/**
* Computes the number of results.
*
* @param qc query context (may be @code null)
* @return number of results
*/
private long size(final QueryContext qc) {
final Value rt = initial(qc);
// skip computation if value is not a document node
if (rt == null || rt.type != NodeType.DOC) return -1;
final Data data = rt.data();
// skip computation if no database instance is available, is out-of-date or
// if context does not contain all database nodes
if (data == null || !data.meta.uptodate || data.resources.docs().size() != rt.size()) return -1;
ArrayList<PathNode> nodes = data.paths.root();
long m = 1;
final int sl = steps.length;
for (int s = 0; s < sl; s++) {
final Step curr = axisStep(s);
if (curr != null) {
nodes = curr.nodes(nodes, data);
if (nodes == null) return -1;
} else if (s + 1 == sl) {
m = steps[s].size();
} else {
// stop if a non-axis step is not placed last
return -1;
}
}
long sz = 0;
for (final PathNode pn : nodes) sz += pn.stats.count;
return sz * m;
}
@Override
public final Expr optimize(final QueryContext qc, final VarScope scp) throws QueryException {
final Value v = initial(qc);
if (v != null && v.isEmpty() || emptyPath(v)) return optPre(qc);
// merge descendant steps
Expr e = mergeSteps(qc);
if (e == this && v != null && v.type == NodeType.DOC) {
// check index access
e = index(qc, v);
// rewrite descendant to child steps
if (e == this) e = children(qc, v);
}
// recompile path if it has changed
if (e != this) return e.compile(qc, scp);
// set atomic type for single attribute steps to speed up predicate tests
if (root == null && steps.length == 1 && steps[0] instanceof Step) {
final Step curr = (Step) steps[0];
if (curr.axis == ATTR && curr.test.kind == Kind.URI_NAME) curr.seqType(SeqType.NOD_ZO);
}
// choose best path implementation and set type information
final Path path = get(info, root, steps);
path.size = path.size(qc);
path.seqType = SeqType.get(steps[steps.length - 1].seqType().type, size);
return path;
}
@Override
public final Expr compile(final QueryContext qc, final VarScope scp) throws QueryException {
if (root != null) root = root.compile(qc, scp);
// no steps
if (steps.length == 0) return root == null ? new Context(info) : root;
final Value init = qc.value, cv = initial(qc);
final boolean doc = cv != null && cv.type == NodeType.DOC;
qc.value = cv;
try {
final int sl = steps.length;
for (int s = 0; s < sl; s++) {
Expr e = steps[s];
// axis step: if input is a document, its type is temporarily generalized
final boolean as = e instanceof Step;
if (as && s == 0 && doc) cv.type = NodeType.NOD;
e = e.compile(qc, scp);
if (e.isEmpty()) return optPre(qc);
steps[s] = e;
// no axis step: invalidate context value
if (!as) qc.value = null;
}
} finally {
if (doc) cv.type = NodeType.DOC;
qc.value = init;
}
// optimize path
return optimize(qc, scp);
}
/**
* Returns a string representation of all found arguments.
*
* @param args array with arguments
* @return string representation
*/
static String foundArgs(final Value[] args) {
// compose found arguments
final StringBuilder sb = new StringBuilder();
for (final Value v : args) {
if (sb.length() != 0) sb.append(", ");
sb.append(v instanceof Jav ? Util.className(((Jav) v).toJava()) : v.seqType());
}
return sb.toString();
}
/**
* Returns the root of the current context or {@code null}.
*
* @param ctx query context
* @return root
*/
final Value root(final QueryContext ctx) {
final Value v = ctx != null ? ctx.value : null;
// no root specified: return context, if it does not reference a document
// as e.g. happens in //a(b|c)
if (root == null) return v == null || v.type != NodeType.DOC ? v : null;
// root is value: return root
if (root.isValue()) return (Value) root;
// no root reference, no context: return null
if (!(root instanceof Root) || v == null) return null;
// return context sequence or root of current context
return v.size() == 1 ? Root.root(v) : v;
}
@Override
public Expr compile(final QueryContext qc, final VarScope scp) throws QueryException {
final Value init = qc.value;
// never compile predicates with empty sequence as context value (#1016)
if (init != null && init.isEmpty()) qc.value = null;
try {
final int pl = preds.length;
for (int p = 0; p < pl; ++p) preds[p] = preds[p].compile(qc, scp).optimizeEbv(qc, scp);
return this;
} finally {
qc.value = init;
}
}
/** Tests the specified instance. */
@Test
public void test() {
final StringBuilder sb = new StringBuilder();
int fail = 0;
for (final Object[] qu : queries) {
final boolean correct = qu.length == 3;
final String query = qu[correct ? 2 : 1].toString();
final Value cmp = correct ? (Value) qu[1] : null;
final QueryProcessor qp = new QueryProcessor(query, context);
try {
final Value val = qp.value();
if (!correct || !new DeepCompare().equal(val, cmp)) {
sb.append("[" + qu[0] + "] " + query);
String s = correct && cmp.size() != 1 ? "#" + cmp.size() : "";
sb.append("\n[E" + s + "] ");
if (correct) {
final String cp = cmp.toString();
sb.append('\'');
sb.append(cp.length() > 1000 ? cp.substring(0, 1000) + "..." : cp);
sb.append('\'');
} else {
sb.append("error");
}
final TokenBuilder types = new TokenBuilder();
for (final Item it : val) types.add(it.type.toString()).add(" ");
s = val.size() == 1 ? "" : "#" + val.size();
sb.append("\n[F" + s + "] '" + val + "', " + types + details() + '\n');
++fail;
}
} catch (final Exception ex) {
final String msg = ex.getMessage();
if (correct || msg == null || msg.contains("mailman")) {
final String cp = correct && cmp.data() != null ? cmp.toString() : "()";
sb.append(
"["
+ qu[0]
+ "] "
+ query
+ "\n[E] "
+ cp
+ "\n[F] "
+ (msg == null ? Util.className(ex) : msg.replaceAll("\r\n?|\n", " "))
+ ' '
+ details()
+ '\n');
ex.printStackTrace();
++fail;
}
} finally {
qp.close();
}
}
if (fail != 0) fail(fail + " Errors. [E] = expected, [F] = found:\n" + sb.toString().trim());
}
/**
* Returns the path nodes that will result from this path.
