/*
 * Copyright (c) 2018-2025 Tada AB and other contributors, as listed below.
 *
 * All rights reserved. This program and the accompanying materials
 * are made available under the terms of the The BSD 3-Clause License
 * which accompanies this distribution, and is available at
 * http://opensource.org/licenses/BSD-3-Clause
 *
 * Contributors:
 *   Chapman Flack
 */
package org.postgresql.pljava.example.saxon;

import java.math.BigDecimal;
import java.math.BigInteger;

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.ResultSet;
import java.sql.ResultSetMetaData;
import static java.sql.ResultSetMetaData.columnNoNulls;
import java.sql.SQLXML;
import java.sql.Statement;
import java.sql.Types;

import java.sql.SQLException;
import java.sql.SQLDataException;
import java.sql.SQLFeatureNotSupportedException;
import java.sql.SQLNonTransientException;
import java.sql.SQLSyntaxErrorException;

import java.time.LocalDate;
import java.time.LocalTime;
import java.time.OffsetTime;
import java.time.LocalDateTime;
import java.time.OffsetDateTime;
import static java.time.ZoneOffset.UTC;

import static java.util.Arrays.asList;
import static java.util.Arrays.fill;
import java.util.Collection;
import java.util.Collections;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.Properties;

import java.util.regex.Matcher;
import java.util.regex.Pattern;

import javax.xml.transform.Source;
import javax.xml.transform.Result;

import static javax.xml.XMLConstants.W3C_XML_SCHEMA_NS_URI;
import static javax.xml.XMLConstants.XML_NS_URI;
import static javax.xml.XMLConstants.XML_NS_PREFIX;
import static javax.xml.XMLConstants.XMLNS_ATTRIBUTE_NS_URI;
import static javax.xml.XMLConstants.XMLNS_ATTRIBUTE;

import net.sf.saxon.event.Receiver;

import net.sf.saxon.lib.ConversionRules;
import net.sf.saxon.lib.NamespaceConstant;

import static net.sf.saxon.om.NameChecker.isValidNCName;

import net.sf.saxon.query.StaticQueryContext;

import net.sf.saxon.regex.RegexIterator;
import net.sf.saxon.regex.RegularExpression;

import net.sf.saxon.s9api.Destination;
import net.sf.saxon.s9api.DocumentBuilder;
import net.sf.saxon.s9api.ItemType;
import net.sf.saxon.s9api.ItemTypeFactory;
import net.sf.saxon.s9api.OccurrenceIndicator;
import net.sf.saxon.s9api.Processor;
import net.sf.saxon.s9api.QName;
import net.sf.saxon.s9api.SAXDestination;
import net.sf.saxon.s9api.SequenceType;
import static net.sf.saxon.s9api.SequenceType.makeSequenceType;
import net.sf.saxon.s9api.XdmAtomicValue;
import static net.sf.saxon.s9api.XdmAtomicValue.makeAtomicValue;
import net.sf.saxon.s9api.XdmEmptySequence;
import net.sf.saxon.s9api.XdmItem;
import net.sf.saxon.s9api.XdmNode;
import static net.sf.saxon.s9api.XdmNodeKind.DOCUMENT;
import net.sf.saxon.s9api.XdmValue;
import net.sf.saxon.s9api.XdmSequenceIterator;
import net.sf.saxon.s9api.XQueryCompiler;
import net.sf.saxon.s9api.XQueryEvaluator;
import net.sf.saxon.s9api.XQueryExecutable;

import net.sf.saxon.s9api.SaxonApiException;

import net.sf.saxon.trans.XPathException;

import net.sf.saxon.serialize.SerializationProperties;

import net.sf.saxon.type.AtomicType;
import net.sf.saxon.type.Converter;

import net.sf.saxon.value.AtomicValue;
import net.sf.saxon.value.Base64BinaryValue;
import net.sf.saxon.value.CalendarValue;
import net.sf.saxon.value.HexBinaryValue;
import net.sf.saxon.value.StringValue;
import static net.sf.saxon.value.StringValue.getStringLength;

import org.postgresql.pljava.ResultSetProvider;

import org.postgresql.pljava.annotation.Function;
import org.postgresql.pljava.annotation.SQLAction;
import org.postgresql.pljava.annotation.SQLType;
import static org.postgresql.pljava.annotation.Function.OnNullInput.CALLED;

/* For the xmltext function, which only needs plain SAX and not Saxon */

import javax.xml.transform.sax.SAXResult;
import org.xml.sax.ContentHandler;
import org.xml.sax.SAXException;

/**
 * Class illustrating use of XQuery with Saxon as the
 * implementation, using its native "s9api".
 *<p>
 * Supplies alternative, XML Query-based (as the SQL/XML standard dictates)
 * implementation of some of SQL/XML, where the implementation in core
 * PostgreSQL is limited to the capabilities of XPath (and XPath 1.0, at that).
 *<p>
 * Without the syntatic sugar built into the core PostgreSQL parser, calls to
 * a function in this class can look a bit more verbose in SQL, but reflect a
 * straightforward rewriting from the standard syntax. For example, suppose
 * there is a table {@code catalog_as_xml} with a single row whose {@code x}
 * column is a (respectably sized) XML document recording the stuff in
 * {@code pg_catalog}. It could be created like this:
 *<pre>
 * CREATE TABLE catalog_as_xml(x) AS
 *   SELECT schema_to_xml('pg_catalog', false, true, '');
 *</pre>
 *<h2>Functions/predicates from ISO 9075-14 SQL/XML</h2>
 *<h3>XMLQUERY</h3>
 *<p>
 * In the syntax of the SQL/XML standard, here is a query that would return
 * an XML element representing the declaration of a function with a specified
 * name:
 *<pre>
 * SELECT XMLQUERY('/pg_catalog/pg_proc[proname eq $FUNCNAME]'
 *                 PASSING BY VALUE x, 'numeric_avg' AS FUNCNAME
 *                 RETURNING CONTENT EMPTY ON EMPTY)
 * FROM catalog_as_xml;
 *</pre>
 *<p>
 * It binds the 'context item' of the query to {@code x}, and the {@code NAME}
 * parameter to the given value, then evaluates the query and returns XML
 * "CONTENT" (a tree structure with a document node at the root, but not
 * necessarily meeting all the requirements of an XML "DOCUMENT"). It can be
 * rewritten as this call to the {@link #xq_ret_content xq_ret_content} method:
 *<pre>
 * SELECT javatest.xq_ret_content('/pg_catalog/pg_proc[proname eq $FUNCNAME]',
 *                                PASSING =&gt; p, nullOnEmpty =&gt; false)
 * FROM catalog_as_xml,
 * LATERAL (SELECT x AS ".", 'numeric_avg' AS "FUNCNAME") AS p;
 *</pre>
 *<p>
 * In the rewritten form, the form of result wanted ({@code RETURNING CONTENT})
 * is implicit in the called function name ({@code xq_ret_content}), and the
 * parameters to pass to the query are moved out to a separate {@code SELECT}
 * that supplies their values, types, and names (with the context item now given
 * the name ".") and is passed by its alias into the query function.
 *<p>
 * Because of an unconditional uppercasing that PL/Java's JDBC driver currently
 * applies to column names, any parameter names, such as {@code FUNCNAME} above,
 * must be spelled in uppercase where used in the XQuery text, or they will not
 * be recognized. Because the unconditional uppercasing is highly likely to be
 * dropped in a future PL/Java release, it is wisest until then to use only
 * parameter names that really are uppercase, both in the XQuery text where they
 * are used and in the SQL expression that supplies them. In PostgreSQL,
 * identifiers that are not quoted are <em>lower</em>cased, so they must be both
 * uppercase and quoted, in the SQL syntax, to be truly uppercase.
 *<p>
 * In the standard, parameters and results (of XML types) can be passed
 * {@code BY VALUE} or {@code BY REF}, where the latter means that the same
 * nodes will retain their XQuery node identities over calls (note that this is
 * a meaning unrelated to what "by value" and "by reference" usually mean in
 * PostgreSQL's documentation). PostgreSQL's implementation of the XML type
 * provides no way for {@code BY REF} semantics to be implemented, so everything
 * happening here happens {@code BY VALUE} implicitly, and does not need to be
 * specified.
 *<h3>XMLEXISTS</h3>
 *<p>
 * The function {@link #xmlexists xmlexists} here implements the
 * standard function of the same name. Because it is the same name, it has to
 * be either schema-qualified or double-quoted in a call to avoid confusion
 * with the reserved word. In the syntax of the SQL/XML standard, here is a
 * query returning a boolean value indicating whether a function with the
 * specified name is declared:
 *<pre>
 * SELECT XMLEXISTS('/pg_catalog/pg_proc[proname eq $FUNCNAME]'
 *                  PASSING BY VALUE x, 'numeric_avg' AS FUNCNAME)
 * FROM catalog_as_xml;
 *</pre>
 *<p>
 * It can be rewritten as this call to the {@link #xmlexists xmlexists} method:
 *<pre>
 * SELECT "xmlexists"('/pg_catalog/pg_proc[proname eq $FUNCNAME]',
 *                    PASSING =&gt; p)
 * FROM catalog_as_xml,
 * LATERAL (SELECT x AS ".", 'numeric_avg' AS "FUNCNAME") AS p;
 *</pre>
 *<h3>XMLTABLE</h3>
 *<p>
 * The function {@link #xmltable xmltable} here implements (much of) the
 * standard function of the same name. Because it is the same name, it has to
 * be either schema-qualified or double-quoted in a call to avoid confusion
 * with the reserved word. A rewritten form of the <a href=
'https://www.postgresql.org/docs/10/static/functions-xml.html#FUNCTIONS-XML-PROCESSING-XMLTABLE'
>first example in the PostgreSQL manual</a> could be:
 *<pre>
 * SELECT xmltable.*
 * FROM
 *  xmldata,
 *
 *  LATERAL (SELECT data AS ".", 'not specified'::text AS "DPREMIER") AS p,
 *
 *  "xmltable"('//ROWS/ROW', PASSING =&gt; p, COLUMNS =&gt; ARRAY[
 *   'data(@id)', null, 'COUNTRY_NAME',
 *   'COUNTRY_ID', 'SIZE[@unit eq "sq_km"]',
 *   'concat(SIZE[@unit ne "sq_km"], " ", SIZE[@unit ne "sq_km"]/@unit)',
 *   'let $e := PREMIER_NAME
 *    return if ( empty($e) )then $DPREMIER else $e'
 *  ]) AS (
 *   id int, ordinality int8, "COUNTRY_NAME" text, country_id text,
 *   size_sq_km float, size_other text, premier_name text
 *  );
 *</pre>
 *<p>
 * In the first column expression, without the {@code data()} function, the
 * result would be a bare attribute node (one not enclosed in an XML element).
 * Many implementations will accept a bare attribute as a column expression
 * result, and simply assume the attribute's value is wanted, but it appears
 * that a strict implementation of the spec must raise {@code err:XPTY0004} in
 * such a case. This implementation is meant to be strict, so the attribute is
 * wrapped in {@code data()} to extract and return its value. (See
 * "About bare attribute nodes" in {@link #assignRowValues assignRowValues}
 * for more explanation.)
 *<p>
 * The {@code DPREMIER} parameter passed from SQL to the XQuery expression is
 * spelled in uppercase (and also, in the SQL expression supplying it, quoted),
 * for the reasons explained above for the {@code xq_ret_content} function.
 *<h3>XMLCAST</h3>
 *<p>
 * An ISO standard cast expression like
 *<pre>
 * XMLCAST(v AS wantedtype)
 *</pre>
 * can be rewritten with this idiom and the {@link #xmlcast xmlcast} function
 * provided here:
 *<pre>
 * (SELECT r FROM (SELECT v) AS o, xmlcast(o) AS (r wantedtype))
 *</pre>
 *<h2>XQuery regular-expression functions in ISO 9075-2 Foundations</h2>
 * The methods {@link #like_regex like_regex},
 * {@link #occurrences_regex occurrences_regex},
 * {@link #position_regex position_regex},
 * {@link #substring_regex substring_regex}, and
 * {@link #translate_regex translate_regex} provide, with slightly altered
 * syntax, the ISO SQL predicate and functions of the same names.
 *<p>
 * For the moment, they will only match newlines in the way W3C XQuery
 * specifies, not in the more-flexible Unicode-compatible way ISO SQL specifies,
 * and for the ones where ISO SQL allows {@code USING CHARACTERS} or
 * {@code USING OCTETS}, only {@code USING CHARACTERS} will work.
 *<h2>Extensions</h2>
 *<h3>XQuery module prolog allowed</h3>
 *<p>
 * Where any function here accepts an XQuery
 *<a href='https://www.w3.org/TR/xquery-31/#id-expressions'
 *>"expression"</a> according to the SQL specification, in fact an XQuery
 *<a href='https://www.w3.org/TR/xquery-31/#dt-main-module'
 *>"main module"</a> will be accepted. Therefore, the query can be preceded by
 * a prolog declaring namespaces, options, local variables and functions, etc.
 *<h3>Saxon extension to XQuery regular expressions</h3>
 *<p>
 * Saxon's implementation of XQuery regular expressions will accept a
 * nonstandard <em>flag</em> string ending with {@code ;j} to use Java regular
 * expressions rather than XQuery ones. That extension is available in the
 * XQuery regular-expression methods provided here.
 * @author Chapman Flack
 */
@SQLAction(
    implementor = "postgresql_xml",  // skip it all if no xml support
    requires = "presentOnClassPath",
    provides = "saxon9api",
    install =
    "SELECT CASE WHEN" +
    " presentOnClassPath('net.sf.saxon.s9api.QName')" +
    "THEN" +
    " CAST(" +
    "  set_config('pljava.implementors', 'saxon9api,' || " +
    "   current_setting('pljava.implementors'), true)" +
    "  AS void" +
    " )" +
    "ELSE" +
    " logMessage('INFO', 'Saxon examples skipped: s9api classes missing')" +
    "END"
)
public class S9 implements ResultSetProvider.Large
{
    private S9(
        XdmSequenceIterator<XdmItem> xsi,
        XQueryEvaluator[] columnXQEs,
        SequenceType[] columnStaticTypes,
        XMLBinary enc)
    {
        m_sequenceIterator = xsi;
        m_columnXQEs = columnXQEs;
        m_columnStaticTypes = columnStaticTypes;
        m_atomize = new AtomizingFunction [ columnStaticTypes.length ];
        m_xmlbinary = enc;
    }

    final XdmSequenceIterator<XdmItem> m_sequenceIterator;
    final XQueryEvaluator[] m_columnXQEs;
    final SequenceType[] m_columnStaticTypes;
    final SequenceType s_01untypedAtomic = makeSequenceType(
        ItemType.UNTYPED_ATOMIC, OccurrenceIndicator.ZERO_OR_ONE);
    final AtomizingFunction[] m_atomize;
    final XMLBinary m_xmlbinary;
    Binding.Assemblage m_outBindings;

    static final Connection s_dbc;
    static final Processor s_s9p = new Processor(false);
    static final ItemTypeFactory s_itf = new ItemTypeFactory(s_s9p);

    static final Pattern s_intervalSigns;
    static final Pattern s_intervalSignSite;

    enum XMLBinary { HEX, BASE64 };
    enum Nulls { ABSENT, NIL };

    static
    {
        try
        {
            s_dbc = DriverManager.getConnection("jdbc:default:connection");

