) p.add(preold(ids[i]));
return p.sort().toArray();
}
@Override
protected void indexDelete(final int pre, final int size) {
final boolean textI = meta.textindex, attrI = meta.attrindex;
if (textI || attrI) {
// collect all keys and ids
indexBegin();
final int l = pre + size;
for (int p = pre; p < l; ++p) {
final int k = kind(p);
// consider nodes which are attribute, text, comment, or proc. instruction
final boolean text = k == TEXT || k == COMM || k == PI;
if (textI && text || attrI && k == ATTR) {
final byte[] key = text(p, text);
if (key.length <= meta.maxlen) {
final TokenObjMap<IntList> m = text ? txtBuffer : atvBuffer;
IntList ids = m.get(key);
if (ids == null) {
ids = new IntList(1);
m.put(key, ids);
}
ids.add(id(p));
}
}
}
indexDelete();
}
}
/**
* Performs a wildcard search for the specified token.
*
* @param token token to look for
* @return iterator
*/
private synchronized IndexIterator wc(final byte[] token) {
final FTIndexIterator it = FTIndexIterator.FTEMPTY;
final FTWildcard wc = new FTWildcard(token);
if (!wc.parse()) return it;
final IntList pr = new IntList();
final IntList ps = new IntList();
final byte[] pref = wc.prefix();
final int pl = pref.length, tl = tp.length;
final int l = Math.min(tl - 1, wc.max());
for (int ti = pl; ti <= l; ti++) {
int i = tp[ti];
if (i == -1) continue;
int c = ti + 1;
int e = -1;
while (c < tl && e == -1) e = tp[c++];
i = find(pref, i, e, ti);
while (i < e) {
final byte[] t = inY.readBytes(i, ti);
if (!startsWith(t, pref)) break;
if (wc.match(t)) {
inZ.cursor(pointer(i, ti));
final int s = size(i, ti);
for (int d = 0; d < s; d++) {
pr.add(inZ.readNum());
ps.add(inZ.readNum());
}
}
i += ti + ENTRY;
}
}
return iter(new FTCache(pr, ps), token);
}
@Override
public IndexIterator iter(final IndexToken token) {
final int id = values.id(token.get());
if (id == 0) return IndexIterator.EMPTY;
final int len = lenList.get(id);
final int[] ids = idsList.get(id), pres;
if (data.meta.updindex) {
final IntList tmp = new IntList();
for (int i = 0; i < len; ++i) tmp.add(data.pre(ids[i]));
pres = tmp.sort().finish();
} else {
pres = ids;
}
return new IndexIterator() {
int p;
@Override
public boolean more() {
return p < len;
}
@Override
public int pre() {
return pres[p++];
}
@Override
public int size() {
return len;
}
};
}
@Override
protected long index(final int pre, final int id, final byte[] value, final int kind) {
final DataAccess store;
final TokenObjMap<IntList> map;
if (kind == ATTR) {
store = values;
map = meta.attrindex ? atvBuffer : null;
} else {
store = texts;
// don't index document names
map = meta.textindex && kind != DOC ? txtBuffer : null;
}
// add text to map to index later
if (meta.updindex && map != null && value.length <= meta.maxlen) {
IntList ids = map.get(value);
if (ids == null) {
ids = new IntList(1);
map.put(value, ids);
}
ids.add(id);
}
// add text to text file
// inline integer value...
final long v = toSimpleInt(value);
if (v != Integer.MIN_VALUE) return v | IO.OFFNUM;
// store text
final long off = store.length();
final byte[] val = COMP.get().pack(value);
store.writeToken(off, val);
return val == value ? off : off | IO.OFFCOMP;
}
@Override
public void add(final ValueCache cache) {
for (final byte[] key : cache) {
final IntList vals = cache.ids(key);
if (!vals.isEmpty()) add(key, vals.sort().finish());
}
finish();
}
/**
* Constructor.
