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

org.antlr.v4.tool.Grammar Maven / Gradle / Ivy

There is a newer version: 4.13.1.5
Show newest version
/*
 * Copyright (c) 2012 The ANTLR Project. All rights reserved.
 * Use of this file is governed by the BSD-3-Clause license that
 * can be found in the LICENSE.txt file in the project root.
 */

package org.antlr.v4.tool;

import org.antlr.v4.Tool;
import org.antlr.v4.analysis.LeftRecursiveRuleTransformer;
import org.antlr.v4.automata.ParserATNFactory;
import org.antlr.v4.misc.CharSupport;
import org.antlr.v4.misc.OrderedHashMap;
import org.antlr.v4.misc.Utils;
import org.antlr.v4.parse.ANTLRParser;
import org.antlr.v4.parse.GrammarASTAdaptor;
import org.antlr.v4.parse.GrammarTreeVisitor;
import org.antlr.v4.parse.TokenVocabParser;
import org.antlr.v4.runtime.CharStream;
import org.antlr.v4.runtime.Lexer;
import org.antlr.v4.runtime.LexerInterpreter;
import org.antlr.v4.runtime.ParserInterpreter;
import org.antlr.v4.runtime.Token;
import org.antlr.v4.runtime.TokenStream;
import org.antlr.v4.runtime.Vocabulary;
import org.antlr.v4.runtime.VocabularyImpl;
import org.antlr.v4.runtime.atn.ATN;
import org.antlr.v4.runtime.atn.ATNDeserializer;
import org.antlr.v4.runtime.atn.ATNSerializer;
import org.antlr.v4.runtime.atn.SemanticContext;
import org.antlr.v4.runtime.dfa.DFA;
import org.antlr.v4.runtime.misc.IntSet;
import org.antlr.v4.runtime.misc.Interval;
import org.antlr.v4.runtime.misc.IntervalSet;
import org.antlr.v4.runtime.misc.NotNull;
import org.antlr.v4.runtime.misc.Nullable;
import org.antlr.v4.runtime.misc.Tuple;
import org.antlr.v4.runtime.misc.Tuple2;
import org.antlr.v4.tool.ast.ActionAST;
import org.antlr.v4.tool.ast.AltAST;
import org.antlr.v4.tool.ast.GrammarAST;
import org.antlr.v4.tool.ast.GrammarASTWithOptions;
import org.antlr.v4.tool.ast.GrammarRootAST;
import org.antlr.v4.tool.ast.PredAST;
import org.antlr.v4.tool.ast.RuleAST;
import org.antlr.v4.tool.ast.TerminalAST;

import java.io.IOException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.HashMap;
import java.util.HashSet;
import java.util.LinkedHashMap;
import java.util.LinkedHashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;