            /*
             * XML Schema thinks an ISO 8601 duration must have no sign
             * anywhere but at the very beginning before the P. PostgreSQL
             * thinks that's the one place a sign must never be, and instead
             * it should appear in front of every numeric field. (PostgreSQL
             * accepts input where the signs vary, and there are cases where it
             * cannot be normalized away: P1M-1D is a thing, and can't be
             * simplified until anchored at a date to know how long the month
             * is! The XML Schema type simply can't represent that, so mapping
             * of such a value must simply fail, as we'll ensure below.)
             * So, here's a regex with a capturing group for a leading -, and
             * one for any field-leading -, and one for the absence of a field-
             * leading -. Any PostgreSQL or XS duration ought to match overall,
             * but the capturing group matches should be either (f,f,t) or
             * (f,t,f) for a PostgreSQL duration, or either (f,f,t) or (t,f,t)
             * for an XS duration. (f,t,t) would be a PostgreSQL interval with
             * mixed signs, and inconvertible.
             */
            s_intervalSigns = Pattern.compile(
            "(-)?+(?:[PYMWDTH](?:(?:(-)|())\\d++)?+)++(?:(?:[.,]\\d*+)?+S)?+");
            /*
             * To convert from the leading-sign form, need to find every spot
             * where a digit follows a [PYMWDTH] to insert a - there.
             */
            s_intervalSignSite = Pattern.compile("(?<=[PYMWDTH])(?=\\d)");
        }
        catch ( SQLException e )
        {
            throw new ExceptionInInitializerError(e);
        }
    }

    static class PredefinedQueryHolders
    {
        static final XQueryCompiler s_xqc = s_s9p.newXQueryCompiler();
        static final QName s_qEXPR = new QName("EXPR");

        static class DocumentWrapUnwrap
        {
            static final XQueryExecutable INSTANCE;

            static
            {
                try
                {
                    INSTANCE = s_xqc.compile(
                        "declare construction preserve;" +
                        "declare variable $EXPR as item()* external;" +
                        "data(document{$EXPR}/child::node())");
                }
                catch ( SaxonApiException e )
                {
                    throw new ExceptionInInitializerError(e);
                }
            }
        }
    }

    /**
     * PostgreSQL (as of 12) lacks the XMLTEXT function, so here it is.
     *<p>
     * As long as PostgreSQL does not have the {@code XML(SEQUENCE)} type,
     * this can only be the {@code XMLTEXT(sve RETURNING CONTENT)} flavor, which
     * does create a text node with {@code sve} as its value, but returns the
     * text node wrapped in a document node.
     *<p>
     * This function doesn't actually require Saxon, but otherwise fits in with
     * the theme here, implementing missing parts of SQL/XML for PostgreSQL.
     * @param sve SQL string value to use in a text node
     * @return XML content, the text node wrapped in a document node
     */
    @Function(implementor="saxon9api", schema="javatest")
    public static SQLXML xmltext(String sve) throws SQLException
    {
        SQLXML rx = s_dbc.createSQLXML();
        ContentHandler ch = rx.setResult(SAXResult.class).getHandler();

        try
        {
            ch.startDocument();
            /*
             * It seems XMLTEXT() should be such a trivial function to write,
             * but already it reveals a subtlety in the SAX API docs. They say
             * the third argument to characters() is "the number of characters
             * to read from the array" and that follows a long discussion of how
             * individual characters can (with code points above U+FFFF) consist
             * of more than one Java char value.
             *
             * And yet, when you try it out (and include some characters above
             * U+FFFF in the input), you discover the third argument isn't the
             * number of characters, has to be the number of Java char values.
             */
            ch.characters(sve.toCharArray(), 0, sve.length());
            ch.endDocument();
        }
        catch ( SAXException e )
        {
            rx.free();
            throw new SQLException(e.getMessage(), e);
        }

        return rx;
    }

    /**
     * An implementation of XMLCAST.
     *<p>
     * Will be declared to take and return type {@code RECORD}, where each must
     * have exactly one component, just because that makes it easy to use
     * existing JDBC metadata queries to find out the operand and target SQL
     * data types.
     *<p>
     * Serving suggestion: rewrite this ISO standard expression
     *<pre>
     * XMLCAST(v AS wantedtype)
     *</pre>
     * to this idiomatic one:
     *<pre>
     * (SELECT r FROM (SELECT v) AS o, xmlcast(o) AS (r wantedtype))
     *</pre>
     * @param operand a one-row, one-column record supplied by the caller, whose
     * one typed value is the operand to be cast.
     * @param base64 true if binary SQL values should be base64-encoded in XML;
     * if false (the default), values will be encoded in hex.
     * @param target a one-row, one-column record supplied by PL/Java from the
     * {@code AS} clause after the function call, whose one column's type is the
     * type to be cast to.
     */
    @Function(
        implementor="saxon9api",
        schema="javatest",
        type="pg_catalog.record",
        onNullInput=CALLED,
        settings="IntervalStyle TO iso_8601"
    )
    public static boolean xmlcast(
        ResultSet operand, @SQLType(defaultValue="false") Boolean base64,
        ResultSet target)
        throws SQLException
    {
        if ( null == operand )
            throw new SQLDataException(
                "xmlcast \"operand\" must be (in this implementation) " +
                "a non-null row type", "22004");

        if ( null == base64 )
            throw new SQLDataException(
                "xmlcast \"base64\" must be true or false, not null", "22004");
        XMLBinary enc = base64 ? XMLBinary.BASE64 : XMLBinary.HEX;

        assert null != target : "PL/Java supplied a null output record???";

        if ( 1 != operand.getMetaData().getColumnCount() )
            throw new SQLDataException(
                "xmlcast \"operand\" must be a row type with exactly " +
                "one component", "22000");

        if ( 1 != target.getMetaData().getColumnCount() )
            throw new SQLDataException(
                "xmlcast \"target\" must be a row type with exactly " +
                "one component", "22000");

        Binding.Parameter op =
            new BindingsFromResultSet(operand, false).iterator().next();

        Binding.Parameter tg =
            new BindingsFromResultSet(target, null).iterator().next();

        int sd = op.typeJDBC();
        int td = tg.typeJDBC();

        int castcase =
            (Types.SQLXML == sd ? 2 : 0) | (Types.SQLXML == td ? 1 : 0);

        switch ( castcase )
        {
        case 0: // neither sd nor td is an XML type
            throw new SQLSyntaxErrorException(
                "at least one of xmlcast \"operand\" or \"target\" must " +
                "be of XML type", "42804");
        case 3: // both XML
            /*
             * In an implementation closely following the spec, this case would
             * be handled in parse analysis and rewritten from an XMLCAST to a
             * plain CAST, and this code would never see it. This is a plain
             * example function without benefit of a parser that can do that.
             * In a DBMS with all the various SQL:2006 XML subtypes, there would
             * be nontrivial work to do here, but casting from PostgreSQL's one
             * XML type to itself is more of a warm-up exercise.
             */
            target.updateSQLXML(1, operand.getSQLXML(1));
            return true;
        case 1: // something non-XML being cast to XML
            assertCanCastAsXmlSequence(sd, "operand");
            Object v = op.valueJDBC();
            if ( null == v )
            {
                target.updateNull(1);
                return true;
            }
            ItemType xsbt =
                mapSQLDataTypeToXMLSchemaDataType(op, enc, Nulls.ABSENT);
            Iterator<XdmItem> tv =
                xmlCastAsSequence(v, enc, xsbt).iterator();
            try
            {
                target.updateSQLXML(1,
                    returnContent(tv, /*nullOnEmpty*/ false));
            }
            catch ( SaxonApiException | XPathException e )
            {
                throw new SQLException(e.getMessage(), "10000", e);
            }
            return true;
        case 2: // XML being cast to something non-XML
            assertCanCastAsXmlSequence(td, "target");
            SQLXML sx = operand.getSQLXML(1);
            if ( null == sx )
            {
                target.updateNull(1);
                return true;
            }
            DocumentBuilder dBuilder = s_s9p.newDocumentBuilder();
            Source source = sx.getSource(null);
            try
            {
                XdmValue xv = dBuilder.build(source);
                XQueryEvaluator xqe =
                    PredefinedQueryHolders.DocumentWrapUnwrap.INSTANCE.load();
                xqe.setExternalVariable(PredefinedQueryHolders.s_qEXPR, xv);
                xv = xqe.evaluate();
                /*
                 * It's zero-or-one, or XPTY0004 was thrown here.
                 */
                if ( 0 == xv.size() )
                {
                    target.updateNull(1);
                    return true;
                }
                XdmAtomicValue av = (XdmAtomicValue)xv;
                xmlCastAsNonXML(
                    av, ItemType.UNTYPED_ATOMIC, tg, target, 1, enc);
            }
            catch ( SaxonApiException | XPathException e )
            {
                throw new SQLException(e.getMessage(), "10000", e);
            }
            return true;
        }

        throw new SQLFeatureNotSupportedException(
            "cannot yet xmlcast from " + op.typePG() +
            " to " + tg.typePG(), "0A000");
    }

    /**
     * A simple example corresponding to {@code XMLQUERY(expression
     * PASSING BY VALUE passing RETURNING CONTENT {NULL|EMPTY} ON EMPTY)}.
     * @param expression An XQuery expression. Must not be {@code null} (in the
     * SQL standard {@code XMLQUERY} syntax, it is not even allowed to be an
     * SQL expression at all, only a string literal).
     * @param nullOnEmpty pass {@code true} to get a null return in place of
     * an empty sequence, or {@code false} to just get the empty sequence.
     * @param passing A row value whose columns will be supplied to the query
     * as parameters. Columns with names (typically supplied with {@code AS})
     * appear as predeclared external variables with matching names (in no
     * namespace) in the query, with types derived from the SQL types of the
     * row value's columns. There may be one (and no more than one)
     * column with {@code AS "."} which, if present, will be bound as the
     * context item. (The name {@code ?column?}, which PostgreSQL uses for an
     * otherwise-unnamed column, is also accepted, which will often allow the
     * context item to be specified with no {@code AS} at all. Beware, though,
     * that PostgreSQL likes to invent column names from any function or type
     * name that may appear in the value expression, so this shorthand will not
     * always work, while {@code AS "."} will.) PL/Java's internal JDBC uppercases all column
     * names, so any uses of the corresponding variables in the query must have
     * the names in upper case. It is safest to also uppercase their appearances
     * in the SQL (for which, in PostgreSQL, they must be quoted), so that the
     * JDBC uppercasing is not being relied on. It is likely to be dropped in a
     * future PL/Java release.
     * @param namespaces An even-length String array where, of each pair of
     * consecutive entries, the first is a namespace prefix and the second is
     * the URI to which to bind it. The zero-length prefix sets the default
     * element and type namespace; if the prefix has zero length, the URI may
     * also have zero length, to declare that unprefixed elements are in no
     * namespace.
     */
    @Function(
        implementor="saxon9api",
        schema="javatest",
        onNullInput=CALLED,
        settings="IntervalStyle TO iso_8601"
    )
    public static SQLXML xq_ret_content(
        String expression, Boolean nullOnEmpty,
        @SQLType(defaultValue={}) ResultSet passing,
        @SQLType(defaultValue={}) String[] namespaces)
        throws SQLException
    {
        /*
         * The expression itself may not be null (in the standard, it isn't
         * even allowed to be dynamic, and can only be a string literal!).
         */
        if ( null == expression )
            throw new SQLDataException(
                "XMLQUERY expression may not be null", "22004");

        if ( null == nullOnEmpty )
            throw new SQLDataException(
                "XMLQUERY nullOnEmpty may not be null", "22004");

        try
        {
            XdmSequenceIterator<XdmItem> x1 =
                evalXQuery(expression, passing, namespaces);
            return null == x1 ? null : returnContent(x1, nullOnEmpty);
        }
        catch ( SaxonApiException | XPathException e )
        {
            throw new SQLException(e.getMessage(), "10000", e);
        }
    }

    /**
     * An implementation of {@code XMLEXISTS(expression
     * PASSING BY VALUE passing)}, using genuine XQuery.
     * @param expression An XQuery expression. Must not be {@code null} (in the
     * SQL standard {@code XMLQUERY} syntax, it is not even allowed to be an
     * SQL expression at all, only a string literal).
     * @param passing A row value whose columns will be supplied to the query
     * as parameters. Columns with names (typically supplied with {@code AS})
     * appear as predeclared external variables with matching names (in no
     * namespace) in the query, with types derived from the SQL types of the
     * row value's columns. There may be one (and no more than one)
     * column with {@code AS "."} which, if present, will be bound as the
     * context item. (The name {@code ?column?}, which PostgreSQL uses for an
     * otherwise-unnamed column, is also accepted, which will often allow the
     * context item to be specified with no {@code AS} at all. Beware, though,
     * that PostgreSQL likes to invent column names from any function or type
     * name that may appear in the value expression, so this shorthand will not
     * always work, while {@code AS "."} will.) PL/Java's internal JDBC uppercases all column
     * names, so any uses of the corresponding variables in the query must have
     * the names in upper case. It is safest to also uppercase their appearances
     * in the SQL (for which, in PostgreSQL, they must be quoted), so that the
     * JDBC uppercasing is not being relied on. It is likely to be dropped in a
     * future PL/Java release.
     * @param namespaces An even-length String array where, of each pair of
     * consecutive entries, the first is a namespace prefix and the second is
     * the URI to which to bind it. The zero-length prefix sets the default
     * element and type namespace; if the prefix has zero length, the URI may
     * also have zero length, to declare that unprefixed elements are in no
     * namespace.
     * @return True if the expression evaluates to a nonempty sequence, false if
     * it evaluates to an empty one. Null if a context item is passed and its
     * SQL value is null.
     */
    @Function(
        implementor="saxon9api",
        schema="javatest",
        onNullInput=CALLED,
        settings="IntervalStyle TO iso_8601"
    )
    public static Boolean xmlexists(
        String expression,
        @SQLType(defaultValue={}) ResultSet passing,
        @SQLType(defaultValue={}) String[] namespaces)
        throws SQLException
    {
        /*
         * The expression itself may not be null (in the standard, it isn't
         * even allowed to be dynamic, and can only be a string literal!).
         */
        if ( null == expression )
            throw new SQLDataException(
                "XMLEXISTS expression may not be null", "22004");

        XdmSequenceIterator<XdmItem> x1 =
            evalXQuery(expression, passing, namespaces);
        if ( null == x1 )
            return null;
        if ( ! x1.hasNext() )
            return false;
        x1.close();
        return true;
    }

    /**
     * Implementation factor of XMLEXISTS and XMLQUERY.
     * @return null if a context item is passed and its SQL value is null
     */
    private static XdmSequenceIterator<XdmItem> evalXQuery(
        String expression, ResultSet passing, String[] namespaces)
        throws SQLException
    {
        Binding.Assemblage bindings = new BindingsFromResultSet(passing, true);

        try
        {
            XQueryCompiler xqc = createStaticContextWithPassedTypes(
                bindings, namespaceBindings(namespaces));

            XQueryEvaluator xqe = xqc.compile(expression).load();

            if ( storePassedValuesInDynamicContext(xqe, bindings, true) )
                return null;

            /*
             * For now, punt on whether the <XQuery expression> is evaluated
             * with XML 1.1 or 1.0 lexical rules....  XXX
             */
            return xqe.iterator();
        }
        catch ( SaxonApiException | XPathException e )
        {
            throw new SQLException(e.getMessage(), "10000", e);
        }
    }

    /**
     * Perform the final steps of <em>something</em> {@code RETURNING CONTENT},
     * with or without {@code nullOnEmpty}.
     *<p>
     * The effects are to be the same as if the supplied sequence were passed
     * as {@code $EXPR} to {@code document{$EXPR}}.
     */
    private static SQLXML returnContent(
        Iterator<XdmItem> x, boolean nullOnEmpty)
    throws SQLException, SaxonApiException, XPathException
    {
        if ( nullOnEmpty  &&  ! x.hasNext() )
            return null;

        SQLXML rsx = s_dbc.createSQLXML();
        /*
         * Keep this simple by requesting a specific type of Result rather
         * than letting PL/Java choose. It happens (though this is a detail of
         * the implementation) that SAXResult won't be a bad choice.
         */
        SAXResult sr = rsx.setResult(SAXResult.class);
        /*
         * Michael Kay recommends the following as equivalent to the SQL/XML-
         * mandated behavior of evaluating document{$x}.
         * https://sourceforge.net/p/saxon/mailman/message/36969060/
         */
        SAXDestination d = new SAXDestination(sr.getHandler());
        Receiver r = d.getReceiver(
            s_s9p.getUnderlyingConfiguration().makePipelineConfiguration(),
            new SerializationProperties());
        r.open();
        while ( x.hasNext() )
            r.append(x.next().getUnderlyingValue());
        r.close();
        return rsx;
    }