*
* @param qc query context
* @return path nodes, or {@code null} if nodes cannot be evaluated
*/
public final ArrayList<PathNode> pathNodes(final QueryContext qc) {
final Value init = initial(qc);
final Data data = init != null && init.type == NodeType.DOC ? init.data() : null;
if (data == null || !data.meta.uptodate) return null;
ArrayList<PathNode> nodes = data.paths.root();
final int sl = steps.length;
for (int s = 0; s < sl; s++) {
final Step curr = axisStep(s);
if (curr == null) return null;
nodes = curr.nodes(nodes, data);
if (nodes == null) return null;
}
return nodes;
}
/**
* Returns a document test. This test will be called by {@link AxisPath#index} if the context
* value only consists of database nodes.
*
* @param rt root value
* @return document test
*/
static Test get(final Value rt) {
// use simple test if database contains only one document
final Data data = rt.data();
if (data.meta.ndocs == 1) return Test.DOC;
// adopt nodes from existing sequence
if (rt instanceof DBNodeSeq) {
final DBNodeSeq seq = (DBNodeSeq) rt;
return seq.all() ? Test.DOC : new InvDocTest(new IntList(seq.pres()), data);
}
// loop through all documents and add pre values of documents
// not more than 2^31 documents supported
final IntList il = new IntList((int) rt.size());
for (final Item it : rt) il.add(((DBNode) it).pre());
return new InvDocTest(il, data);
}
@Override
protected Expr opt(final CompileContext cc) throws QueryException {
if (allAreValues() && exprs[0].size() < UNROLL_LIMIT) {
// unroll the loop
cc.info(QueryText.OPTUNROLL_X, this);
final Value seq = (Value) exprs[0];
final int len = (int) seq.size();
// fn:for-each(...)
final Expr[] results = new Expr[len];
for (int i = 0; i < len; i++) {
results[i] = new DynFuncCall(info, sc, exprs[1], seq.itemAt(i)).optimize(cc);
}
return new List(info, results).optimize(cc);
}
return this;
}
/**
* Converts descendant to child steps.
*
* @param qc query context
* @param rt root value
* @return original or new expression
*/
private Expr children(final QueryContext qc, final Value rt) {
// skip if index does not exist or is out-dated, or if several namespaces occur in the input
final Data data = rt.data();
if (data == null || !data.meta.uptodate || data.nspaces.globalNS() == null) return this;
Path path = this;
final int sl = steps.length;
for (int s = 0; s < sl; s++) {
// don't allow predicates in preceding location steps
final Step prev = s > 0 ? axisStep(s - 1) : null;
if (prev != null && prev.preds.length != 0) break;
// ignore axes other than descendant, or numeric predicates
final Step curr = axisStep(s);
if (curr == null || curr.axis != DESC || curr.has(Flag.FCS)) continue;
// check if child steps can be retrieved for current step
ArrayList<PathNode> nodes = pathNodes(data, s);
if (nodes == null) continue;
// cache child steps
final ArrayList<QNm> qnm = new ArrayList<>();
while (nodes.get(0).parent != null) {
QNm nm = new QNm(data.elemNames.key(nodes.get(0).name));
// skip children with prefixes
if (nm.hasPrefix()) return this;
for (final PathNode p : nodes) {
if (nodes.get(0).name != p.name) nm = null;
}
qnm.add(nm);
nodes = PathSummary.parent(nodes);
}
qc.compInfo(OPTCHILD, steps[s]);
// build new steps
int ts = qnm.size();
final Expr[] stps = new Expr[ts + sl - s - 1];
for (int t = 0; t < ts; t++) {
final Expr[] preds = t == ts - 1 ? ((Preds) steps[s]).preds : new Expr[0];
final QNm nm = qnm.get(ts - t - 1);
final NameTest nt =
nm == null ? new NameTest(false) : new NameTest(nm, Kind.NAME, false, null);
stps[t] = Step.get(info, CHILD, nt, preds);
}
while (++s < sl) stps[ts++] = steps[s];
path = get(info, root, stps);
break;
}
// check if all steps yield results; if not, return empty sequence
final ArrayList<PathNode> nodes = pathNodes(qc);
if (nodes != null && nodes.isEmpty()) {
qc.compInfo(OPTPATH, path);
return Empty.SEQ;
}
return path;
}
/**
* Returns an equivalent expression which accesses an index. If the expression cannot be
* rewritten, the original expression is returned.