public class Grammar implements AttributeResolver {
	public static final String GRAMMAR_FROM_STRING_NAME = "";
	/**
	 * This value is used in the following situations to indicate that a token
	 * type does not have an associated name which can be directly referenced in
	 * a grammar.
	 *
	 * 
    *
  • This value is the name and display name for the token with type * {@link Token#INVALID_TYPE}.
  • *
  • This value is the name for tokens with a type not represented by a * named token. The display name for these tokens is simply the string * representation of the token type as an integer.
  • *
*/ public static final String INVALID_TOKEN_NAME = ""; /** * This value is used as the name for elements in the array returned by * {@link #getRuleNames} for indexes not associated with a rule. */ public static final String INVALID_RULE_NAME = ""; public static final Set parserOptions = new HashSet(); static { parserOptions.add("superClass"); parserOptions.add("contextSuperClass"); parserOptions.add("TokenLabelType"); parserOptions.add("abstract"); parserOptions.add("tokenVocab"); parserOptions.add("language"); parserOptions.add("accessLevel"); parserOptions.add("exportMacro"); } public static final Set lexerOptions = parserOptions; public static final Set ruleOptions = new HashSet(); static { ruleOptions.add("baseContext"); } public static final Set ParserBlockOptions = new HashSet(); static { ParserBlockOptions.add("sll"); } public static final Set LexerBlockOptions = new HashSet(); /** Legal options for rule refs like id<key=value> */ public static final Set ruleRefOptions = new HashSet(); static { ruleRefOptions.add(LeftRecursiveRuleTransformer.PRECEDENCE_OPTION_NAME); ruleRefOptions.add(LeftRecursiveRuleTransformer.TOKENINDEX_OPTION_NAME); } /** Legal options for terminal refs like ID<assoc=right> */ public static final Set tokenOptions = new HashSet(); static { tokenOptions.add("assoc"); tokenOptions.add(LeftRecursiveRuleTransformer.TOKENINDEX_OPTION_NAME); } public static final Set actionOptions = new HashSet(); public static final Set semPredOptions = new HashSet(); static { semPredOptions.add(LeftRecursiveRuleTransformer.PRECEDENCE_OPTION_NAME); semPredOptions.add("fail"); } public static final Set doNotCopyOptionsToLexer = new HashSet(); static { doNotCopyOptionsToLexer.add("superClass"); doNotCopyOptionsToLexer.add("TokenLabelType"); doNotCopyOptionsToLexer.add("abstract"); doNotCopyOptionsToLexer.add("tokenVocab"); } public static final Map grammarAndLabelRefTypeToScope = new HashMap(); static { grammarAndLabelRefTypeToScope.put("parser:RULE_LABEL", Rule.predefinedRulePropertiesDict); grammarAndLabelRefTypeToScope.put("parser:TOKEN_LABEL", AttributeDict.predefinedTokenDict); grammarAndLabelRefTypeToScope.put("combined:RULE_LABEL", Rule.predefinedRulePropertiesDict); grammarAndLabelRefTypeToScope.put("combined:TOKEN_LABEL", AttributeDict.predefinedTokenDict); } public String name; public GrammarRootAST ast; /** Track token stream used to create this grammar */ @NotNull public final org.antlr.runtime.TokenStream tokenStream; /** If we transform grammar, track original unaltered token stream. * This is set to the same value as tokenStream when tokenStream is * initially set. * * If this field differs from tokenStream, then we have transformed * the grammar. */ @NotNull public final org.antlr.runtime.TokenStream originalTokenStream; public String text; // testing only public String fileName; /** Was this parser grammar created from a COMBINED grammar? If so, * this is what we extracted. */ public LexerGrammar implicitLexer; /** If this is an extracted/implicit lexer, we point at original grammar */ public Grammar originalGrammar; /** If we're imported, who imported us? If null, implies grammar is root */ public Grammar parent; public List importedGrammars; /** All rules defined in this specific grammar, not imported. Also does * not include lexical rules if combined. */ public OrderedHashMap rules = new OrderedHashMap(); public List indexToRule = new ArrayList(); /** * This maps a context name → a collection of {@link RuleAST} nodes in * the original grammar. The union of accessors and labels identified by * these ASTs define the accessor methods and fields of the generated * context classes. * *

* The keys of this map match the result of {@link Rule#getBaseContext}.

*

* The values in this map are clones of the nodes in the original grammar * (provided by {@link GrammarAST#dupTree}) to ensure that grammar * transformations do not affect the values generated for the contexts. The * duplication is performed after nodes from imported grammars are merged * into the AST.