    /**
     * An implementation of (much of) XMLTABLE, using genuine XML Query.
     *<p>
     * The {@code columns} array must supply a valid XML Query expression for
     * every column in the column definition list that follows the call of this
     * function in SQL, except that the column for ordinality, if wanted, is
     * identified by a {@code null} entry in {@code columns}. Syntax sugar in
     * the standard allows an omitted column expression to imply an element test
     * for an element with the same name as the column; that doesn't work here.
     *<p>
     * For now, this implementation lacks the ability to specify defaults for
     * when a column expression produces an empty sequence. It is possible to
     * do defaults explicitly by rewriting a query expression <em>expr</em> as
     * {@code let $e := }<em>expr</em>{@code return if(empty($e))then $D else $e}
     * and supplying the default <em>D</em> as another query parameter, though
     * such defaults will be evaluated only once when {@code xmltable} is called
     * and will not be able to refer to other values in an output row.
     * @param rows The single XQuery expression whose result sequence generates
     * the rows of the resulting table. Must not be null.
     * @param columns Array of XQuery expressions, exactly as many as result
     * columns in the column definition list that follows the SQL call to this
     * function. This array must not be null. It is allowed for one element (and
     * no more than one) to be null, marking the corresponding column to be
     * "FOR ORDINALITY" (the column must be of "exact numeric with scale zero"
     * type; PostgreSQL supports 64-bit row counters, so {@code int8} is
     * recommended).
     * @param passing A row value whose columns will be supplied to the query
     * as parameters, just as described for
     * {@link #xq_ret_content xq_ret_content()}. If a context item is supplied,
     * it is the context item for the {@code rows} query (the {@code columns}
     * queries get their context item from the {@code rows} query's result). Any
     * named parameters supplied here are available both in the {@code rows}
     * expression and (though this goes beyond the standard) in every expression
     * of {@code columns}, with their values unchanging from row to row.
     * @param namespaces An even-length String array where, of each pair of
     * consecutive entries, the first is a namespace prefix and the second is
     * to URI to which to bind it, just as described for
     * {@link #xq_ret_content xq_ret_content()}.
     * @param base64 whether the effective, in-scope 'xmlbinary' setting calls
     * for base64 or (the default, false) hexadecimal.
     */
    @Function(
        implementor="saxon9api",
        schema="javatest",
        onNullInput=CALLED,
        settings="IntervalStyle TO iso_8601"
    )
    public static ResultSetProvider xmltable(
        String rows, String[] columns,
        @SQLType(defaultValue={}) ResultSet passing,
        @SQLType(defaultValue={}) String[] namespaces,
        @SQLType(defaultValue="false") Boolean base64)
        throws SQLException
    {
        if ( null == rows )
            throw new SQLDataException(
                "XMLTABLE row expression may not be null", "22004");

        if ( null == columns )
            throw new SQLDataException(
                "XMLTABLE columns expression array may not be null", "22004");

        if ( null == base64 )
            throw new SQLDataException(
                "XMLTABLE base64 parameter may not be null", "22004");
        XMLBinary enc = base64 ? XMLBinary.BASE64 : XMLBinary.HEX;

        Binding.Assemblage rowBindings =
            new BindingsFromResultSet(passing, true);

        Iterable<Map.Entry<String,String>> namespacepairs =
            namespaceBindings(namespaces);

        XQueryEvaluator[] columnXQEs = new XQueryEvaluator[ columns.length ];
        SequenceType[] columnStaticTypes = new SequenceType[ columns.length ];

        try
        {
            XQueryCompiler rowXQC = createStaticContextWithPassedTypes(
                rowBindings, namespacepairs);

            XQueryExecutable rowXQX = rowXQC.compile(rows);

            Binding.Assemblage columnBindings =
                new BindingsFromXQX(rowXQX, rowBindings);

            XQueryCompiler columnXQC = createStaticContextWithPassedTypes(
                columnBindings, namespacepairs);

            boolean ordinalitySeen = false;
            for ( int i = 0; i < columns.length; ++ i )
            {
                String expr = columns[i];
                if ( null == expr )
                {
                    if ( ordinalitySeen )
                        throw new SQLSyntaxErrorException(
                            "No more than one column expression may be null " +
                            "(=> \"for ordinality\")", "42611");
                    ordinalitySeen = true;
                    continue;
                }
                XQueryExecutable columnXQX = columnXQC.compile(expr);
                columnStaticTypes[i] = makeSequenceType(
                    columnXQX.getResultItemType(),
                    columnXQX.getResultCardinality());
                columnXQEs[i] = columnXQX.load();
                storePassedValuesInDynamicContext(
                    columnXQEs[i], columnBindings, false);
            }

            XQueryEvaluator rowXQE = rowXQX.load();
            XdmSequenceIterator<XdmItem> rowIterator;
            if ( storePassedValuesInDynamicContext(rowXQE, rowBindings, true) )
                rowIterator = (XdmSequenceIterator<XdmItem>)
                    XdmEmptySequence.getInstance().iterator();
            else
                rowIterator = rowXQE.iterator();
            return new S9(rowIterator, columnXQEs, columnStaticTypes, enc);
        }
        catch ( SaxonApiException | XPathException e )
        {
            throw new SQLException(e.getMessage(), "10000", e);
        }
    }

    /**
     * Called when PostgreSQL has no need for more rows of the tabular result.
     */
    @Override
    public void close()
    {
        m_sequenceIterator.close();
    }

    /**
     * <a id='assignRowValues'>Produce and return one row</a> of
     * the {@code XMLTABLE} result table per call.
     *<p>
     * The row expression has already been compiled and its evaluation begun,
     * producing a sequence iterator. The column XQuery expressions have all
     * been compiled and are ready to evaluate, and the compiler's static
     * analysis has bounded the data types they will produce. Because of the
     * way the set-returning function protocol works, we don't know the types
     * of the SQL output columns yet, until the first call of this function,
     * when the {@code receive} parameter's {@code ResultSetMetaData} can be
     * inspected to find out. So that will be the first thing done when called
     * with {@code currentRow} of zero.
     *<p>
     * Each call will then: (a) get the next value from the row expression's
     * sequence iterator, then for each column, (b) evaluate that column's
     * XQuery expression on the row value, and (c) assign that column's result
     * to the SQL output column, casting to the proper type (which the SQL/XML
     * spec has very exacting rules on how to do).
     *<p>
     * A note before going any further: this implementation, while fairly
     * typical of a PostgreSQL set-returning user function, is <em>not</em> the
     * way the SQL/XML spec defines {@code XMLTABLE}. The official behavior of
     * {@code XMLTABLE} is defined in terms of a rewriting, at the SQL level,
     * into a much-expanded SQL query where each result column appears as an
     * {@code XMLQUERY} call applying the column expression, wrapped in an
     * {@code XMLCAST} to the result column type (with a
     * {@code CASE WHEN XMLEXISTS} thrown in to support column defaults).
     *<p>
     * As an ordinary user function, this example cannot rely on any fancy
     * query rewriting during PostgreSQL's parse analysis. The slight syntax
     * desugaring needed to transform a standard {@code XMLTABLE} call into a
     * call of this "xmltable" is not too hard to learn and do by hand, but no
     * one would ever want to write out by hand the whole longwinded "official"
     * expansion prescribed in the spec. So this example is a compromise.
     *<p>
     * The main thing lost in the compromise is the handling of column defaults.
     * The full rewriting with per-column SQL expressions means that each
     * column default expression can be evaluated exactly when/if needed, which
     * is often the desired behavior. This implementation as an ordinary
     * function, whose arguments all get evaluated ahead of the call, can't
     * really do that. Otherwise, there's nothing in the spec that's inherently
     * unachievable in this implementation.
     *<p>
     * Which brings us to the matter of casting each column expression result
     * to the proper type for its SQL result column.
     *<p>
     * Like any spec, {@code SQL/XML} does not mandate that an implementation
     * must be done in exactly the way presented in the spec (rewritten so each
     * column value is produced by an {@code XMLQUERY} wrapped in an
     * {@code XMLCAST}). The requirement is to produce the equivalent result.
     *<p>
     * A look at the rewritten query shows that each column XQuery result value
     * must be representable as some value in SQL's type system, not once, but
     * twice: first as the result returned by {@code XMLQUERY} and passed along
     * to {@code XMLCAST}, and finally with the output column's type as the
     * result of the {@code XMLCAST}.
     *<p>
     * Now, the output column type can be whatever is wanted. Importantly, it
     * can be either an XML type, or any ordinary SQL scalar type, like a
     * {@code float} or a {@code date}. Likewise, the XQuery column expression
     * may have produced some atomic value (like an {@code xs:double} or
     * {@code xs:date}), or some XML node, or any sequence of any of those.
     *<p>
     * What are the choices for the type in the middle: the SQL value returned
     * by {@code XMLQUERY} and passed on to {@code XMLCAST}?
     *<p>
     * There are two. An ISO-standard SQL {@code XMLQUERY} can specify
     * {@code RETURNING SEQUENCE} or {@code RETURNING CONTENT}. The first option
     * produces the type {@code XML(SEQUENCE)}, a useful type that PostgreSQL
     * does not currently have. {@code XML(SEQUENCE)} can hold exactly whatever
     * an XQuery expression can produce: a sequence of any length, of any
     * mixture of atomic values and XML nodes (even such oddities as attribute
     * nodes outside of any element), in any order. An {@code XML(SEQUENCE)}
     * value need not look anything like what "XML" normally brings to mind.
     *<p>
     * With the other option, {@code RETURNING CONTENT}, the result of
     * {@code XMLQUERY} has to be something that PostgreSQL's {@code xml} type
     * could store: a serialized document with XML structure, but without the
     * strict requirements of exactly one root element with no text outside it.
     * At the limit, a completely non-XMLish string of ordinary text is
     * perfectly acceptable XML {@code CONTENT}, as long as it uses the right
     * {@code &...;} escapes for any characters that could look like XML markup.
     *<p>
     * {@code XMLCAST} is able to accept either form as input, and deliver it
     * to the output column as whatever type is needed. But the spec leaves no
     * wiggle room as to which form to use:
     *<ul>
     *<li>If the result column type is {@code XML(SEQUENCE)}, then the
     * {@code XMLQUERY} is to specify {@code RETURNING SEQUENCE}. It produces
     * the column's result type directly, so the {@code XMLCAST} has nothing
     * to do.
     *<li>In every other case (<em>every</em> other case), the {@code XMLQUERY}
     * is to specify {@code RETURNING CONTENT}.
     *</ul>
     *<p>
     * At first blush, that second rule should sound crazy. Imagine a column
     * definition like
     *<pre>
     * growth float8 PATH 'math:pow(1.0 + $RATE, count(year))'
     *</pre>
     * The expression produces an {@code xs:double}, which can be assigned
     * directly to a PostgreSQL {@code float8}, but the rule in the spec will
     * have it first converted to a decimal string representation, made into
     * a text node, wrapped in a document node, and returned as XML, to be
     * passed along to {@code XMLCAST}, which parses it, discards the wrapping
     * document node, parses the text content as a double, and returns that as
     * a proper value of the result column type (which, in this example, it
     * already is).
     *<p>
     * The spec does not go into why this rule was chosen. The only rationale
     * that makes sense to me is that the {@code XML(SEQUENCE)} data type
     * is an SQL feature (X190) that not every implementation will support,
     * so the spec has to define {@code XMLTABLE} using a rewritten query that
     * can work on systems that do not have that type. (PostgreSQL itself, at
     * present, does not have it.)
     *<p>
     * The first rule, when {@code XML(SEQUENCE)} is the result column type,
     * will naturally never be in play except on a system that has that type, in
     * which case it can be used directly. But even such a system must still
     * produce, in all other cases, results that match what a system without
     * that type would produce. All those cases are therefore defined as if
     * going the long way through {@code XML(CONTENT)}.
     *<p>
     * Whenever the XQuery expression can be known to produce a (possibly empty
     * or) singleton sequence of an atomic type, the long round trip can be
     * shown to be idempotent, and we can skip right to casting the atomic type
     * to the SQL result column type. A few other cases could be short-circuited
     * the same way. But in general, for cases involving nodes or non-singleton
     * sequences, it is safest to follow the spec punctiliously; the steps are
     * defined in terms of XQuery constructs like {@code document {...}} and
     * {@code data()}, which have specs of their own with many traps for the
     * unwary, and the XQuery library provides implementations of them that are
     * already tested and correct.
     *<p>
     * Though most of the work can be done by the XQuery library, it may be
     * helpful to look closely at just what the specification entails.
     *<p>
     * Again, but for the case of an {@code XML(SEQUENCE)} result column, in all
     * other cases the result must pass through
     * {@code XMLQUERY(... RETURNING CONTENT EMPTY ON EMPTY)}. That, in turn, is
     * defined as equivalent to {@code XMLQUERY(... RETURNING SEQUENCE)} with
     * the result then passed to {@code XMLDOCUMENT(... RETURNING CONTENT)},
     * whose behavior is that of a
     * <a href='https://www.w3.org/TR/xquery-31/#id-documentConstructors'>
     * document node constructor</a> in XQuery, with
     * <a href='https://www.w3.org/TR/xquery-31/#dt-construction-mode'>
     * construction mode</a> {@code preserve}. The first step of that behavior
     * is the same as Step 1e in the processing of
     * <a href='https://www.w3.org/TR/xquery-31/#id-content'>direct element
     * constructor content</a>. The remaining steps are those laid out for the
     * document node constructor.
     *<p>
     * Clarity demands flattening this nest of specifications into a single
     * ordered list of the steps to apply:
     *<ul>
     *<li>Any item in the sequence that is an array is flattened (its elements
     * become items in the sequence).
     *<li>If any item is a function, {@code err:XQTY0105} is raised.
     *<li>Any sequence {@code $s} of adjacent atomic values is replaced by
     * {@code string-join($s, ' ')}.
     *<li>Any XML node in the sequence is copied (as detailed in the spec).
     *<li>After all the above, any document node that may exist in the resulting
     * sequence is flattened (replaced by its children).
     *<li>A single text node is produced for any run of adjacent text nodes in
     * the sequence (including any that have newly become adjacent by the
     * flattening of document nodes), by concatenation with no separator (unlike
     * the earlier step where atomic values were concatenated with a space as
     * the separator).
     *<li>If the sequence directly contains any attribute or namespace node,
     * {@code err:XPTY0004} is raised. <b>More on this below.</b>
     *<li>The sequence resulting from the preceding steps is wrapped in one
     * new document node (as detailed in the spec).
     *</ul>
     *<p>
     * At this point, the result could be returned to SQL as a value of
     * {@code XML(CONTENT(ANY))} type, to be passed to an {@code XMLCAST}
     * invocation. This implementation avoids that, and simply proceeds with the
     * existing Java in-memory representation of the document tree, to the
     * remaining steps entailed in an {@code XMLCAST} to the output column type:
     *<ul>
     *<li>If the result column type is an XML type, rewriting would turn the
     * {@code XMLCAST} into a simple {@code CAST} and that's that. Otherwise,
     * the result column has some non-XML, SQL type, and:
     *<li>The algorithm "Removing XQuery document nodes from an XQuery sequence"
     * is applied. By construction, we know the only such node is the one the
     * whole sequence was recently wrapped in, two steps ago (you get your
     * house back, you get your dog back, you get your truck back...).
     *<li>That sequence of zero or more XML nodes is passed to the
     *<a href='https://www.w3.org/TR/xpath-functions-31/#func-data'>fn:data</a>
     * function, producing a sequence of zero or more atomic values, which will
     * all have type {@code xs:untypedAtomic} (because the document-wrapping
     * stringified any original atomic values and wrapped them in text nodes,
     * for which the
     * <a href='https://www.w3.org/TR/xpath-datamodel-31/#acc-summ-typed-value'>
     * typed-value</a> is {@code xs:untypedAtomic} by definition). This sequence
     * also has cardinality zero-or-more, and may be shorter or longer than the
     * original.
     *<li>If the sequence is empty, the result column is assigned {@code NULL}
     * (or the column's default value, if one was specified). Otherwise, the
     * sequence is known to have length one or more, and:
     *<li>The spec does not say this (which may be an oversight or bug), but the
     * sequence must be checked for length greater than one, raising
     * {@code err:XPTY0004} in that case. The following steps require it to be a
     * singleton.
     *<li>It is labeled as a singleton sequence of {@code xs:anyAtomicType} and
     * used as input to an XQuery {@code cast as} expression. (Alternatively, it
     * could be labeled a one-or-more sequence of {@code xs:anyAtomicType},
     * leaving the length check to be done by {@code cast as}, which would raise
     * the same error {@code err:XPTY0004}, if longer than one.)
     *<li>The {@code cast as} is to the XQuery type determined as in
     * {@code determineXQueryFormalType} below, based on the SQL type of the
     * result column; or, if the SQL type is a date/time type with no time zone,
     * there is a first {@code cast as} to a specific XSD date/time type, which
     * is (if it has a time zone) first adjusted to UTC, then stripped of its
     * time zone, followed by a second {@code cast as} from that type to the one
     * determined from the result column type. Often, that will be the same type
     * as was used for the time zone adjustment, and the second {@code cast as}
     * will have nothing to do.
     *<li>The XQuery value resulting from the cast is converted and assigned to
     * the SQL-typed result column, a step with many details but few surprises,
     * therefore left for the morbidly curious to explore in the code. The flip
     * side of the time zone removal described above happens here: if the SQL
     * column type expects a time zone and the incoming value lacks one, it is
     * given a zone of UTC.
     *</ul>
     *<p>
     * The later steps above, those following the length-one check, are
     * handled by {@code xmlCastAsNonXML} below.
     *<p>
     * The earlier steps, from the start through the {@code XMLCAST} early steps
     * of document-node unwrapping, can all be applied by letting the original
     * result sequence be {@code $EXPR} in the expression:
     *<pre>
     * declare construction preserve;
     * data(document { $EXPR } / child::node())
     *</pre>
     * which may seem a bit of an anticlimax after seeing how many details lurk
     * behind those tidy lines of code.
     *<p>
     * <strong>About bare attribute nodes</strong>
     *<p>
     * One consequence of the rules above deserves special attention.
     * Consider something like:
     *<pre>
     * XMLTABLE('.' PASSING '&lt;a foo="bar"/&gt;' COLUMNS c1 VARCHAR PATH 'a/@foo');
     *</pre>
     *<p>
     * The result of the column expression is an XML attribute node all on its
     * own, with name {@code foo} and value {@code bar}, not enclosed in any
     * XML element. In the data type {@code XML(SEQUENCE)}, an attribute node
     * can appear standalone like that, but not in {@code XML(CONTENT)}.
     *<p>
     * Db2, Oracle, and even the XPath-based pseudo-XMLTABLE built into
     * PostgreSQL, will all accept that query and produce the result "bar".
     *<p>
     * However, a strict interpretation of the spec cannot produce that result,
     * because the result column type ({@code VARCHAR}) is not
     * {@code XML(SEQUENCE)}, meaning the result must be as if passed through
     * {@code XMLDOCUMENT(... RETURNING CONTENT)}, and the XQuery
     * {@code document { ... }} constructor is required to raise
     * {@code err:XPTY0004} upon encountering any bare attribute node. The
     * apparently common, convenient behavior of returning the attribute node's
     * value component is not, strictly, conformant.
     *<p>
     * This implementation will raise {@code err:XPTY0004}. That can be avoided
     * by simply wrapping any such bare attribute in {@code data()}:
     *<pre>
     * ... COLUMNS c1 VARCHAR PATH 'a/data(@foo)');
     *</pre>
     *<p>
     * It is possible the spec has an editorial mistake and did not intend to
     * require an error for this usage, in which case this implementation can
     * be changed to match a future clarification of the spec.
     */
    @Override
    public boolean assignRowValues(ResultSet receive, long currentRow)
    throws SQLException
    {
        if ( 0 == currentRow )
        {
            m_outBindings = new BindingsFromResultSet(receive, m_columnXQEs);
            int i = -1;
            AtomizingFunction atomizer = null;
            for ( Binding.Parameter p : m_outBindings )
            {
                SequenceType staticType = m_columnStaticTypes [ ++ i ];
                /*
                 * A null in m_columnXQEs identifies the ORDINALITY column,
                 * if any. Assign nothing to m_atomize[i], it won't be used.
                 */
                if ( null == m_columnXQEs [ i ] )
                    continue;

