All Downloads are FREE. Search and download functionalities are using the official Maven repository.

org.h2.bnf.Bnf Maven / Gradle / Ivy

There is a newer version: 2.3.232
Show newest version
/*
 * Copyright 2004-2010 H2 Group. Multiple-Licensed under the H2 License,
 * Version 1.0, and under the Eclipse Public License, Version 1.0
 * (http://h2database.com/html/license.html).
 * Initial Developer: H2 Group
 */
package org.h2.bnf;

import java.io.ByteArrayInputStream;
import java.io.IOException;
import java.io.InputStreamReader;
import java.io.Reader;
import java.sql.ResultSet;
import java.sql.SQLException;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.StringTokenizer;
import org.h2.server.web.DbContextRule;
import org.h2.tools.Csv;
import org.h2.util.New;
import org.h2.util.StringUtils;
import org.h2.util.Utils;

/**
 * This class can read a file that is similar to BNF (Backus-Naur form).
 * It is made specially to support SQL grammar.
 */
public class Bnf {

    /**
     * The rule map. The key is lowercase, and all spaces
     * are replaces with underscore.
     */
    private final HashMap ruleMap = New.hashMap();
    private String syntax;
    private String currentToken;
    private String[] tokens;
    private char firstChar;
    private int index;
    private Rule lastRepeat;
    private ArrayList statements;
    private String currentTopic;

    /**
     * Create an instance using the grammar specified in the CSV file.
     *
     * @param csv if not specified, the help.csv is used
     * @return a new instance
     */
    public static Bnf getInstance(Reader csv) throws SQLException, IOException {
        Bnf bnf = new Bnf();
        if (csv == null) {
            byte[] data = Utils.getResource("/org/h2/res/help.csv");
            csv = new InputStreamReader(new ByteArrayInputStream(data));
        }
        bnf.parse(csv);
        return bnf;
    }

    private void addFixedRule(String name, int fixedType) {
        Rule rule = new RuleFixed(fixedType);
        addRule(name, "Fixed", rule);
    }

    private RuleHead addRule(String topic, String section, Rule rule) {
        RuleHead head = new RuleHead(section, topic, rule);
        String key = StringUtils.toLowerEnglish(topic.trim().replace(' ', '_'));
        if (ruleMap.get(key) != null) {
            throw new AssertionError("already exists: " + topic);
        }
        ruleMap.put(key, head);
        return head;
    }

    private void parse(Reader csv) throws SQLException, IOException {
        Rule functions = null;
        statements = New.arrayList();
        ResultSet rs = Csv.getInstance().read(csv, null);
        for (int id = 0; rs.next(); id++) {
            String section = rs.getString("SECTION").trim();
            if (section.startsWith("System")) {
                continue;
            }
            String topic = rs.getString("TOPIC");
            syntax = rs.getString("SYNTAX").trim();
            currentTopic = section;
            tokens = tokenize();
            index = 0;
            Rule rule = parseRule();
            if (section.startsWith("Command")) {
                rule = new RuleList(rule, new RuleElement(";\n\n", currentTopic), false);
            }
            RuleHead head = addRule(topic, section, rule);
            if (section.startsWith("Function")) {
                if (functions == null) {
                    functions = rule;
                } else {
                    functions = new RuleList(rule, functions, true);
                }
            } else if (section.startsWith("Commands")) {
                statements.add(head);
            }
        }
        addRule("@func@", "Function", functions);
        addFixedRule("@ymd@", RuleFixed.YMD);
        addFixedRule("@hms@", RuleFixed.HMS);
        addFixedRule("@nanos@", RuleFixed.NANOS);
        addFixedRule("anything_except_single_quote", RuleFixed.ANY_EXCEPT_SINGLE_QUOTE);
        addFixedRule("anything_except_double_quote", RuleFixed.ANY_EXCEPT_DOUBLE_QUOTE);
        addFixedRule("anything_until_end_of_line", RuleFixed.ANY_UNTIL_EOL);
        addFixedRule("anything_until_end_comment", RuleFixed.ANY_UNTIL_END);
        addFixedRule("anything_except_two_dollar_signs", RuleFixed.ANY_EXCEPT_2_DOLLAR);
        addFixedRule("anything", RuleFixed.ANY_WORD);
        addFixedRule("@hex_start@", RuleFixed.HEX_START);
        addFixedRule("@concat@", RuleFixed.CONCAT);
        addFixedRule("@az_@", RuleFixed.AZ_UNDERSCORE);
        addFixedRule("@af@", RuleFixed.AF);
        addFixedRule("@digit@", RuleFixed.DIGIT);
        addFixedRule("@open_bracket@", RuleFixed.OPEN_BRACKET);
        addFixedRule("@close_bracket@", RuleFixed.CLOSE_BRACKET);
    }

    /**
     * Parse the syntax and let the rule call the visitor.
     *
     * @param visitor the visitor
     * @param s the syntax to parse
     */
    public void visit(BnfVisitor visitor, String s) {
        this.syntax = s;
        tokens = tokenize();
        index = 0;
        Rule rule = parseRule();
        rule.setLinks(ruleMap);
        rule.accept(visitor);
    }

    /**
     * Convert convert ruleLink to rule_link.
     *
     * @param token the token
     * @return the rule map key
     */
    public static String getRuleMapKey(String token) {
        StringBuilder buff = new StringBuilder();
        for (char ch : token.toCharArray()) {
            if (Character.isUpperCase(ch)) {
                buff.append('_').append(Character.toLowerCase(ch));
            } else {
                buff.append(ch);
            }
        }
        return buff.toString();
    }