*
* <p>The following types of queries can be rewritten (in the examples, the equality comparison is
* used, which will be rewritten to {@link ValueAccess} instances):
*
* <pre>
* 1. A[text() = '...'] -> IA('...')
* 2. A[. = '...'] -> IA('...', A)
* 3. text()[. = '...'] -> IA('...')
* 4. A[B = '...'] -> IA('...', B)/parent::A
* 1. A[B/text() = '...'] -> IA('...')/parent::B/parent::A
* 2. A[B/C = '...'] -> IA('...', C)/parent::B/parent::A
* 7. A[@a = '...'] -> IA('...', @a)/parent::A
* 8. @a[. = '...'] -> IA('...', @a)</pre>
*
* Queries of type 1, 3, 5 will not yield any results if the string to be compared is empty.
*
* @param qc query context
* @param rt root value
* @return original or new expression
* @throws QueryException query exception
*/
private Expr index(final QueryContext qc, final Value rt) throws QueryException {
// only rewrite paths with data reference
final Data data = rt.data();
if (data == null) return this;
// cache index access costs
IndexInfo index = null;
// cheapest predicate and step
int iPred = 0, iStep = 0;
// check if path can be converted to an index access
final int sl = steps.length;
for (int s = 0; s < sl; s++) {
// only accept descendant steps without positional predicates
final Step step = axisStep(s);
if (step == null || !step.axis.down || step.has(Flag.FCS)) break;
// check if path is iterable (i.e., will be duplicate-free)
final boolean iter = pathNodes(data, s) != null;
final IndexContext ictx = new IndexContext(data, iter);
// choose cheapest index access
final int pl = step.preds.length;
for (int p = 0; p < pl; p++) {
final IndexInfo ii = new IndexInfo(ictx, qc, step);
if (!step.preds[p].indexAccessible(ii)) continue;
if (ii.costs == 0) {
// no results...
qc.compInfo(OPTNOINDEX, this);
return Empty.SEQ;
}
if (index == null || index.costs > ii.costs) {
index = ii;
iPred = p;
iStep = s;
}
}
}
// skip rewriting if no index access is possible, or if it is too expensive
if (index == null || index.costs > data.meta.size) return this;
// rewrite for index access
qc.compInfo(index.info);
// replace expressions for index access
final Step indexStep = index.step;
// collect remaining predicates
final int pl = indexStep.preds.length;
final ExprList newPreds = new ExprList(pl - 1);
for (int p = 0; p < pl; p++) {
if (p != iPred) newPreds.add(indexStep.preds[p]);
}
// invert steps that occur before index step and add them as predicate
final Test test = InvDocTest.get(rt);
final ExprList invSteps = new ExprList();
if (test != Test.DOC || !data.meta.uptodate || predSteps(data, iStep)) {
for (int s = iStep; s >= 0; s--) {
final Axis ax = axisStep(s).axis.invert();
if (s == 0) {
// add document test for collections and axes other than ancestors
if (test != Test.DOC || ax != Axis.ANC && ax != Axis.ANCORSELF)
invSteps.add(Step.get(info, ax, test));
} else {
final Step prev = axisStep(s - 1);
invSteps.add(Step.get(info, ax, prev.test, prev.preds));
}
}
}
if (!invSteps.isEmpty()) newPreds.add(get(info, null, invSteps.finish()));
// create resulting expression
final ExprList resultSteps = new ExprList();
final Expr resultRoot;
if (index.expr instanceof Path) {
final Path p = (Path) index.expr;
resultRoot = p.root;
resultSteps.add(p.steps);
} else {
resultRoot = index.expr;
}
if (!newPreds.isEmpty()) {
int ls = resultSteps.size() - 1;
Step step;
if (ls < 0 || !(resultSteps.get(ls) instanceof Step)) {
// add at least one self axis step
step = Step.get(info, Axis.SELF, Test.NOD);
ls++;
} else {
step = (Step) resultSteps.get(ls);
}
// add remaining predicates to last step
resultSteps.set(ls, step.addPreds(newPreds.finish()));
}
// add remaining steps
for (int s = iStep + 1; s < sl; s++) resultSteps.add(steps[s]);
return get(info, resultRoot, resultSteps.finish());
}