                if ( Types.SQLXML == p.typeJDBC() )
                    continue;

                /*
                 * Ok, the output column type is non-XML; choose an atomizer,
                 * either a simple identity if the result type is statically
                 * known to be zero-or-one atomic, or the long way through the
                 * general-purpose one. If the type is statically known to be
                 * the empty sequence (weird, but not impossible), the identity
                 * atomizer suffices and we're on to the next column.
                 */
                OccurrenceIndicator occur = staticType.getOccurrenceIndicator();
                if ( OccurrenceIndicator.ZERO == occur )
                {
                    m_atomize [ i ] = (v, col) -> v;
                    continue;
                }

                /* So, it isn't known to be empty. If the column
                 * expression type isn't known to be atomic, or isn't known to
                 * be zero-or-one, then the general-purpose atomizer--a trip
                 * through data(document { ... } / child::node())--must be used.
                 * This atomizer will definitely produce a sequence of length
                 * zero or one, raising XPTY0004 otherwise. So the staticType
                 * can be replaced by xs:anyAtomicType?. xmlCastAsNonXML will
                 * therefore be passed xs:anyAtomicType, as in the spec.
                 *    BUT NO ... Saxon is more likely to find a converter from
                 * xs:untypedAtomic than from xs:anyAtomicType.
                 */
                ItemType itemType = staticType.getItemType();
                if ( occur.allowsMany()
                    || ! ItemType.ANY_ATOMIC_VALUE.subsumes(itemType)
                    /*
                     * The following tests may be punctilious to a fault. If we
                     * have a bare Saxon atomic type of either xs:base64Binary
                     * or xs:hexBinary type, Saxon will happily and successfully
                     * convert it to a binary string; but if we have the same
                     * thing as a less-statically-determinate type that we'll
                     * put through the atomizer, the conversion will fail unless
                     * its encoding matches the m_xmlbinary setting. That could
                     * seem weirdly unpredictable to a user, so we'll just
                     * (perversely) disallow the optimization (which would
                     * succeed) in the cases where the specified, unoptimized
                     * behavior would be to fail.
                     */
                    || ItemType.HEX_BINARY.subsumes(itemType)
                        && (XMLBinary.HEX != m_xmlbinary)
                    || ItemType.BASE64_BINARY.subsumes(itemType)
                        && (XMLBinary.BASE64 != m_xmlbinary)
                   )
                {
                    if ( null == atomizer )
                    {
                        XQueryEvaluator docWrapUnwrap = PredefinedQueryHolders
                            .DocumentWrapUnwrap.INSTANCE.load();
                        atomizer = (v, col) ->
                        {
                            docWrapUnwrap.setExternalVariable(
                                PredefinedQueryHolders.s_qEXPR, v);
                            v = docWrapUnwrap.evaluate();
                            /*
                             * It's already zero-or-one, or XPTY0004 was thrown
                             */
                            return v;
                        };
                    }
                    m_atomize [ i ] = atomizer;
                    /*
                     * The spec wants anyAtomicType below instead of
                     * untypedAtomic. But Saxon's getConverter is more likely
                     * to fail to find a converter from anyAtomicType to an
                     * arbitrary type, than from untypedAtomic. So use that.
                     */
                    m_columnStaticTypes [ i ] = s_01untypedAtomic;
                }
                else
                {
                    /*
                     * We know we'll be getting zero-or-one atomic value, so
                     * the atomizing function can be the identity.
                     */
                    m_atomize [ i ] = (v, col) -> v;
                }
            }
        }

        if ( ! m_sequenceIterator.hasNext() )
            return false;

        ++ currentRow; // for use as 1-based ordinality column

        XdmItem it = m_sequenceIterator.next();

        int i = 0;
        for ( Binding.Parameter p : m_outBindings )
        {
            XQueryEvaluator xqe = m_columnXQEs [ i ];
            AtomizingFunction atomizer = m_atomize [ i ];
            SequenceType staticType = m_columnStaticTypes [ i++ ];

            if ( null == xqe )
            {
                receive.updateLong( i, currentRow);
                continue;
            }

            try
            {
                xqe.setContextItem(it);

                if ( null == atomizer ) /* => result type was found to be XML */
                {
                    receive.updateSQLXML(
                        i, returnContent(xqe.iterator(), false));
                    continue;
                }

                XdmValue x1 = xqe.evaluate();
                x1 = atomizer.apply(x1, i);

                /*
                 * The value is now known to be atomic and either exactly
                 * one or zero-or-one. May as well just use size() to see if
                 * it's empty.
                 */
                if ( 0 == x1.size() )
                {
                    receive.updateNull(i); // XXX Handle defaults some day
                    continue;
                }
                XdmAtomicValue av = (XdmAtomicValue)x1.itemAt(0);
                xmlCastAsNonXML(
                    av, staticType.getItemType(), p, receive, i, m_xmlbinary);
            }
            catch ( SaxonApiException | XPathException e )
            {
                throw new SQLException(e.getMessage(), "10000", e);
            }
        }
        return true;
    }

    /**
     * Store the values of any passed parameters and/or context item into the
     * dynamic context, returning true if the overall query should
     * short-circuit and return null.
     *<p>
     * The specification requires the overall query to return null if a
     * context item is specified in the bindings and its value is null.
     * @param xqe XQuery evaluator into which to store the values.
     * @param passing The bindings whose values should be installed.
     * @param setContextItem True to handle the context item, if present in the
     * bindings. False to skip any processing of the context item, in cases
     * where the caller will handle that.
     * @return True if the overall query's return should be null, false if the
     * query should proceed to evaluation.
     */
    private static boolean storePassedValuesInDynamicContext(
        XQueryEvaluator xqe, Binding.Assemblage passing, boolean setContextItem)
        throws SQLException, SaxonApiException
    {
        /*
         * Is there or is there not a context item?
         */
        if ( ! setContextItem  ||  null == passing.contextItem() )
        {
            /* "... there is no context item in XDC." */
        }
        else
        {
            Object cve = passing.contextItem().valueJDBC();
            if ( null == cve )
                return true;
            XdmValue ci;
            if ( cve instanceof XdmNode ) // XXX support SEQUENCE input someday
            {
                ci = (XdmNode)cve;
            }
            else
                ci = xmlCastAsSequence(
                    cve, XMLBinary.HEX, passing.contextItem().typeXS());
            switch ( ci.size() )
            {
            case 0:
                /* "... there is no context item in XDC." */
                break;
            case 1:
                xqe.setContextItem(ci.itemAt(0));
                break;
            default:
                throw new SQLDataException(
                    "invalid XQuery context item", "2200V");
            }
        }

        /*
         * For each <XML query variable> XQV:
         */
        for ( Binding.Parameter p : passing )
        {
            String name = p.name();
            Object v = p.valueJDBC();
            XdmValue vv;
            if ( null == v )
                vv = XdmEmptySequence.getInstance();
            else if ( v instanceof XdmNode ) // XXX support SEQUENCE someday
            {
                vv = (XdmNode)v;
            }
            else
                vv = xmlCastAsSequence(
                    v, XMLBinary.HEX, p.typeXS().getItemType());
            xqe.setExternalVariable(new QName(name), vv);
        }

        return false;
    }

    /**
     * Return a s9api {@link XQueryCompiler XQueryCompiler} with static context
     * preconfigured as the Syntax Rules dictate.
     * @param pt The single-row ResultSet representing the passed parameters
     * and context item, if any.
     * @param nameToIndex A Map, supplied empty, that on return will map
     * variable names for the dynamic context to column indices in {@code pt}.
     * If a context item was supplied, its index will be entered in the map
     * with the null key.
     */
    private static XQueryCompiler createStaticContextWithPassedTypes(
        Binding.Assemblage pt, Iterable<Map.Entry<String,String>> namespaces)
        throws SQLException, XPathException
    {
        XQueryCompiler xqc = s_s9p.newXQueryCompiler();
        xqc.declareNamespace(
            "sqlxml", "http://standards.iso.org/iso9075/2003/sqlxml");
        // https://sourceforge.net/p/saxon/mailman/message/20318550/ :
        xqc.declareNamespace("xdt", W3C_XML_SCHEMA_NS_URI);

        for ( Map.Entry<String,String> e : namespaces )
            xqc.declareNamespace(e.getKey(), e.getValue());

        /*
         * This business of predeclaring global external named variables
         * is not an s9api-level advertised ability in Saxon, hence the
         * various getUnderlying.../getStructured... methods here to access
         * the things that make it happen.
         */
        StaticQueryContext sqc = xqc.getUnderlyingStaticContext();

        for ( Binding.Parameter p : pt )
        {
            String name = p.name();
            int ct = p.typeJDBC();
            assertCanCastAsXmlSequence(ct, name);
            SequenceType st = p.typeXS();
            sqc.declareGlobalVariable(
                new QName(name).getStructuredQName(),
                st.getUnderlyingSequenceType(), null, true);
        }

        /*
         * Apply syntax rules to the context item, if any.
         */
        Binding.ContextItem ci = pt.contextItem();
        if ( null != ci )
        {
            int ct = ci.typeJDBC();
            assertCanCastAsXmlSequence(ct, "(context item)");
            ItemType it = ci.typeXS();
            xqc.setRequiredContextItemType(it);
        }

        return xqc;
    }

    /**
     * Check that something's type is "convertible to XML(SEQUENCE)
     * according to the Syntax Rules of ... <XML cast specification>."
     * That turns out not to be a very high bar; not much is excluded
     * by those rules except collection, row, structured, or
     * reference typed <value expression>s.
     * @param jdbcType The {@link Types JDBC type} to be checked.
     * @param what A string to include in the exception message if the
     * check fails.
     * @throws SQLException if {@code jdbcType} is one of the prohibited types.
     */
    private static void assertCanCastAsXmlSequence(int jdbcType, String what)
    throws SQLException
    {
        if ( Types.ARRAY == jdbcType || Types.STRUCT == jdbcType
            || Types.REF == jdbcType )
            throw new SQLSyntaxErrorException(
                "The type of \"" + what + "\" is not suitable for " +
                "XMLCAST to XML(SEQUENCE).", "42804");
    }

    /**
     * The "determination of an XQuery formal type notation" algorithm.
     *<p>
     * This is relied on for parameters and context items passed to
     * {@code XMLQUERY} and therefore, {@code XMLTABLE} (and also, in the spec,
     * {@code XMLDOCUMENT} and {@code XMLPI}). Note that it does <em>not</em>
     * take an {@code XMLBinary} parameter, but rather imposes hexadecimal form
     * unconditionally, so in the contexts where this is called, any
     * {@code xmlbinary} setting is ignored.
     * @param b a {@code Binding} from which the JDBC type can be retrieved
     * @param forContextItem whether the type being derived is for a context
     * item or (if false) for a named parameter.
     * @return a {@code SequenceType} (always a singleton in the
     * {@code forContextItem} case)
     */
    private static SequenceType determineXQueryFormalType(
        Binding b, boolean forContextItem)
        throws SQLException
    {
        int sd = b.typeJDBC();
        OccurrenceIndicator suffix;
        /*
         * The SQL/XML standard uses a formal type notation straight out of
         * the XQuery 1.0 and XPath 2.0 Formal Semantics document, and that is
         * strictly more fine-grained and expressive than anything you can
         * actually say in the form of XQuery SequenceTypes. This method will
         * simply return the nearest approximation in the form of a sequence
         * type; some of the standard's distinct formal type notations will
         * collapse into the same SequenceType.
         *  That also means the various cases laid out in the standard will,
         * here, all simply assign some ItemType to 'it', and therefore the
         * tacking on of the occurrence suffix can be factored out for the
         * very end.
         */
        ItemType it;

        if ( forContextItem )
            suffix = OccurrenceIndicator.ONE;
        // else if sd is XML(SEQUENCE) - we don't have this type yet
        //  suffix = OccurrenceIndicator.ZERO_OR_MORE;
        /*
         * Go through the motions of checking isNullable, though PL/Java's JDBC
         * currently hardcodes columnNullableUnknown. Maybe someday it won't.
         */
        else if ( b.knownNonNull() )
            suffix = OccurrenceIndicator.ONE;
        else
            suffix = OccurrenceIndicator.ZERO_OR_ONE;

