/** * Algorithm of Tarjan for computing the strongly connected components of a graph. * * @param v current node * @throws QueryException if a variable directly calls itself */ private void tarjan(final int v) throws QueryException { final int ixv = 2 * v, llv = ixv + 1, idx = next++; while (list.size() <= llv) list.add(-1); list.set(ixv, idx); list.set(llv, idx); stack.push(v); for (final int w : adjacentTo(v)) { final int ixw = 2 * w, llw = ixw + 1; if (list.size() <= ixw || list.get(ixw) < 0) { // Successor w has not yet been visited; recurse on it tarjan(w); list.set(llv, Math.min(list.get(llv), list.get(llw))); } else if (stack.contains(w)) { // Successor w is in stack S and hence in the current SCC list.set(llv, Math.min(list.get(llv), list.get(ixw))); } } // If v is a root node, pop the stack and generate an SCC if (list.get(llv) == list.get(ixv)) { int w; Scope[] out = null; do { w = stack.pop(); final Scope scp = scopes.get(w); out = out == null ? new Scope[] {scp} : Array.add(out, scp); } while (w != v); result.add(out); } }
/** * 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); }