*
* @param pr pre values
* @param ps positions
*/
private FTCache(final IntList pr, final IntList ps) {
final int s = pr.size();
final double[] v = new double[s];
for (int i = 0; i < s; i++) v[i] = (long) pr.get(i) << 32 | ps.get(i);
order = Array.createOrder(v, true);
pre = pr;
pos = ps;
}
/**
* Removes values from the index.
*
* @param key key
* @param vals sorted values
*/
void delete(final byte[] key, final int... vals) {
final int id = values.id(key), vl = vals.length, l = lenList.get(id), s = l - vl;
final int[] ids = idsList.get(id);
for (int i = 0, n = 0, v = 0; i < l; i++) {
if (v == vl || ids[i] != vals[v]) ids[n++] = ids[i];
else v++;
}
lenList.set(id, s);
if (s == 0) idsList.set(id, null);
}
/**
* Returns an iterator for an index entry.
*
* @param off offset on entries
* @param size number of id/pos entries
* @param da data source
* @param token index token
* @return iterator
*/
private static FTIndexIterator iter(
final long off, final int size, final DataAccess da, final byte[] token) {
da.cursor(off);
final IntList pr = new IntList(size);
final IntList ps = new IntList(size);
for (int c = 0; c < size; c++) {
pr.add(da.readNum());
ps.add(da.readNum());
}
return iter(new FTCache(pr, ps), token);
}
/** Finishes the index creation. */
void finish() {
if (reorder == null) return;
for (int i = 1; i < reorder.size(); i++) {
if (reorder.get(i)) Arrays.sort(idsList.get(i), 0, lenList.get(i));
}
reorder = null;
}
@Override
public int size() {
// returns the actual number of indexed entries
int s = 0;
for (int c = 1; c < s; c++) if (lenList.get(c) > 0) s++;
return s;
}
/**
* Adds values to the index.
*
* @param key key to be indexed
* @param vals sorted values
*/
void add(final byte[] key, final int... vals) {
// token index: add values. otherwise, reference existing values
final int id = type == IndexType.TOKEN ? values.put(key) : values.id(key), vl = vals.length;
// updatable index: if required, resize existing arrays
while (idsList.size() < id + 1) idsList.add(null);
if (lenList.size() < id + 1) lenList.set(id, 0);
final int len = lenList.get(id), size = len + vl;
int[] ids = idsList.get(id);
if (ids == null) {
ids = vals;
} else {
if (ids.length < size) ids = Arrays.copyOf(ids, Array.newSize(size));
System.arraycopy(vals, 0, ids, len, vl);
if (ids[len - 1] > vals[0]) {
if (reorder == null) reorder = new BoolList(values.size());
reorder.set(id, true);
}
}
idsList.set(id, ids);
lenList.set(id, size);
}
/**
* Returns a string representation of the index structure.
*
* @param all include database contents in the representation. During updates, database lookups
* must be avoided, as the data structures will be inconsistent.
* @return string
*/
public String toString(final boolean all) {
final TokenBuilder tb = new TokenBuilder();
tb.addExt(type).add(" INDEX, '").add(data.meta.name).add("':\n");
final int s = lenList.size();
for (int m = 1; m < s; m++) {
final int len = lenList.get(m);
if (len == 0) continue;
final int[] ids = idsList.get(m);
tb.add(" ").addInt(m);
if (all)
tb.add(", key: \"").add(data.text(data.pre(ids[0]), type == IndexType.TEXT)).add('"');
tb.add(", ids");
if (all) tb.add("/pres");
tb.add(": ");
for (int n = 0; n < len; n++) {
if (n != 0) tb.add(",");
tb.addInt(ids[n]);
if (all) tb.add('/').addInt(data.pre(ids[n]));
}
tb.add("\n");
}
return tb.toString();
}
@Override
public byte[] info(final MainOptions options) {
final TokenBuilder tb = new TokenBuilder();
tb.add(LI_STRUCTURE).add(HASH).add(NL);
tb.add(LI_NAMES).add(data.meta.names(type)).add(NL);
final IndexStats stats = new IndexStats(options.get(MainOptions.MAXSTAT));
final int s = values.size();
for (int p = 1; p <= s; p++) {
final int oc = lenList.get(p);
if (oc > 0 && stats.adding(oc)) stats.add(values.key(p), oc);
}
stats.print(tb);
return tb.finish();
}
/**
* Inserts a data instance at the specified pre value. Note that the specified data instance must
* differ from this instance.