        // Define ET... for {DOCUMENT|CONTENT}(XMLSCHEMA) case ... not supported

        // if SD is XML(DOCUMENT(UNTYPED)) - not currently tracked, can't tell
        //  it = s_itf.getDocumentTest(item type for xdt:untyped);
        // else if SD is XML(DOCUMENT(ANY)) - not currently tracked, can't tell
        //  it = s_itf.getDocumentTest(item type for xs:anyType);
        // else if SD is XML(DOCUMENT(XMLSCHEMA)) - unsupported and can't tell
        //  it = s_itf.getDocumentTest(the ET... we didn't define earlier)
        // else if SD is XML(CONTENT(UNTYPED)) - which we're not tracking ...
        //  at s9api granularity, there's no test for this that's not same as:
        // else if SD is XML(CONTENT(ANY)) - which we must assume for ANY XML
        if ( Types.SQLXML == sd )
            it = s_itf.getNodeKindTest(DOCUMENT);
        // else if SD is XML(CONTENT(XMLSCHEMA)) - we don't track and can't tell
        //  at s9api granularity, there's no test that means this anyway.
        // else if SD is XML(SEQUENCE) - we really should have this type, but no
        //  it = it.ANY_ITEM
        else // it ain't XML, it's some SQL type
        {
            ItemType xmlt = mapSQLDataTypeToXMLSchemaDataType(
                b, XMLBinary.HEX, Nulls.ABSENT);
            // ItemType pt = xmlt.getUnderlyingItemType().getPrimitiveType()
            //  .somehowGetFromUnderlyingPTBackToS9apiPT() - ugh, the hard part
            /*
             * The intention here is to replace any derived type with the
             * primitive type it is based on, *except* for three types that are
             * technically derived: integer (from decimal), yearMonthDuration
             * and dayTimeDuration (from duration). Those are not replaced, so
             * they stand, as if they were honorary primitive types.
             *
             * For now, it's simplified greatly by mapSQLDataType... skipping
             * the construction of a whole derived XML Schema snippet, and just
             * returning the type we want anyway. Also, no need to dive under
             * the s9api layer to try to make getPrimitiveType work.
             */
            it = xmlt;
        }

        SequenceType xftn = makeSequenceType(it, suffix);
        return xftn;
    }

    @SuppressWarnings("fallthrough")
    private static ItemType mapSQLDataTypeToXMLSchemaDataType(
        Binding b, XMLBinary xmlbinary, Nulls nulls)
        throws SQLException
    {
        /*
         * Nearly all of the fussing about specified in the standard
         * for this method is to create XML Schema derived types that
         * accurately reflect the typmod information for the SQL type
         * in question. Then, in determineXQueryFormalType (the only
         * client of this method so far!), all of that is thrown away
         * and our painstakingly specified derived type is replaced with
         * the primitive type we based it on. That simplifies a lot. :)
         * For now, forget the derived XML Schema declarations, and just
         * return the primitive types they would be based on.
         *
         * The need for the nulls parameter vanishes if no XML Schema snippets
         * are to be generated.
         *
         * If the full XML Schema snippet generation ever proves to be
         * needed, one hacky way to get it would be with a SELECT
         * query_to_xmlschema('SELECT null::type-in-question', false, false,
         * '') where the same derivations are already implemented (though it
         * produces some different results; that work may have been done from
         * an earlier version of the standard).
         */
        switch ( b.typeJDBC() )
        {
        case Types.CHAR:
        case Types.VARCHAR:
        case Types.CLOB:
            return ItemType.STRING;

        case Types.BINARY:
        case Types.VARBINARY:
        case Types.BLOB:
            return XMLBinary.HEX == xmlbinary ?
                ItemType.HEX_BINARY : ItemType.BASE64_BINARY;

        case Types.NUMERIC:
        case Types.DECIMAL:
            /*
             * Go through the motions to get the scale and do this right,
             * though PL/Java's getScale currently hardcodes a -1 return.
             * Maybe someday it won't.
             */
            int scale = b.scale();
            return 0 == scale ? ItemType.INTEGER : ItemType.DECIMAL;

        case Types.INTEGER:
            return ItemType.INT;
        case Types.SMALLINT:
            return ItemType.SHORT;
        case Types.BIGINT:
            return ItemType.LONG;

        case Types.REAL:
            return ItemType.FLOAT; // could check P, MINEXP, MAXEXP here.
        case Types.FLOAT:
            assert false; // PG should always report either REAL or DOUBLE
            /*FALLTHROUGH*/
        case Types.DOUBLE:
            return ItemType.DOUBLE;

        case Types.BOOLEAN:
            return ItemType.BOOLEAN;

        case Types.DATE:
            return ItemType.DATE;

        case Types.TIME:
            return ItemType.TIME;

        case Types.TIME_WITH_TIMEZONE:
            return ItemType.TIME; // restrictive facet would make sense here

        case Types.TIMESTAMP:
            return ItemType.DATE_TIME;

        case Types.TIMESTAMP_WITH_TIMEZONE:
            return ItemType.DATE_TIME_STAMP; // xsd 1.1 equivalent of facet!

        // There's no JDBC Types.INTERVAL; handle it after switch

        // Good luck finding out from JDBC if it's a domain

        // PG doesn't have DISTINCT types per se

        // PL/Java's JDBC doesn't support PostgreSQL's arrays as ARRAY

        // PG doesn't seem to have multisets (JDBC doesn't grok them either)

        // Types.SQLXML we could recognize, but for determineFormalTypes it has
        // been handled already, and it's not yet clear what would be
        // appropriate to return (short of the specified XMLSchema snippet),
        // probably just document.

        // So punt all these for now; what hasn't been handled in this switch
        // can be handled specially after the switch falls through, and what
        // isn't, isn't supported just now.
        }

        String typeName = b.typePG();
        if ( "interval".equals(typeName) )
        {
            /*
             * XXX This isn't right yet; it needs to be refined to a
             * YEAR_MONTH_DURATION or a DAY_TIME_DURATION in the appropriate
             * cases, and for that it needs access to the typmod information
             * for the type, which getColumnTypeName doesn't now provide.
             */
            return ItemType.DURATION;
        }

        throw new SQLNonTransientException(String.format(
            "Mapping SQL type \"%s\" to XML type not supported", typeName),
            "0N000");
    }

    /**
     * Implement that portion of the {@code <XML cast>} specification where
     * the target data type is sequence, and (for now, anyway) the source is
     * not an XML type; the only caller, so far, handles that case separately.
     * @param v The SQL value to be cast (in the form of an Object from JDBC).
     * @param enc Whether binary values should be encoded in hex or base 64.
     * @param xst The formal static XS type derived from the SQL type of v.
     * @return An {@code XdmValue}, {@code null} if {@code v} is null.
     */
    private static XdmValue xmlCastAsSequence(
        Object v, XMLBinary enc, ItemType xst)
        throws SQLException
    {
        if ( null == v )
            return null;
        /*
         * What happens next in the standard is one of the most breathtaking
         * feats of obscurantism in the whole document. It begins, plausibly
         * enough, by using mapValuesOfSQLTypesToValuesOfXSTypes to produce
         * the lexical form of the XS type (but with XML metacharacters escaped,
         * if it's a string type). Then:
         * 1. That lexical form is to be fed to an XML parser, producing an
         *    XQuery document node that NEVER can be a well-formed document (it
         *    is expected to satisfy document { text ? } where the text node is
         *    just the lexical value form we started with, now with the escaped
         *    metacharacters unescaped again as a consequence of parsing). For
         *    some source types, mapValuesOfSQLTypesToValuesOfXSTypes can
         *    produce a string that parses to XML with element content: row
         *    types, arrays, multisets, XML. Clearly, those cases can't satisfy
         *    the formal type assumed here, and they are cases this routine
         *    won't be expected to handle: XML handled separately by the caller,
         *    arrays/structs/etc. being ruled out by assertCanCastAsXmlSequence.
         * 2. That document node is made the $TEMP parameter of an XML Query,
         *    '$TEMP cast as XSBTN' (where XSBTN is a QName for the result type
         *    chosen according to the rules) and the sequence resulting from
         *    that query is the result of the cast.
         *
         * Step (1) can only succeed if the XML parser doesn't insist on well-
         * formed documents, as the stock JRE parser emphatically does. And the
         * ultimate effect of that whole dance is that the cast in (2) casts a
         * document node to the target type, which means the document node gets
         * atomized, which, for a document node, means everything is thrown away
         * save the concatenated values of its descendant text nodes (or node,
         * in this case; haven't we seen that value somewhere before?), assigned
         * the type xs:untypedAtomic, and then that's operated on by the cast.
         *
         * Because this implementation's in PL/Java, the value v received here
         * has already been mapped from an SQL type to a Java type according to
         * JDBC's rules as PL/Java implements them, so there's one degree of
         * removal from the specified algorithm anyway. And the s9api
         * XdmAtomicValue already has constructors from several of the expected
         * Java types, as well as one taking a lexical form and explicit type.
         * Beause this is /example/ code, rather than slavishly implementing the
         * specified algorithm, it will assume that that is either roughly or
         * exactly equivalent to what these s9api constructors in fact do, and
         * just use them; conformance-testing code could then check for exact
         * equivalence if there's enough interest to write it.
         *
         * So, we will NOT start with this:
         *
         *   String xmlv = mapValuesOfSQLTypesToValuesOfXSTypes(
         *       v, enc, Nulls.ABSENT, true);
         *
         * Instead, we'll derive this type first ...
         */
        ItemType xsbt;
        // year-month interval type => xsbt = YEAR_MONTH_DURATION
        // day-time interval type => xsbt = DAY_TIME_DURATION
        xsbt = xst; // we have a winner!
        // xs non-built-in atomic type => xsbt = getPrimitiveType(ugh).

        /*
         * ... and then use this method instead:
         */
        try
        {
            return mapJDBCofSQLvalueToXdmAtomicValue(v, enc, xsbt);
        }
        catch ( SaxonApiException | XPathException e )
        {
            throw new SQLException(e.getMessage(), "10000", e);
        }
    }

    @FunctionalInterface
    interface CastingFunction
    {
        AtomicValue apply(AtomicValue v) throws XPathException;
    }

    @FunctionalInterface
    interface CasterSupplier
    {
        CastingFunction get() throws SQLException, XPathException;
    }

    @FunctionalInterface
    interface AtomizingFunction
    {
        /**
         * @param v sequence to be atomized
         * @param columnIndex only to include in exception if result has more
         * than one item
         */
        XdmValue apply(XdmValue v, int columnIndex)
        throws SaxonApiException, XPathException;
    }

    private static XPathException noPrimitiveCast(ItemType vt, ItemType xt)
    {
        return new XPathException(
            "Casting from " + vt.getTypeName() + " to " + xt.getTypeName() +
            " can never succeed", "XPTY0004");
    }

    /**
     * Handle the case of XMLCAST to a non-XML target type when the cast operand
     * is already a single atomic value.
     *<p>
     * The caller, if operating on a sequence, must itself handle the case of
     * an empty sequence (returning null, per General Rule 4c in :2011), or a
     * sequence of length greater than one (raising XPTY0004, which is not
     * specified in :2011, but the exclusion of such a sequence is implicit in
     * rules 4g and 4h; Db2 silently drops all but the first item, unlike
     * Oracle, which raises XPTY0004).
     * @param av The atomic operand value
     * @param p The parameter binding, recording the needed type information
     * @param rs ResultSet into which the value will be stored
     * @param col Index of the result column
     */
    private static void xmlCastAsNonXML(
        XdmAtomicValue av, ItemType vt,
        Binding.Parameter p, ResultSet rs, int col, XMLBinary enc)
        throws SQLException, XPathException
    {
        XdmAtomicValue bv;
        ItemType xt = p.typeXT(enc);

        CastingFunction caster = p.atomicCaster(vt, () ->
        {
            ConversionRules rules = vt.getConversionRules();
            Converter c1;
            ItemType t1;
            Converter c2;

            switch ( p.typeJDBC() )
            {
            case Types.TIMESTAMP:
                t1 = ItemType.DATE_TIME;
                break;
            case Types.TIME:
                t1 = ItemType.TIME;
                break;
            case Types.DATE:
                t1 = ItemType.DATE;
                break;
            default:
                c1 = rules.getConverter(
                    (AtomicType)vt.getUnderlyingItemType(),
                    (AtomicType)xt.getUnderlyingItemType());
                if ( null == c1 )
                    throw noPrimitiveCast(vt, xt);
                return (AtomicValue v) -> c1.convert(v).asAtomic();
            }
            /*
             * Nothing left here but the rest of the three date/timey cases
             * partly handled above.
             */
            c1 = rules.getConverter(
                (AtomicType)vt.getUnderlyingItemType(),
                (AtomicType)t1.getUnderlyingItemType());
            c2 = rules.getConverter(
                (AtomicType)t1.getUnderlyingItemType(),
                (AtomicType)xt.getUnderlyingItemType());
            if ( null == c1  ||  null == c2 )
                throw noPrimitiveCast(vt, xt);
            return (AtomicValue v) ->
            {
                v = c1.convert(v).asAtomic();
                v = ((CalendarValue)v).adjustTimezone(0).removeTimezone();
                return c2.convert(v).asAtomic();
            };
        });

        bv = makeAtomicValue(caster.apply(av.getUnderlyingValue()));

        if ( ItemType.STRING.subsumes(xt) )
            rs.updateString(col, bv.getStringValue());

        else if ( ItemType.HEX_BINARY.subsumes(xt) )
            rs.updateBytes(col,
                ((HexBinaryValue)bv.getUnderlyingValue()).getBinaryValue());
        else if ( ItemType.BASE64_BINARY.subsumes(xt) )
            rs.updateBytes(col,
                ((Base64BinaryValue)bv.getUnderlyingValue()).getBinaryValue());

        else if ( ItemType.DECIMAL.subsumes(xt) )
            rs.updateObject(col, bv.getValue());

        /*
         * The standard calls for throwing "data exception - numeric value out
         * of range" rather than forwarding a float or double inf, -inf, or nan
         * to SQL, but PostgreSQL supports those values, and these conversions
         * preserve them.
         *  Because of the collapsing in typeXT(), xt will never be FLOAT,
         * only DOUBLE. JDBC is supposed to handle assigning a double to a float
         * column, anyway.
         */
        else if ( ItemType.DOUBLE.subsumes(xt) )
            rs.updateObject(col, bv.getValue());

        else if ( ItemType.DATE.subsumes(xt) )
            rs.updateObject(col, bv.getLocalDate());
        else if ( ItemType.DATE_TIME.subsumes(xt) )
        {
            if ( ((CalendarValue)bv.getUnderlyingValue()).hasTimezone() )
                rs.updateObject(col, bv.getOffsetDateTime());
            else
            {
                LocalDateTime jv = bv.getLocalDateTime();
                rs.updateObject(col,
                    Types.TIMESTAMP_WITH_TIMEZONE == p.typeJDBC() ?
                        jv.atOffset(UTC) : jv);
            }
        }
        else if ( ItemType.TIME.subsumes(xt) ) // no handy tz/notz distinction
        {
            if ( ((CalendarValue)bv.getUnderlyingValue()).hasTimezone() )
                rs.updateObject(col, OffsetTime.parse(bv.getStringValue()));
            else
            {
                LocalTime jv = LocalTime.parse(bv.getStringValue());
                rs.updateObject(col,
                    Types.TIME_WITH_TIMEZONE == p.typeJDBC() ?
                        jv.atOffset(UTC) : jv);
            }
        }

        else if ( ItemType.YEAR_MONTH_DURATION.subsumes(xt) )
            rs.updateString(col, toggleIntervalRepr(bv.getStringValue()));
        else if ( ItemType.DAY_TIME_DURATION.subsumes(xt) )
            rs.updateString(col, toggleIntervalRepr(bv.getStringValue()));
        else if ( ItemType.DURATION.subsumes(xt) ) // need this case for now
            rs.updateString(col, toggleIntervalRepr(bv.getStringValue()));

        else if ( ItemType.BOOLEAN.subsumes(xt) )
            rs.updateObject(col, bv.getValue());
        else
            throw new SQLNonTransientException(String.format(
                "Mapping XML type \"%s\" to SQL value not supported", xt),
                "0N000");
    }