*/ public final Map> contextASTs = new HashMap>(); int ruleNumber = 0; // used to get rule indexes (0..n-1) int stringLiteralRuleNumber = 0; // used to invent rule names for 'keyword', ';', ... (0..n-1) /** The ATN that represents the grammar with edges labelled with tokens * or epsilon. It is more suitable to analysis than an AST representation. */ public ATN atn; public Map stateToGrammarRegionMap; public Map decisionDFAs = new HashMap(); public List decisionLOOK; @NotNull public final Tool tool; /** Token names and literal tokens like "void" are uniquely indexed. * with -1 implying EOF. Characters are different; they go from * -1 (EOF) to \uFFFE. For example, 0 could be a binary byte you * want to lexer. Labels of DFA/ATN transitions can be both tokens * and characters. I use negative numbers for bookkeeping labels * like EPSILON. Char/String literals and token types overlap in the same * space, however. */ int maxTokenType = Token.MIN_USER_TOKEN_TYPE -1; /** * Map token like {@code ID} (but not literals like {@code 'while'}) to its * token type. */ public final Map tokenNameToTypeMap = new LinkedHashMap(); /** * Map token literals like {@code 'while'} to its token type. It may be that * {@code WHILE="while"=35}, in which case both {@link #tokenNameToTypeMap} * and this field will have entries both mapped to 35. */ public final Map stringLiteralToTypeMap = new LinkedHashMap(); /** * Reverse index for {@link #stringLiteralToTypeMap}. Indexed with raw token * type. 0 is invalid. */ public final List typeToStringLiteralList = new ArrayList(); /** * Map a token type to its token name. Indexed with raw token type. 0 is * invalid. */ public final List typeToTokenList = new ArrayList(); /** * The maximum channel value which is assigned by this grammar. Values below * {@link Token#MIN_USER_CHANNEL_VALUE} are assumed to be predefined. */ int maxChannelType = Token.MIN_USER_CHANNEL_VALUE - 1; /** * Map channel like {@code COMMENTS_CHANNEL} to its constant channel value. * Only user-defined channels are defined in this map. */ public final Map channelNameToValueMap = new LinkedHashMap(); /** * Map a constant channel value to its name. Indexed with raw channel value. * The predefined channels {@link Token#DEFAULT_CHANNEL} and * {@link Token#HIDDEN_CHANNEL} are not stored in this list, so the values * at the corresponding indexes is {@code null}. */ public final List channelValueToNameList = new ArrayList(); /** Map a name to an action. * The code generator will use this to fill holes in the output files. * I track the AST node for the action in case I need the line number * for errors. */ public Map namedActions = new HashMap(); /** Tracks all user lexer actions in all alternatives of all rules. * Doesn't track sempreds. maps tree node to action index (alt number 1..n). */ public LinkedHashMap lexerActions = new LinkedHashMap(); /** All sempreds found in grammar; maps tree node to sempred index; * sempred index is 0..n-1 */ public LinkedHashMap sempreds = new LinkedHashMap(); /** Map the other direction upon demand */ public LinkedHashMap indexToPredMap; public static final String AUTO_GENERATED_TOKEN_NAME_PREFIX = "T__"; public Grammar(Tool tool, @NotNull GrammarRootAST ast) { if ( ast==null ) { throw new NullPointerException("ast"); } if (ast.tokenStream == null) { throw new IllegalArgumentException("ast must have a token stream"); } this.tool = tool; this.ast = ast; this.name = (ast.getChild(0)).getText(); this.tokenStream = ast.tokenStream; this.originalTokenStream = this.tokenStream; initTokenSymbolTables(); } /** For testing */ public Grammar(String grammarText) throws org.antlr.runtime.RecognitionException { this(GRAMMAR_FROM_STRING_NAME, grammarText, null); } public Grammar(String grammarText, LexerGrammar tokenVocabSource) throws org.antlr.runtime.RecognitionException { this(GRAMMAR_FROM_STRING_NAME, grammarText, tokenVocabSource, null); } /** For testing */ public Grammar(String grammarText, ANTLRToolListener listener) throws org.antlr.runtime.RecognitionException { this(GRAMMAR_FROM_STRING_NAME, grammarText, listener); } /** For testing; builds trees, does sem anal */ public Grammar(String fileName, String grammarText) throws org.antlr.runtime.RecognitionException { this(fileName, grammarText, null); } /** For testing; builds trees, does sem anal */ public Grammar(String fileName, String grammarText, @Nullable ANTLRToolListener listener) throws org.antlr.runtime.RecognitionException { this(fileName, grammarText, null, listener); } /** For testing; builds trees, does sem anal */ public Grammar(String fileName, String grammarText, Grammar tokenVocabSource, @Nullable ANTLRToolListener listener) throws org.antlr.runtime.RecognitionException { this.text = grammarText; this.fileName = fileName; this.tool = new Tool(); this.tool.addListener(listener); org.antlr.runtime.ANTLRStringStream in = new org.antlr.runtime.ANTLRStringStream(grammarText); in.name = fileName; this.ast = tool.parse(fileName, in); if ( ast==null ) { throw new UnsupportedOperationException(); } if (ast.tokenStream == null) { throw new IllegalStateException("expected ast to have a token stream"); } this.tokenStream = ast.tokenStream; this.originalTokenStream = this.tokenStream; // ensure each node has pointer to surrounding grammar final Grammar thiz = this; org.antlr.runtime.tree.TreeVisitor v = new org.antlr.runtime.tree.TreeVisitor(new GrammarASTAdaptor()); v.visit(ast, new org.antlr.runtime.tree.TreeVisitorAction() { @Override public Object pre(Object t) { ((GrammarAST)t).g = thiz; return t; } @Override public Object post(Object t) { return t; } }); initTokenSymbolTables(); if (tokenVocabSource != null) { importVocab(tokenVocabSource); } tool.process(this, false); } protected void initTokenSymbolTables() { tokenNameToTypeMap.put("EOF", Token.EOF); // reserve a spot for the INVALID token typeToTokenList.add(null); } public void loadImportedGrammars() { if ( ast==null ) return; GrammarAST i = (GrammarAST)ast.getFirstChildWithType(ANTLRParser.IMPORT); if ( i==null ) return; Set visited = new HashSet(); visited.add(this.name); importedGrammars = new ArrayList(); for (Object c : i.getChildren()) { GrammarAST t = (GrammarAST)c; String importedGrammarName = null; if ( t.getType()==ANTLRParser.ASSIGN ) { t = (GrammarAST)t.getChild(1); importedGrammarName = t.getText(); } else if ( t.getType()==ANTLRParser.ID ) { importedGrammarName = t.getText(); } if ( visited.contains(importedGrammarName) ) { // ignore circular refs continue; } Grammar g; try { g = tool.loadImportedGrammar(this, t); } catch (IOException ioe) { tool.errMgr.grammarError(ErrorType.ERROR_READING_IMPORTED_GRAMMAR, importedGrammarName, t.getToken(), importedGrammarName, name); continue; } // did it come back as error node or missing? if ( g == null ) continue; g.parent = this; importedGrammars.add(g); g.loadImportedGrammars(); // recursively pursue any imports in this import } } public void defineAction(GrammarAST atAST) { if ( atAST.getChildCount()==2 ) { String name = atAST.getChild(0).getText(); namedActions.put(name, (ActionAST)atAST.getChild(1)); } else { String scope = atAST.getChild(0).getText(); String gtype = getTypeString(); if ( scope.equals(gtype) || (scope.equals("parser")&>ype.equals("combined")) ) { String name = atAST.getChild(1).getText(); namedActions.put(name, (ActionAST)atAST.getChild(2)); } } } /** * Define the specified rule in the grammar. This method assigns the rule's * {@link Rule#index} according to the {@link #ruleNumber} field, and adds * the {@link Rule} instance to {@link #rules} and {@link #indexToRule}. * * @param r The rule to define in the grammar. * @return {@code true} if the rule was added to the {@link Grammar} * instance; otherwise, {@code false} if a rule with this name already * existed in the grammar instance. */ public boolean defineRule(@NotNull Rule r) { if ( rules.get(r.name)!=null ) { return false; } rules.put(r.name, r); r.index = ruleNumber++; indexToRule.add(r); return true; } /** * Undefine the specified rule from this {@link Grammar} instance. The * instance {@code r} is removed from {@link #rules} and * {@link #indexToRule}. This method updates the {@link Rule#index} field * for all rules defined after {@code r}, and decrements {@link #ruleNumber} * in preparation for adding new rules. *