*
* @param ipre value at which to insert new data
* @param ipar parent pre value of node
* @param clip data clip
*/
public final void insert(final int ipre, final int ipar, final DataClip clip) {
meta.update();
// update value and document indexes
if (meta.updindex) indexBegin();
resources.insert(ipre, clip);
final int dsize = clip.size();
final int buf = Math.min(dsize, IO.BLOCKSIZE >> IO.NODEPOWER);
// resize buffer to cache more entries
buffer(buf);
// find all namespaces in scope to avoid duplicate declarations
final TokenMap nsScope = nspaces.scope(ipar, this);
// loop through all entries
final IntList preStack = new IntList();
final NSNode nsRoot = nspaces.current();
final HashSet<NSNode> newNodes = new HashSet<NSNode>();
final IntList flagPres = new IntList();
// indicates if database only contains a dummy node
final Data data = clip.data;
int c = 0;
for (int dpre = clip.start; dpre < clip.end; ++dpre, ++c) {
if (c != 0 && c % buf == 0) insert(ipre + c - buf);
final int pre = ipre + c;
final int dkind = data.kind(dpre);
final int dpar = data.parent(dpre, dkind);
// ipar < 0 if document nodes on top level are added
final int dis = dpar >= 0 ? dpre - dpar : ipar >= 0 ? pre - ipar : 0;
final int par = dis == 0 ? -1 : pre - dis;
if (c == 0) nspaces.root(par, this);
while (!preStack.isEmpty() && preStack.peek() > par) nspaces.close(preStack.pop());
switch (dkind) {
case DOC:
// add document
nspaces.prepare();
final int s = data.size(dpre, dkind);
doc(pre, s, data.text(dpre, true));
meta.ndocs++;
preStack.push(pre);
break;
case ELEM:
// add element
nspaces.prepare();
boolean ne = false;
if (data.nsFlag(dpre)) {
final Atts at = data.ns(dpre);
for (int a = 0; a < at.size(); ++a) {
// see if prefix has been declared/ is part of current ns scope
final byte[] old = nsScope.get(at.name(a));
if (old == null || !eq(old, at.value(a))) {
// we have to keep track of all new NSNodes that are added
// to the Namespace structure, as their pre values must not
// be updated. I.e. if an NSNode N with pre value 3 existed
// prior to inserting and two new nodes are inserted at
// location pre == 3 we have to make sure N and only N gets
// updated.