    /**
     * Like the "Mapping values of SQL data types to values of XML Schema
     * data types" algorithm, except after the SQL values have already been
     * converted to Java values according to JDBC rules.
     *<p>
     * Also, this uses Saxon s9api constructors for the XML Schema values, which
     * accept the Java types directly. As a consequence, where the target type
     * {@code xst} is {@code xs:hexBinary} or {@code xs:base64Binary}, that type
     * will be produced, regardless of the passed {@code encoding}. This might
     * not be strictly correct, but is probably safest until an oddity in the
     * spec can be clarified: {@code determineXQueryFormalType} will always
     * declare {@code xs:hexBinary} as the type for an SQL byte string, and it
     * would violate type safety to construct a value here that honors the
     * {@code encoding} parameter but isn't of the declared formal type.
     */
    private static XdmAtomicValue mapJDBCofSQLvalueToXdmAtomicValue(
        Object dv, XMLBinary encoding, ItemType xst)
        throws SQLException, SaxonApiException, XPathException
    {
        if ( ItemType.STRING.equals(xst) )
            return new XdmAtomicValue((String)dv);

        if ( ItemType.HEX_BINARY.equals(xst) )
            return makeAtomicValue(new HexBinaryValue((byte[])dv));
        if ( ItemType.BASE64_BINARY.equals(xst) )
            return makeAtomicValue(new Base64BinaryValue((byte[])dv));

        if ( ItemType.INTEGER.equals(xst) )
            return new XdmAtomicValue(((BigInteger)dv).toString(), xst);
        if ( ItemType.DECIMAL.equals(xst) )
            return new XdmAtomicValue((BigDecimal)dv);
        if ( ItemType.INT.equals(xst) )
            return new XdmAtomicValue((Integer)dv);
        if ( ItemType.SHORT.equals(xst) )
            return new XdmAtomicValue((Short)dv);
        if ( ItemType.LONG.equals(xst) )
            return new XdmAtomicValue((Long)dv);
        if ( ItemType.FLOAT.equals(xst) )
            return new XdmAtomicValue((Float)dv);
        if ( ItemType.DOUBLE.equals(xst) )
            return new XdmAtomicValue((Double)dv);

        if ( ItemType.BOOLEAN.equals(xst) )
            return new XdmAtomicValue((Boolean)dv);

        if ( ItemType.DATE.equals(xst) )
        {
            if ( dv instanceof LocalDate )
                return new XdmAtomicValue((LocalDate)dv);
            return new XdmAtomicValue(dv.toString(), xst);
        }

        if ( ItemType.TIME.equals(xst) )
            return new XdmAtomicValue(dv.toString(), xst);

        if ( ItemType.DATE_TIME.equals(xst) )
        {
            if ( dv instanceof LocalDateTime )
                return new XdmAtomicValue((LocalDateTime)dv);
            return new XdmAtomicValue(dv.toString(), xst);
        }

        if ( ItemType.DATE_TIME_STAMP.equals(xst) )
        {
            if ( dv instanceof OffsetDateTime )
                return new XdmAtomicValue((OffsetDateTime)dv);
            return new XdmAtomicValue(dv.toString(), xst);
        }

        if ( ItemType.DURATION.equals(xst) )
            return new XdmAtomicValue(toggleIntervalRepr((String)dv), xst);

        throw new SQLNonTransientException(String.format(
            "Mapping SQL value to XML type \"%s\" not supported", xst),
            "0N000");
    }

    /*
     * Toggle the lexical representation of an interval/duration between the
     * form PostgreSQL likes and the form XML Schema likes. Only negative values
     * are affected. Positive values are returned unchanged, as are those that
     * don't fit any expected form; those will probably be reported as malformed
     * by whatever tries to consume them.
     */
    static String toggleIntervalRepr(String lex)
    {
        Matcher m = s_intervalSigns.matcher(lex);
        if ( ! m.matches() )
            return lex; // it's weird, just don't touch it
        if ( -1 == m.start(1) )
        {
            if ( -1 != m.start(2)  &&  -1 == m.start(3) ) // it's PG negative
                return '-' + lex.replace("-", "");        // make it XS negative
        }
        else if ( -1 == m.start(2)  &&  -1 != m.start(3) )// it's XS negative
            return m.usePattern(s_intervalSignSite)       // make it PG negative
                .reset(lex.substring(1)).replaceAll("-");
        return lex; // it's either positive, or weird, just don't touch it
    }

    static Iterable<Map.Entry<String,String>> namespaceBindings(String[] nbs)
    throws SQLException
    {
        if ( 1 == nbs.length % 2 )
            throw new SQLSyntaxErrorException(
                "Namespace binding array must have even length", "42000");
        Map<String,String> m = new HashMap<>();

        for ( int i = 0; i < nbs.length; i += 2 )
        {
            String prefix = nbs[i];
            String uri = nbs[1 + i];

            if ( null == prefix  ||  null == uri )
                throw new SQLDataException(
                    "Namespace binding array elements must not be null",
                    "22004");

            if ( ! "".equals(prefix) )
            {
                if ( ! isValidNCName(prefix) )
                    throw new SQLSyntaxErrorException(
                        "Not an XML NCname: \"" + prefix + '"', "42602");
                if ( XML_NS_PREFIX.equals(prefix)
                    || XMLNS_ATTRIBUTE.equals(prefix) )
                    throw new SQLSyntaxErrorException(
                        "Namespace prefix may not be xml or xmlns", "42939");
                if ( XML_NS_URI.equals(uri)
                    || XMLNS_ATTRIBUTE_NS_URI.equals(uri) )
                    throw new SQLSyntaxErrorException(
                        "Namespace URI has a disallowed value", "42P17");
                if ( "".equals(uri) )
                    throw new SQLSyntaxErrorException(
                        "URI for non-default namespace may not be zero-length",
                        "42P17");
            }

            String was = m.put(prefix.intern(), uri.intern());

            if ( null != was )
                throw new SQLSyntaxErrorException(
                    "Namespace prefix \"" + prefix + "\" multiply bound (" +
                    "to \"" + was + "\" and \"" + uri + "\")", "42712");
        }

        return Collections.unmodifiableSet(m.entrySet());
    }

    static class Binding
    {
        String typePG() throws SQLException
        {
            if ( null != m_typePG )
                return m_typePG;
            return m_typePG = implTypePG();
        }

        int typeJDBC() throws SQLException
        {
            if ( null != m_typeJDBC )
                return m_typeJDBC;
            int tj = implTypeJDBC();
            /*
             * The JDBC types TIME_WITH_TIMEZONE and TIMESTAMP_WITH_TIMEZONE
             * first appear in JDBC 4.2 / Java 8. PL/Java's JDBC driver does
             * not yet return those values. As a workaround until it does,
             * recheck here using the PG type name string, if TIME or TIMESTAMP
             * is the JDBC type that the driver returned.
             *
             * Also for backward compatibility, the driver still returns
             * Types.OTHER for XML, rather than Types.SQLXML. Check and fix that
             * here too.
             */
            switch ( tj )
            {
            case Types.OTHER:
                if ( "xml".equals(typePG()) )
                    tj = Types.SQLXML;
                break;
            case Types.TIME:
                if ( "timetz".equals(typePG()) )
                    tj = Types.TIME_WITH_TIMEZONE;
                break;
            case Types.TIMESTAMP:
                if ( "timestamptz".equals(typePG()) )
                    tj = Types.TIMESTAMP_WITH_TIMEZONE;
                break;
            default:
            }
            return m_typeJDBC = tj;
        }

        Object valueJDBC() throws SQLException
        {
            if ( m_valueJDBCValid )
                return m_valueJDBC;
            /*
             * When JDBC 4.2 added support for the JSR 310 date/time types, for
             * back-compatibility purposes, it did not change what types a plain
             * getObject(...) would return for them, which could break existing
             * code. Instead, it's necessary to use the form of getObject that
             * takes a Class<?>, and ask for the new classes explicitly.
             *
             * Similarly, PL/Java up through 1.5.0 has always returned a String
             * from getObject for a PostgreSQL xml type. Here, the JDBC standard
             * provides that a SQLXML object should be returned, and that should
             * happen in a future major PL/Java release, but for now, the plain
             * getObject will still return String, so it is also necessary to
             * ask for the SQLXML type explicitly. In fact, we will ask for
             * XdmNode, as it might be referred to more than once (if a
             * parameter), and a SQLXML can't be read more than once, nor would
             * there be any sense in building an XdmNode from it more than once.
             */
            switch ( typeJDBC() )
            {
            case Types.DATE:
                return setValueJDBC(implValueJDBC(LocalDate.class));
            case Types.TIME:
                return setValueJDBC(implValueJDBC(LocalTime.class));
            case Types.TIME_WITH_TIMEZONE:
                return setValueJDBC(implValueJDBC(OffsetTime.class));
            case Types.TIMESTAMP:
                return setValueJDBC(implValueJDBC(LocalDateTime.class));
            case Types.TIMESTAMP_WITH_TIMEZONE:
                return setValueJDBC(implValueJDBC(OffsetDateTime.class));
            case Types.SQLXML:
                return setValueJDBC(implValueJDBC(XdmNode.class));
            default:
            }
            return setValueJDBC(implValueJDBC());
        }

        boolean knownNonNull() throws SQLException
        {
            if ( null != m_knownNonNull )
                return m_knownNonNull;
            return m_knownNonNull = implKnownNonNull();
        }

        int scale() throws SQLException
        {
            if ( null != m_scale )
                return m_scale;
            return m_scale = implScale();
        }

        static class ContextItem extends Binding
        {
            /**
             * Return the XML Schema type of this input binding for a context
             * item.
             *<p>
             * Because it is based on {@code determinXQueryFormalType}, this
             * method is not parameterized by {@code XMLBinary}, and will always
             * map a binary-string SQL type to {@code xs:hexBinary}.
             */
            ItemType typeXS() throws SQLException
            {
                if ( null != m_typeXS )
                    return m_typeXS;
                SequenceType st = implTypeXS(true);
                assert OccurrenceIndicator.ONE == st.getOccurrenceIndicator();
                return m_typeXS = st.getItemType();
            }

            protected ItemType m_typeXS;
        }

        static class Parameter extends Binding
        {
            String name()
            {
                return m_name;
            }

            SequenceType typeXS() throws SQLException
            {
                if ( null != m_typeXS )
                    return m_typeXS;
                return m_typeXS = implTypeXS(false);
            }

            /**
             * Return the XML Schema type collapsed according to the Syntax Rule
             * deriving {@code XT} for {@code XMLCAST}.
             *<p>
             * The intent of the rule is unclear, but it involves collapsing
             * certain sets of more-specific types that {@code typeXS} might
             * return into common supertypes, for use only in an intermediate
             * step of {@code xmlCastAsNonXML}. Unlike {@code typeXS}, this
             * method must be passed an {@code XMLBinary} parameter reflecting
             * the hex/base64 choice currently in scope.
             * @param enc whether to use {@code xs:hexBinary} or
             * {@code xs:base64Binary} as the XML Schema type corresponding to a
             * binary-string SQL type.
             */
            ItemType typeXT(XMLBinary enc) throws SQLException
            {
                throw new UnsupportedOperationException(
                    "typeXT() on synthetic binding");
            }

            /**
             * Memoize and return a casting function from a given
             * {@code ItemType} to the type of this parameter.
             *<p>
             * Used only by {@code xmlCastAsNonXML}, which does all the work
             * of constructing the function; this merely allows it to be
             * remembered, if many casts to the same output parameter will be
             * made (as by {@code xmltable}).
             */
            CastingFunction atomicCaster(ItemType it, CasterSupplier s)
            throws SQLException, XPathException
            {
                throw new UnsupportedOperationException(
                    "atomicCaster() on synthetic binding");
            }

            protected SequenceType m_typeXS;

            private final String m_name;

            /**
             * @param name The SQL name of the parameter
             * @param checkName True if the name must be a valid NCName (as for
             * an input parameter from SQL to the XML query context), or false
             * if the name doesn't matter (as when it describes a result, or the
             * sole input value of an XMLCAST.
             * @throws SQLException if the name of a checked input parameter
             * isn't a valid NCName.
             */
            protected Parameter(String name, boolean checkName)
            throws SQLException
            {
                if ( checkName  &&  ! isValidNCName(name) )
                    throw new SQLSyntaxErrorException(
                        "Not an XML NCname: \"" + name + '"', "42602");
                m_name = name;
            }
        }

        protected String m_typePG;
        protected Integer m_typeJDBC;
        protected Boolean m_knownNonNull;
        protected Integer m_scale;
        private Object m_valueJDBC;
        private boolean m_valueJDBCValid;
        protected Object setValueJDBC(Object v)
        {
            m_valueJDBCValid = true;
            return m_valueJDBC = v;
        }

        protected String implTypePG() throws SQLException
        {
            throw new UnsupportedOperationException(
                "typePG() on synthetic binding");
        }

        protected int implTypeJDBC() throws SQLException
        {
            throw new UnsupportedOperationException(
                "typeJDBC() on synthetic binding");
        }

        protected boolean implKnownNonNull() throws SQLException
        {
            throw new UnsupportedOperationException(
                "knownNonNull() on synthetic binding");
        }

        protected int implScale() throws SQLException
        {
            throw new UnsupportedOperationException(
                "scale() on synthetic binding");
        }

        protected Object implValueJDBC() throws SQLException
        {
            throw new UnsupportedOperationException(
                "valueJDBC() on synthetic binding");
        }

        /*
         * This implementation just forwards to the type-less version, then
         * fails if that did not return the wanted type. Override if a smarter
         * behavior is possible.
         */
        protected <T> T implValueJDBC(Class<T> type) throws SQLException
        {
            return type.cast(implValueJDBC());
        }

        protected SequenceType implTypeXS(boolean forContextItem)
        throws SQLException
        {
            return determineXQueryFormalType(this, forContextItem);
        }

        static class Assemblage implements Iterable<Parameter>
        {
            ContextItem contextItem() { return m_contextItem; }

            @Override
            public Iterator<Parameter> iterator()
            {
                return m_params.iterator();
            }

            protected ContextItem m_contextItem;
            protected Collection<Parameter> m_params = Collections.emptyList();
        }
    }

    static class BindingsFromResultSet extends Binding.Assemblage
    {
        /**
         * Construct the bindings from a ResultSet representing input parameters
         * to an XML query.
         * @param rs ResultSet representing the input parameters. Column names
         * "." and "?COLUMN?" are treated specially, and used to supply the
         * query's context item; every other column name must be a valid NCName,
         * and neither any named parameter nor the context item may be mentioned
         * more than once.
         * @param checkNames True if the input parameter names matter (a name of
         * "." or "?COLUMN?" will define the context item, and any other name
         * must be a valid NCName); false to skip such checking (as for the
         * single input value to XMLCAST, whose name doesn't matter).
         * @throws SQLException if names are duplicated or invalid.
         */
        BindingsFromResultSet(ResultSet rs, boolean checkNames)
        throws SQLException
        {
            m_resultSet = rs;
            m_rsmd = rs.getMetaData();

            int nParams = m_rsmd.getColumnCount();
            ContextItem contextItem = null;
            Map<String,Binding.Parameter> n2b = new HashMap<>();

            if ( 0 < nParams )
                m_dBuilder = s_s9p.newDocumentBuilder();

            for ( int i = 1; i <= nParams; ++i )
            {
                String label = m_rsmd.getColumnLabel(i);
                if ( checkNames  &&
                    ("?COLUMN?".equals(label)  ||  ".".equals(label)) )
                {
                    if ( null != contextItem )
                    throw new SQLSyntaxErrorException(
                        "Context item supplied more than once (at " +
                        contextItem.m_idx + " and " + i + ')', "42712");
                    contextItem = new ContextItem(i);
                    continue;
                }

                Parameter was =
                    (Parameter)n2b.put(
                        label, new Parameter(label, i, checkNames));
                if ( null != was )
                    throw new SQLSyntaxErrorException(
                        "Name \"" + label + "\" duplicated at positions " +
                        was.m_idx + " and " + i, "42712");
            }

            m_contextItem = contextItem;
            m_params = n2b.values();
        }

        /**
         * Construct the bindings from a ResultSet representing output
         * parameters (as from XMLTABLE).
         * @param rs ResultSet representing the result parameters. Names have
         * no particular significance and are not subject to any checks.
         * @param exprs Compiled evaluators for the supplied column expressions.
         * The number of these must match the number of columns in {@code rs}.
         * One of these (and no more than one; the caller will have enforced
         * that) is allowed to be null, making the corresponding column
         * "FOR ORDINALITY". An ordinality column will be checked to ensure it
         * has an SQL type that is (ahem) "exact numeric with scale 0 (zero)."
         * May be null if this is some other general-purpose output result set,
         * not for an XMLTABLE.
         * @throws SQLException if numbers of columns and expressions don't
         * match, or there is an ordinality column and its type is not suitable.
         */
        @SuppressWarnings("fallthrough")
        BindingsFromResultSet(ResultSet rs, XQueryEvaluator[] exprs)
        throws SQLException
        {
            m_resultSet = rs;
            m_rsmd = rs.getMetaData();

            int nParams = m_rsmd.getColumnCount();
            if ( null != exprs  &&  nParams != exprs.length )
                throw new SQLSyntaxErrorException(
                    "Not as many supplied column expressions as output columns",
                    "42611");