    /**
     * Get the rule head for the given title.
     *
     * @param title the title
     * @return the rule head, or null
     */
    public RuleHead getRuleHead(String title) {
        return ruleMap.get(title);
    }

    private Rule parseRule() {
        read();
        return parseOr();
    }

    private Rule parseOr() {
        Rule r = parseList();
        if (firstChar == '|') {
            read();
            r = new RuleList(r, parseOr(), true);
        }
        lastRepeat = r;
        return r;
    }

    private Rule parseList() {
        Rule r = parseToken();
        if (firstChar != '|' && firstChar != ']' && firstChar != '}' && firstChar != 0) {
            r = new RuleList(r, parseList(), false);
        }
        lastRepeat = r;
        return r;
    }

    private Rule parseToken() {
        Rule r;
        if ((firstChar >= 'A' && firstChar <= 'Z') || (firstChar >= 'a' && firstChar <= 'z')) {
            // r = new RuleElement(currentToken+ " syntax:" + syntax);
            r = new RuleElement(currentToken, currentTopic);
        } else if (firstChar == '[') {
            read();
            Rule r2 = parseOr();
            r = new RuleOptional(r2);
            if (firstChar != ']') {
                throw new AssertionError("expected ], got " + currentToken + " syntax:" + syntax);
            }
        } else if (firstChar == '{') {
            read();
            r = parseOr();
            if (firstChar != '}') {
                throw new AssertionError("expected }, got " + currentToken + " syntax:" + syntax);
            }
        } else if ("@commaDots@".equals(currentToken)) {
            r = new RuleList(new RuleElement(",", currentTopic), lastRepeat, false);
            r = new RuleRepeat(r, true);
        } else if ("@dots@".equals(currentToken)) {
            r = new RuleRepeat(lastRepeat, false);
        } else {
            r = new RuleElement(currentToken, currentTopic);
        }
        lastRepeat = r;
        read();
        return r;
    }

    private void read() {
        if (index < tokens.length) {
            currentToken = tokens[index++];
            firstChar = currentToken.charAt(0);
        } else {
            currentToken = "";
            firstChar = 0;
        }
    }

    private String[] tokenize() {
        ArrayList list = New.arrayList();
        syntax = StringUtils.replaceAll(syntax, "yyyy-MM-dd", "@ymd@");
        syntax = StringUtils.replaceAll(syntax, "hh:mm:ss", "@hms@");
        syntax = StringUtils.replaceAll(syntax, "nnnnnnnnn", "@nanos@");
        syntax = StringUtils.replaceAll(syntax, "function", "@func@");
        syntax = StringUtils.replaceAll(syntax, "0x", "@hexStart@");
        syntax = StringUtils.replaceAll(syntax, ",...", "@commaDots@");
        syntax = StringUtils.replaceAll(syntax, "...", "@dots@");
        syntax = StringUtils.replaceAll(syntax, "||", "@concat@");
        syntax = StringUtils.replaceAll(syntax, "a-z|_", "@az_@");
        syntax = StringUtils.replaceAll(syntax, "A-Z|_", "@az_@");
        syntax = StringUtils.replaceAll(syntax, "a-f", "@af@");
        syntax = StringUtils.replaceAll(syntax, "A-F", "@af@");
        syntax = StringUtils.replaceAll(syntax, "0-9", "@digit@");
        syntax = StringUtils.replaceAll(syntax, "'['", "@openBracket@");
        syntax = StringUtils.replaceAll(syntax, "']'", "@closeBracket@");
        StringTokenizer tokenizer = getTokenizer(syntax);
        while (tokenizer.hasMoreTokens()) {
            String s = tokenizer.nextToken();
            // avoid duplicate strings
            s = StringUtils.cache(s);
            if (s.length() == 1) {
                if (" \r\n".indexOf(s.charAt(0)) >= 0) {
                    continue;
                }
            }
            list.add(s);
        }
        return list.toArray(new String[list.size()]);
    }

    /**
     * Get the list of tokens that can follow.
     * This is the main autocomplete method.
     * The returned map for the query 'S' may look like this:
     * 
     * key: 1#SELECT, value: ELECT
     * key: 1#SET, value: ET
     * 
* * @param query the start of the statement * @return the map of possible token types / tokens */ public HashMap getNextTokenList(String query) { Sentence sentence = new Sentence(); sentence.setQuery(query); for (RuleHead head : statements) { if (!head.getSection().startsWith("Commands")) { continue; } sentence.start(); head.getRule().addNextTokenList(sentence); } return sentence.getNext(); } /** * Cross-link all statements with each other. * This method is called after updating the topics. */ public void linkStatements() { for (RuleHead r : ruleMap.values()) { r.getRule().setLinks(ruleMap); } } /** * Update a topic with a context specific rule. * This is used for autocomplete support. * * @param topic the topic * @param rule the database context rule */ public void updateTopic(String topic, DbContextRule rule) { topic = StringUtils.toLowerEnglish(topic); RuleHead head = ruleMap.get(topic); if (head == null) { head = new RuleHead("db", topic, rule); ruleMap.put(topic, head); statements.add(head); } else { head.setRule(rule); } } /** * Get the list of possible statements. * * @return the list of statements */ public ArrayList getStatements() { return statements; } /** * Get the tokenizer for the given syntax. * * @param s the syntax * @return the tokenizer */ public StringTokenizer getTokenizer(String s) { return new StringTokenizer(s, " [](){}|.,\r\n<>:-+*/=<\">!'$", true); } }




© 2015 - 2024 Weber Informatics LLC | Privacy Policy