newNodes.add(nspaces.add(at.name(a), at.value(a), pre));
ne = true;
}
}
}
byte[] nm = data.name(dpre, dkind);
elem(
dis,
tagindex.index(nm, null, false),
data.attSize(dpre, dkind),
data.size(dpre, dkind),
nspaces.uri(nm, true),
ne);
preStack.push(pre);
break;
case TEXT:
case COMM:
case PI:
// add text
text(pre, dis, data.text(dpre, true), dkind);
break;
case ATTR:
// add attribute
nm = data.name(dpre, dkind);
// check if prefix already in nsScope or not
final byte[] attPref = prefix(nm);
// check if prefix of attribute has already been declared, otherwise
// add declaration to parent node
if (data.nsFlag(dpre) && nsScope.get(attPref) == null) {
nspaces.add(
par,
preStack.isEmpty() ? -1 : preStack.peek(),
attPref,
data.nspaces.uri(data.uri(dpre, dkind)),
this);
// save pre value to set ns flag later for this node. can't be done
// here as direct table access would interfere with the buffer
flagPres.add(par);
}
attr(
pre,
dis,
atnindex.index(nm, null, false),
data.text(dpre, false),
nspaces.uri(nm, false),
false);
break;
}
}
// finalize and update namespace structure
while (!preStack.isEmpty()) nspaces.close(preStack.pop());
nspaces.root(nsRoot);
if (bp != 0) insert(ipre + c - 1 - (c - 1) % buf);
// reset buffer to old size
buffer(1);
// set ns flags
for (int f = 0; f < flagPres.size(); f++) {
final int fl = flagPres.get(f);
table.write2(fl, 1, name(fl) | 1 << 15);
}
// increase size of ancestors
int p = ipar;
while (p >= 0) {
final int k = kind(p);
size(p, k, size(p, k) + dsize);
p = parent(p, k);
}
if (meta.updindex) {
// add the entries to the ID -> PRE mapping:
idmap.insert(ipre, id(ipre), dsize);
indexEnd();
}
if (!cache) updateDist(ipre + dsize, dsize);
// propagate PRE value shifts to namespaces
if (ipar != -1) nspaces.insert(ipre, dsize, newNodes);
}
/**
* Optimizes the structures of a database.
*
* @param data data
* @param enforceText enforce creation or deletion of text index
* @param enforceAttr enforce creation or deletion of attribute index
* @param enforceToken enforce creation or deletion of token index
* @param enforceFt enforce creation or deletion of full-text index
* @param cmd calling command instance (may be {@code null})
* @throws IOException I/O Exception during index rebuild
*/
public static void optimize(
final Data data,
final boolean enforceText,
final boolean enforceAttr,
final boolean enforceToken,
final boolean enforceFt,
final Optimize cmd)
throws IOException {
// initialize structural indexes
final MetaData md = data.meta;
if (!md.uptodate) {
data.paths.init();
data.elemNames.init();
data.attrNames.init();
md.dirty = true;
final IntList pars = new IntList(), elms = new IntList();
int n = 0;
for (int pre = 0; pre < md.size; ++pre) {
final byte kind = (byte) data.kind(pre);
final int par = data.parent(pre, kind);
while (!pars.isEmpty() && pars.peek() > par) {
pars.pop();
elms.pop();
}
final int level = pars.size();
if (kind == Data.DOC) {
data.paths.put(0, Data.DOC, level);
pars.push(pre);
elms.push(0);
++n;
} else if (kind == Data.ELEM) {
final int id = data.nameId(pre);
data.elemNames.index(data.elemNames.key(id), null, true);
data.paths.put(id, Data.ELEM, level);
pars.push(pre);
elms.push(id);
} else if (kind == Data.ATTR) {
final int id = data.nameId(pre);
final byte[] val = data.text(pre, false);
data.attrNames.index(data.attrNames.key(id), val, true);
data.paths.put(id, Data.ATTR, level, val, md);
} else {
final byte[] val = data.text(pre, true);
if (kind == Data.TEXT && level > 1) data.elemNames.index(elms.peek(), val);
data.paths.put(0, kind, level, val, md);
}
if (cmd != null) cmd.pre = pre;
}
md.ndocs = n;
md.uptodate = true;
}
// rebuild value indexes
optimize(IndexType.TEXT, data, md.createtext, md.textindex, enforceText, cmd);
optimize(IndexType.ATTRIBUTE, data, md.createattr, md.attrindex, enforceAttr, cmd);
optimize(IndexType.TOKEN, data, md.createtoken, md.tokenindex, enforceToken, cmd);
optimize(IndexType.FULLTEXT, data, md.createft, md.ftindex, enforceFt, cmd);
}
@Override
public int costs(final IndexToken it) {
return lenList.get(values.id(it.get()));
}