            Binding.Parameter[] ps = new Binding.Parameter[ nParams ];

            for ( int i = 1; i <= nParams; ++i )
            {
                String label = m_rsmd.getColumnLabel(i);
                Parameter p = new Parameter(label, i, false);
                ps [ i - 1 ] = p;
                if ( null != exprs  &&  null == exprs [ i - 1 ] )
                {
                    switch ( p.typeJDBC() )
                    {
                    case Types.INTEGER:
                    case Types.SMALLINT:
                    case Types.BIGINT:
                        break;
                    case Types.NUMERIC:
                    case Types.DECIMAL:
                        int scale = p.scale();
                        if ( 0 == scale  ||  -1 == scale )
                            break;
                        /*FALLTHROUGH*/
                    default:
                        throw new SQLSyntaxErrorException(
                            "Column FOR ORDINALITY must have an exact numeric" +
                            " type with scale zero.", "42611");
                    }
                }
            }

            m_params = asList(ps);
        }

        private ResultSet m_resultSet;
        private ResultSetMetaData m_rsmd;
        DocumentBuilder m_dBuilder;

        <T> T typedValueAtIndex(int idx, Class<T> type) throws SQLException
        {
            if ( XdmNode.class != type )
                return m_resultSet.getObject(idx, type);
            try
            {
                SQLXML sx = m_resultSet.getObject(idx, SQLXML.class);
                return type.cast(
                    m_dBuilder.build(sx.getSource((Class<Source>)null)));
            }
            catch ( SaxonApiException e )
            {
                throw new SQLException(e.getMessage(), "10000", e);
            }
        }

        class ContextItem extends Binding.ContextItem
        {
            final int m_idx;

            ContextItem(int index) { m_idx = index; }

            protected String implTypePG() throws SQLException
            {
                return m_rsmd.getColumnTypeName(m_idx);
            }

            protected int implTypeJDBC() throws SQLException
            {
                return m_rsmd.getColumnType(m_idx);
            }

            protected int implScale() throws SQLException
            {
                return m_rsmd.getScale(m_idx);
            }

            protected Object implValueJDBC() throws SQLException
            {
                return m_resultSet.getObject(m_idx);
            }

            protected <T> T implValueJDBC(Class<T> type) throws SQLException
            {
                return typedValueAtIndex(m_idx, type);
            }
        }

        class Parameter extends Binding.Parameter
        {
            final int m_idx;
            private ItemType m_typeXT;
            private CastingFunction m_atomCaster;
            private ItemType m_lastCastFrom;

            Parameter(String name, int index, boolean isInput)
            throws SQLException
            {
                super(name, isInput);
                m_idx = index;
            }

            @Override
            ItemType typeXT(XMLBinary enc) throws SQLException
            {
                if ( null != m_typeXT )
                    return m_typeXT;

                ItemType it =
                    mapSQLDataTypeToXMLSchemaDataType(this, enc, Nulls.ABSENT);
                if ( ! ItemType.ANY_ATOMIC_VALUE.subsumes(it) )
                    return m_typeXT = it;

                if ( it.equals(ItemType.INTEGER) )
                {
                    int tj = typeJDBC();
                    if ( Types.NUMERIC == tj || Types.DECIMAL == tj )
                        it = ItemType.DECIMAL;
                }
                else if ( ItemType.INTEGER.subsumes(it) )
                    it = ItemType.INTEGER;
                else if ( ItemType.FLOAT.subsumes(it) )
                    it = ItemType.DOUBLE;
                else if ( ItemType.DATE_TIME_STAMP.subsumes(it) )
                    it = ItemType.DATE_TIME;

                return m_typeXT = it;
            }

            @Override
            CastingFunction atomicCaster(ItemType it, CasterSupplier s)
            throws SQLException, XPathException
            {
                if ( null == m_atomCaster || ! it.equals(m_lastCastFrom) )
                {
                    m_atomCaster = s.get();
                    m_lastCastFrom = it;
                }
                return m_atomCaster;
            }

            protected String implTypePG() throws SQLException
            {
                return m_rsmd.getColumnTypeName(m_idx);
            }

            protected int implTypeJDBC() throws SQLException
            {
                return m_rsmd.getColumnType(m_idx);
            }

            protected boolean implKnownNonNull() throws SQLException
            {
                return columnNoNulls == m_rsmd.isNullable(m_idx);
            }

            protected int implScale() throws SQLException
            {
                return m_rsmd.getScale(m_idx);
            }

            protected Object implValueJDBC() throws SQLException
            {
                return m_resultSet.getObject(m_idx);
            }

            protected <T> T implValueJDBC(Class<T> type) throws SQLException
            {
                return typedValueAtIndex(m_idx, type);
            }
        }
    }

    static class BindingsFromXQX extends Binding.Assemblage
    {
        /**
         * Construct a new assemblage of bindings for the static context of an
         * XMLTABLE column expression. It will have the same named-parameter
         * bindings passed to the row expression, but the static type of the
         * context item will be the result type of the row expression. The
         * {@code ContextItem} in this assemblage will have no associated value;
         * the caller is responsible for retrieving that from the row evaluator
         * and storing it in the column expression context every iteration.
         * @param xqx The result of compiling the row expression; its
         * compiler-determined static result type will be used as the static
         * context item type.
         * @param params The bindings supplied to the row expression. Its named
         * parameters will be copied as the named parameters here.
         */
        BindingsFromXQX(XQueryExecutable xqx, Binding.Assemblage params)
        {
            m_params = params.m_params;
            m_contextItem = new ContextItem(xqx.getResultItemType());
        }

        static class ContextItem extends Binding.ContextItem
        {
            ContextItem(ItemType it)
            {
                m_typeXS = it;
                /*
                 * There needs to be a dummy JDBC type to return when queried
                 * for purposes of assertCanCastAsXmlSequence. It can literally
                 * be any type outside of the few that method rejects. Because
                 * the XS type is already known, nothing else will need to ask
                 * for this, or care.
                 */
                m_typeJDBC = Types.OTHER;
            }
        }
    }

    /*
     * The XQuery-regular-expression-based functions added in 9075-2:2006.
     *
     * For each function below, a parameter is marked //strict if the spec
     * explicitly says the result is NULL when that parameter is NULL. The
     * parameters not marked //strict (including the non-standard w3cNewlines
     * added here) all have non-null defaults, so by executive decision, these
     * functions will all get the onNullInput=RETURNS_NULL treatment, so none of
     * the null-checking has to be done here. At worst, that may result in a
     * mystery NULL return rather than an error, if someone explicitly passes
     * NULL to one of the parameters with a non-null default.
     */

    /*
     * Check valid range of 'from' and supported 'usingOctets'.
     *
     * Every specified function that has a start position FROM and a USING
     * clause starts with a check that the start position is in range. This
     * function factors out that test, returning true if the start position is
     * /out of range/ (triggering the caller to return the special result
     * defined for that case), returning false if the value is in range, or
     * throwing an exception if the length unit specified in the USING clause
     * isn't supported.
     */
    private static boolean usingAndLengthCheck(
        String in, int from, boolean usingOctets, String function)
    throws SQLException
    {
        if ( usingOctets )
            throw new SQLFeatureNotSupportedException(
                '"' + function + "\" does not yet support USING OCTETS",
                "0A000");
        return ( 1 > from  ||  from > getStringLength(in) );
    }

    private static void newlinesCheck(boolean w3cNewlines, String function)
    throws SQLException
    {
        if ( ! w3cNewlines )
            throw new SQLFeatureNotSupportedException(
                '"' + function + "\" does not yet support the ISO SQL newline" +
                " conventions, only the original W3C XQuery ones" +
                " (HINT: pass w3cNewlines => true)", "0A000");
    }

    private static RegularExpression compileRE(String pattern, String flags)
    throws SQLException
    {
        try
        {
            return s_s9p.getUnderlyingConfiguration()
                .compileRegularExpression(pattern, flags, "XP30", null);
        }
        catch ( XPathException e )
        {
            if ( NamespaceConstant.ERR.equals(e.getErrorCodeNamespace()) )
            {
                if ( "FORX0001".equals(e.getErrorCodeLocalPart()) )
                    throw new SQLDataException(
                        "invalid XQuery option flag", "2201T", e);
                if ( "FORX0002".equals(e.getErrorCodeLocalPart()) )
                    throw new SQLDataException(
                        "invalid XQuery regular expression", "2201S", e);
            }
            throw new SQLException(
                "compiling XQuery regular expression: " + e.getMessage(), e);
        }
    }

    private static CharSequence replace(
        RegularExpression re, CharSequence in, CharSequence with)
        throws SQLException
    {
        /*
         * Report the standard-mandated error if replacing a zero-length match.
         * Strictly speaking, this is a test of the length of the match, not of
         * the input string. Here, though, this private method is only called by
         * translate_regex, which always passes only the portion of the input
         * string that matched, so the test is equivalent.
         *  As to why the SQL committee would make such a point of disallowing
         * replacement of a zero-length match, that's a good question. See
         * s_intervalSignSite in this very file for an example where replacing
         * a zero-length match is just what's wanted. (But that pattern relies
         * on lookahead/lookbehind operators, which XQuery regular expressions
         * don't have.)
         *  When the underlying library is Saxon, there is an Easter egg: if a
         * regular expression is compiled with a 'flags' string ending in ";j",
         * a Java regular expression is produced instead of an XQuery one (with
         * standards conformance cast to the wind). That can be detected with
         * getFlags() on the regular expression: not looking for ";j", because
         * that has been stripped out, but for "d" which is a Java regex flag
         * that Saxon sets by default, and is not a valid XQuery regex flag.
         *  If the caller has used Saxon's Easter egg to get a Java regex, here
         * is another Easter egg to go with it, allowing zero-length matches
         * to be replaced if that's what the caller wants to do.
         */
        if ( 0 == in.length()  &&  ! re.getFlags().contains("d") )
            throw new SQLDataException(
                "attempt to replace a zero-length string", "2201U");
        try
        {
            return re.replace(in, with);
        }
        catch ( XPathException e )
        {
            if ( NamespaceConstant.ERR.equals(e.getErrorCodeNamespace()) )
            {
                if ( "FORX0003".equals(e.getErrorCodeLocalPart()) )
                    throw new SQLDataException(
                        "attempt to replace a zero-length string", "2201U", e);
                if ( "FORX0004".equals(e.getErrorCodeLocalPart()) )
                    throw new SQLDataException(
                        "invalid XQuery replacement string", "2201V", e);
            }
            throw new SQLException(
                "replacing regular expression match: " + e.getMessage(), e);
        }
    }

    interface MatchVector
    {
        int groups();
        int position(int group);
        int length(int group);
    }

    interface ListOfMatchVectors
    {
        /**
         * Return the MatchVector for one occurrence of a match.
         *<p>
         * Any previously-returned MatchVector is invalid after another get.
         * In multiple calls to get, the occurrence parameter must be strictly
         * increasing.
         * After get has returned null, it should not be called again.
         */
        MatchVector get(int occurrence) throws SQLException;
        void close();
    }

    static class LOMV
    implements ListOfMatchVectors, MatchVector, RegexIterator.MatchHandler
    {
        private RegexIterator m_ri;
        private int m_pos;
        private int m_occurrence;

        LOMV(int startPos, RegexIterator ri)
        {
            m_ri = ri;
            m_pos = startPos;
        }

        static ListOfMatchVectors of(
            String pattern, String flags, String in, int from)
            throws SQLException
        {
            RegularExpression re = compileRE(pattern, flags);
            return of(re, in, from);
        }

        static ListOfMatchVectors of(RegularExpression re, String in, int from)
        {
            RegexIterator ri =
                re.analyze(in.substring(in.offsetByCodePoints(0, from - 1)));
            return new LOMV(from, ri);
        }

        private int[] m_begPositions;
        private int[] m_endPositions;

        @Override // ListOfMatchVectors
        public MatchVector get(int occurrence) throws SQLException
        {
            try
            {
                StringValue sv;
                for ( ;; )
                {
                    sv = m_ri.next();
                    if ( null == sv )
                        return null;
                    if ( m_ri.isMatching() )
                        if ( ++ m_occurrence == occurrence )
                            break;
                    m_pos += sv.getStringLength();
                }

                if ( null == m_begPositions )
                {
                    int groups = m_ri.getNumberOfGroups();
                    /*
                     * Saxon's Apache-derived XQuery engine will report a number
                     * of groups counting $0 (so it will be 1 even if no capture
                     * groups were defined in the expression). In contrast, the
                     * Java regex engine that you get with the Saxon ";j" Easter
                     * egg does not count $0 (so arrays need groups+1 entries).
                     * It's hard to tell from here which flavor was used, plus
                     * the Saxon behavior might change some day, so just spend
                     * the extra + 1 every time.
                     */
                    m_begPositions = new int [ groups + 1 ];
                    m_endPositions = new int [ groups + 1 ];
                }

                m_begPositions [ 0 ] = m_pos;

                fill(m_begPositions, 1, m_begPositions.length, 0);
                fill(m_endPositions, 1, m_endPositions.length, 0);
                m_ri.processMatchingSubstring(this);

                m_endPositions [ 0 ] = m_pos;

                return this;
            }
            catch ( XPathException e )
            {
                throw new SQLException(
                    "evaluating XQuery regular expression: " + e.getMessage(),
                    e);
            }
        }

        @Override
        public void close()
        {
            m_ri.close();
        }

        @Override // MatchVector
        public int groups()
        {
            return m_begPositions.length - 1;
        }

        @Override
        public int position(int groupNumber)
        {
            return m_begPositions [ groupNumber ];
        }

        @Override
        public int length(int groupNumber)
        {
            return
                m_endPositions [ groupNumber ] - m_begPositions [ groupNumber ];
        }

        @Override // MatchHandler
        public void characters(CharSequence s)
        {
            m_pos += getStringLength(s);
        }

        @Override
        public void onGroupStart(int groupNumber)
        {
            m_begPositions [ groupNumber ] = m_pos;
        }

        @Override
        public void onGroupEnd(int groupNumber)
        {
            m_endPositions [ groupNumber ] = m_pos;
        }
    }

    /**
     * Function form of the ISO SQL
     * <a id='like_regex'>{@code <regex like predicate>}</a>.
     *<p>
     * Rewrite the standard form
     *<pre>
     * value LIKE_REGEX pattern FLAG flags
     *</pre>
     * into this form:
     *<pre>
     * like_regex(value, pattern, flag =&gt; flags)
     *</pre>
     * where the {@code flag} parameter defaults to no flags if omitted.
     *<p>
     * The SQL standard specifies that pattern elements sensitive to newlines
     * (namely {@code ^}, {@code $}, {@code \s}, {@code \S}, and {@code .}) are
     * to support the various representations of newline set out in
     * <a href='http://www.unicode.org/reports/tr18/#RL1.6'>Unicode Technical
     * Standard #18, RL1.6</a>. That behavior differs from the standard W3C
     * XQuery newline handling, as described for
     * <a href='https://www.w3.org/TR/xpath-functions-31/#flags'>the flags
     * {@code m} and {@code s}</a> and for
     * <a href='https://www.w3.org/TR/xmlschema11-2/#cces-mce'>the
     * multicharacter escapes {@code \s} and {@code \S}</a>. As an extension to
     * ISO SQL, passing {@code w3cNewlines => true} requests the standard W3C
     * XQuery behavior rather than the UTS#18 behevior for newlines. If the
     * underlying XQuery library only provides the W3C behavior, calls without
     * {@code w3cNewlines => true} will throw exceptions.
     * @param value The string to be tested against the pattern.
     * @param pattern The XQuery regular expression.
     * @param flag Optional string of
     * <a href='https://www.w3.org/TR/xpath-functions-31/#flags'>flags adjusting
     * the regular expression behavior</a>.
     * @param w3cNewlines Pass true to allow the regular expression to recognize
     * newlines according to the W3C XQuery rules rather than those of ISO SQL.
     * @return True if the supplied value matches the pattern. Null if any
     * parameter is null.
     * @throws SQLException SQLDataException with SQLSTATE 2201S if the regular
     * expression is invalid, 2201T if the flags string is invalid;
     * SQLFeatureNotSupportedException (0A000) if (in the current
     * implementation) w3cNewlines is false or omitted.
     */
    @Function(implementor="saxon9api", schema="javatest")
    public static boolean like_regex(
        String value,                          //strict
        String pattern,                        //strict
        @SQLType(defaultValue="") String flag, //strict
        @SQLType(defaultValue="false") boolean w3cNewlines
    )
        throws SQLException
    {
        newlinesCheck(w3cNewlines, "like_regex");
        return compileRE(pattern, flag).containsMatch(value);
    }

    /**
     * Syntax-sugar-free form of the ISO SQL
     * <a id='occurrences_regex'>{@code OCCURRENCES_REGEX}</a> function:
     * how many times does a pattern occur in a string?
     *<p>
     * Rewrite the standard form
     *<pre>
     * OCCURRENCES_REGEX(pattern FLAG flags IN str FROM position USING units)
     *</pre>
     * into this form:
     *<pre>
     * occurrences_regex(pattern, flag =&gt; flags, "in" =&gt; str,
     *                   "from" =&gt; position, usingOctets =&gt; true|false)
     *</pre>
     * where all of the named parameters are optional except pattern and "in",
     * and the standard {@code USING CHARACTERS} becomes
     * {@code usingOctets => false}, which is the default, and
     * {@code USING OCTETS} becomes {@code usingOctets => true}. See also
     * {@link #like_regex like_regex} regarding the {@code w3cNewlines}
     * parameter.
     * @param pattern XQuery regular expression to seek in the input string.
     * @param in The input string.
     * @param flag Optional string of
     * <a href='https://www.w3.org/TR/xpath-functions-31/#flags'>flags adjusting
     * the regular expression behavior</a>.
     * @param from Starting position in the input string, 1 by default.
     * @param usingOctets Whether position is counted in characters (actual
     * Unicode characters, not any smaller encoded unit, not even Java char),
     * which is the default, or (when true) in octets of the string's encoded
     * form.
     * @param w3cNewlines Pass true to allow the regular expression to recognize
     * newlines according to the W3C XQuery rules rather than those of ISO SQL.
     * @return The number of occurrences of the pattern in the input string,
     * starting from the specified position. Null if any parameter is null; -1
     * if the start position is less than 1 or beyond the end of the string.
     * @throws SQLException SQLDataException with SQLSTATE 2201S if the regular
     * expression is invalid, 2201T if the flags string is invalid;
     * SQLFeatureNotSupportedException (0A000) if (in the current
     * implementation) usingOctets is true, or w3cNewlines is false or omitted.
     */
    @Function(implementor="saxon9api", schema="javatest")
    public static int occurrences_regex(
        String pattern,                        //strict
        @SQLType(name="\"in\"") String in,     //strict
        @SQLType(defaultValue="") String flag, //strict
        @SQLType(name="\"from\"", defaultValue="1") int from,
        @SQLType(defaultValue="false") boolean usingOctets,
        @SQLType(defaultValue="false") boolean w3cNewlines
    )
        throws SQLException
    {
        if ( usingAndLengthCheck(in, from, usingOctets, "occurrences_regex") )
            return -1; // note: not the same as in position_regex!
        newlinesCheck(w3cNewlines, "occurrences_regex");

        ListOfMatchVectors lomv = LOMV.of(pattern, flag, in, from);

        for ( int i = 1 ;; ++ i )
            if ( null == lomv.get(i) )
                return i - 1;
    }

    /**
     * Syntax-sugar-free form of the ISO SQL
     * <a id='position_regex'>{@code POSITION_REGEX}</a> function:
     * where does a pattern, or part of it, occur in a string?
     *<p>
     * Rewrite the standard forms
     *<pre>
     * POSITION_REGEX(START pattern FLAG flags IN str FROM position
     *                OCCURRENCE n GROUP m)
     * POSITION_REGEX(AFTER pattern FLAG flags IN str FROM position
     *                OCCURRENCE n GROUP m)
     *</pre>
     * into these forms, respectively:
     *<pre>
     * position_regex(pattern, flag =&gt; flags, "in" =&gt; str,
     *                "from" =&gt; position, occurrence =&gt; n,
     *                "group" =&gt; m)
     * position_regex(pattern, flag =&gt; flags, "in" =&gt; str,
     *                "from" =&gt; position, occurrence =&gt; n,
     *                "group" =&gt; m, after =&gt; true)
     *</pre>
     * where all of the named parameters are optional except pattern and "in".
     * See also {@link #occurrences_regex occurrences_regex} regarding the
     * {@code usingOctets} parameter, and {@link #like_regex like_regex}
     * regarding {@code w3cNewlines}.
     * @param pattern XQuery regular expression to seek in the input string.
     * @param in The input string.
     * @param flag Optional string of
     * <a href='https://www.w3.org/TR/xpath-functions-31/#flags'>flags adjusting
     * the regular expression behavior</a>.
     * @param from Starting position in the input string, 1 by default.
     * @param usingOctets Whether position is counted in characters (actual
     * Unicode characters, not any smaller encoded unit, not even Java char),
     * which is the default, or (when true) in octets of the string's encoded
     * form.
     * @param after Whether to return the position where the match starts
     * (when false, the default), or just after the match ends (when true).
     * @param occurrence If specified as an integer n (default 1), returns the
     * position starting (or after) the nth match of the pattern in the string.
     * @param group If zero (the default), returns the position starting (or
     * after) the match of the whole pattern overall, otherwise if an integer m,
     * the position starting or after the mth parenthesized group in (the nth
     * occurrence of) the pattern.
     * @param w3cNewlines Pass true to allow the regular expression to recognize
     * newlines according to the W3C XQuery rules rather than those of ISO SQL.
     * @return The position, in the specified units, starting or just after,
     * the nth occurrence (or mth capturing group of the nth occurrence) of the
     * pattern in the input string, starting from the specified position. Null
     * if any parameter is null; zero if the start position is less than 1 or
     * beyond the end of the string, if occurrence is less than 1 or greater
     * than the number of matches, or if group is less than zero or greater than
     * the number of parenthesized capturing groups in the pattern.
     * @throws SQLException SQLDataException with SQLSTATE 2201S if the regular
     * expression is invalid, 2201T if the flags string is invalid;
     * SQLFeatureNotSupportedException (0A000) if (in the current
     * implementation) usingOctets is true, or w3cNewlines is false or omitted.
     */
    @Function(implementor="saxon9api", schema="javatest")
    public static int position_regex(
        String pattern,                                         //strict
        @SQLType(name="\"in\"") String in,                      //strict
        @SQLType(defaultValue="") String flag,                  //strict
        @SQLType(name="\"from\"", defaultValue="1") int from,
        @SQLType(defaultValue="false") boolean usingOctets,
        @SQLType(defaultValue="false") boolean after,
        @SQLType(defaultValue="1") int occurrence,              //strict
        @SQLType(name="\"group\"", defaultValue="0") int group, //strict
        @SQLType(defaultValue="false") boolean w3cNewlines
    )
        throws SQLException
    {
        if ( 1 > occurrence )
            return 0;
        if ( 0 > group ) // test group > ngroups after compiling regex
            return 0;
        if ( usingAndLengthCheck(in, from, usingOctets, "position_regex") )
            return 0; // note: not the same as in occurrences_regex!
        newlinesCheck(w3cNewlines, "position_regex");

        ListOfMatchVectors lomv = LOMV.of(pattern, flag, in, from);

        MatchVector mv = lomv.get(occurrence);
        if ( null == mv  ||  mv.groups() < group )
            return 0;

        return mv.position(group) + (after ? mv.length(group) : 0);
    }

    /**
     * Syntax-sugar-free form of the ISO SQL
     * <a id='substring_regex'>{@code SUBSTRING_REGEX}</a> function:
     * return a substring specified by a pattern match in a string.
     *<p>
     * Rewrite the standard form
     *<pre>
     * SUBSTRING_REGEX(pattern FLAG flags IN str FROM position
     *                 OCCURRENCE n GROUP m)
     *</pre>
     * into this form:
     *<pre>
     * substring_regex(pattern, flag =&gt; flags, "in" =&gt; str,
     *                 "from" =&gt; position, occurrence =&gt; n,
     *                 "group" =&gt; m)
     *</pre>
     * where all of the named parameters are optional except pattern and "in".
     * See also {@link #position_regex position_regex} regarding the
     * {@code occurrence} and {@code "group"} parameters,
     * {@link #occurrences_regex occurrences_regex} regarding
     * {@code usingOctets}, and {@link #like_regex like_regex}
     * regarding {@code w3cNewlines}.
     * @param pattern XQuery regular expression to seek in the input string.
     * @param in The input string.
     * @param flag Optional string of
     * <a href='https://www.w3.org/TR/xpath-functions-31/#flags'>flags adjusting
     * the regular expression behavior</a>.
     * @param from Starting position in the input string, 1 by default.
     * @param usingOctets Whether position is counted in characters (actual
     * Unicode characters, not any smaller encoded unit, not even Java char),
     * which is the default, or (when true) in octets of the string's encoded
     * form.
     * @param occurrence If specified as an integer n (default 1), returns the
     * nth match of the pattern in the string.
     * @param group If zero (the default), returns the match of the whole
     * pattern overall, otherwise if an integer m, the match of the mth
     * parenthesized group in (the nth occurrence of) the pattern.
     * @param w3cNewlines Pass true to allow the regular expression to recognize
     * newlines according to the W3C XQuery rules rather than those of ISO SQL.
     * @return The substring matching the nth occurrence (or mth capturing group
     * of the nth occurrence) of the pattern in the input string, starting from
     * the specified position. Null if any parameter is null, if the start
     * position is less than 1 or beyond the end of the string, if occurrence is
     * less than 1 or greater than the number of matches, or if group is less
     * than zero or greater than the number of parenthesized capturing groups in
     * the pattern.
     * @throws SQLException SQLDataException with SQLSTATE 2201S if the regular
     * expression is invalid, 2201T if the flags string is invalid;
     * SQLFeatureNotSupportedException (0A000) if (in the current
     * implementation) usingOctets is true, or w3cNewlines is false or omitted.
     */
    @Function(implementor="saxon9api", schema="javatest")
    public static String substring_regex(
        String pattern,                                          //strict
        @SQLType(name="\"in\"") String in,                       //strict
        @SQLType(defaultValue="") String flag,                   //strict
        @SQLType(name="\"from\"", defaultValue="1") int from,
        @SQLType(defaultValue="false") boolean usingOctets,
        @SQLType(defaultValue="1") int occurrence,               //strict
        @SQLType(name="\"group\"", defaultValue="0") int group,  //strict
        @SQLType(defaultValue="false") boolean w3cNewlines
    )
        throws SQLException
    {
        if ( 1 > occurrence )
            return null;
        if ( 0 > group ) // test group > ngroups after compiling regex
            return null;
        if ( usingAndLengthCheck(in, from, usingOctets, "substring_regex") )
            return null;
        newlinesCheck(w3cNewlines, "substring_regex");

        ListOfMatchVectors lomv = LOMV.of(pattern, flag, in, from);

        MatchVector mv = lomv.get(occurrence);
        if ( null == mv  ||  mv.groups() < group )
            return null;

        int codePointPos = mv.position(group);
        int codePointLen = mv.length(group);

        int utf16pos = in.offsetByCodePoints(0, codePointPos - 1);
        int utf16end = in.offsetByCodePoints(utf16pos, codePointLen);

        return in.substring(utf16pos, utf16end);
    }

    /**
     * Syntax-sugar-free form of the ISO SQL
     * <a id='translate_regex'>{@code TRANSLATE_REGEX}</a> function:
     * return a string constructed from the input string by replacing one
     * specified occurrence, or all occurrences, of a matching pattern.
     *<p>
     * Rewrite the standard forms
     *<pre>
     * TRANSLATE_REGEX(pattern FLAG flags IN str WITH repl FROM position
     *                 OCCURRENCE ALL)
     * TRANSLATE_REGEX(pattern FLAG flags IN str WITH repl FROM position
     *                 OCCURRENCE n)
     *</pre>
     * into these forms, respectively:
     *<pre>
     * translate_regex(pattern, flag =&gt; flags, "in" =&gt; str,
     *                 "with" =&gt; repl, "from" =&gt; position)
     * translate_regex(pattern, flag =&gt; flags, "in" =&gt; str,
     *                 "with" =&gt; repl, "from" =&gt; position,
     *                 occurrence =&gt; n)
     *</pre>
     * where all of the named parameters are optional except pattern and "in"
     * (the default for "with" is the empty string, resulting in matches being
     * deleted).
     * See also {@link #position_regex position_regex} regarding the
     * {@code occurrence} parameter,
     * {@link #occurrences_regex occurrences_regex} regarding
     * {@code usingOctets}, and {@link #like_regex like_regex}
     * regarding {@code w3cNewlines}.
     *<p>
     * For the specified occurrence (or all occurrences), the matching portion
     * <em>s</em> of the string is replaced as by the XQuery function
     * <a href='https://www.w3.org/TR/xpath-functions-31/#func-replace'
     * >replace</a>(<em>s, pattern, repl, flags</em>). The <em>repl</em> string
     * may contain {@code $0} to refer to the entire matched substring, or
     * {@code $}<em>m</em> to refer to the <em>m</em>th parenthesized capturing
     * group in the pattern.
     * @param pattern XQuery regular expression to seek in the input string.
     * @param in The input string.
     * @param flag Optional string of
     * <a href='https://www.w3.org/TR/xpath-functions-31/#flags'>flags adjusting
     * the regular expression behavior</a>.
     * @param with The replacement string, possibly with $m references.
     * @param from Starting position in the input string, 1 by default.
     * @param usingOctets Whether position is counted in characters (actual
     * Unicode characters, not any smaller encoded unit, not even Java char),
     * which is the default, or (when true) in octets of the string's encoded
     * form.
     * @param occurrence If specified as an integer n (default 0 for "ALL"),
     * replace the nth match of the pattern in the string.
     * @param w3cNewlines Pass true to allow the regular expression to recognize
     * newlines according to the W3C XQuery rules rather than those of ISO SQL.
     * @return The input string with one occurrence or all occurences of the
     * pattern replaced, as described above. Null if any parameter is null, or
     * if the start position is less than 1 or beyond the end of the string.
     * The input string unchanged if occurrence is less than zero or exceeds the
     * number of matches.
     * @throws SQLException SQLDataException with SQLSTATE 2201S if the regular
     * expression is invalid, 2201T if the flags string is invalid; 2201U if
     * replacing where the pattern has matched a substring of zero length; 2201V
     * if the replacement string has improper form (a backslash must be used to
     * escape any dollar sign or backslash intended literally);
     * SQLFeatureNotSupportedException (0A000) if (in the current
     * implementation) usingOctets is true, or w3cNewlines is false or omitted.
     */
    @Function(implementor="saxon9api", schema="javatest")
    public static String translate_regex(
        String pattern,                                          //strict
        @SQLType(name="\"in\"") String in,                       //strict
        @SQLType(defaultValue="") String flag,                   //strict
        @SQLType(name="\"with\"", defaultValue="") String with,  //strict
        @SQLType(name="\"from\"", defaultValue="1") int from,
        @SQLType(defaultValue="false") boolean usingOctets,
        @SQLType(defaultValue="0" /* ALL */) int occurrence,
        @SQLType(defaultValue="false") boolean w3cNewlines
    )
        throws SQLException
    {
        if ( usingAndLengthCheck(in, from, usingOctets, "translate_regex") )
            return null;
        newlinesCheck(w3cNewlines, "translate_regex");
        if ( 0 > occurrence )
            return in;

        RegularExpression re = compileRE(pattern, flag);

        ListOfMatchVectors lomv = LOMV.of(re, in, from);

        MatchVector mv;
        int codePointPos;
        int codePointLen;
        int utf16pos;
        int utf16end;

        if ( 0 < occurrence )
        {
            mv = lomv.get(occurrence);
            if ( null == mv )
                return in;

            codePointPos = mv.position(0);
            codePointLen = mv.length(0);

            utf16pos = in.offsetByCodePoints(0, codePointPos - 1);
            utf16end = in.offsetByCodePoints(utf16pos, codePointLen);

            return
                in.substring(0, utf16pos)
                + replace(re, in.substring(utf16pos, utf16end), with)
                + in.substring(utf16end);
        }

        StringBuilder sb = new StringBuilder();
        utf16end = 0;

        for ( int i = 1; null != (mv = lomv.get(i)); ++ i )
        {
            codePointPos = mv.position(0);
            codePointLen = mv.length(0);

            utf16pos = in.offsetByCodePoints(0, codePointPos - 1);

            sb.append(in.substring(utf16end, utf16pos));

            utf16end = in.offsetByCodePoints(utf16pos, codePointLen);

            sb.append(replace(re, in.substring(utf16pos, utf16end), with));
        }

        return sb.append(in.substring(utf16end)).toString();
    }
}