* This method does nothing if the current {@link Grammar} does not contain * the instance {@code r} at index {@code r.index} in {@link #indexToRule}. *

* * @param r * @return {@code true} if the rule was removed from the {@link Grammar} * instance; otherwise, {@code false} if the specified rule was not defined * in the grammar. */ public boolean undefineRule(@NotNull Rule r) { if (r.index < 0 || r.index >= indexToRule.size() || indexToRule.get(r.index) != r) { return false; } assert rules.get(r.name) == r; rules.remove(r.name); indexToRule.remove(r.index); for (int i = r.index; i < indexToRule.size(); i++) { assert indexToRule.get(i).index == i + 1; indexToRule.get(i).index--; } ruleNumber--; return true; } // public int getNumRules() { // int n = rules.size(); // List imports = getAllImportedGrammars(); // if ( imports!=null ) { // for (Grammar g : imports) n += g.getNumRules(); // } // return n; // } public Rule getRule(String name) { Rule r = rules.get(name); if ( r!=null ) return r; return null; /* List imports = getAllImportedGrammars(); if ( imports==null ) return null; for (Grammar g : imports) { r = g.getRule(name); // recursively walk up hierarchy if ( r!=null ) return r; } return null; */ } public ATN getATN() { if ( atn==null ) { ParserATNFactory factory = new ParserATNFactory(this); atn = factory.createATN(); } return atn; } public Rule getRule(int index) { return indexToRule.get(index); } public Rule getRule(String grammarName, String ruleName) { if ( grammarName!=null ) { // scope override Grammar g = getImportedGrammar(grammarName); if ( g ==null ) { return null; } return g.rules.get(ruleName); } return getRule(ruleName); } protected String getBaseContextName(String ruleName) { Rule referencedRule = rules.get(ruleName); if (referencedRule != null) { ruleName = referencedRule.getBaseContext(); } return ruleName; } public List getUnlabeledAlternatives(RuleAST ast) throws org.antlr.runtime.RecognitionException { AltLabelVisitor visitor = new AltLabelVisitor(new org.antlr.runtime.tree.CommonTreeNodeStream(new GrammarASTAdaptor(), ast)); visitor.rule(); return visitor.getUnlabeledAlternatives(); } public Map>> getLabeledAlternatives(RuleAST ast) throws org.antlr.runtime.RecognitionException { AltLabelVisitor visitor = new AltLabelVisitor(new org.antlr.runtime.tree.CommonTreeNodeStream(new GrammarASTAdaptor(), ast)); visitor.rule(); return visitor.getLabeledAlternatives(); } /** Get list of all imports from all grammars in the delegate subtree of g. * The grammars are in import tree preorder. Don't include ourselves * in list as we're not a delegate of ourselves. */ public List getAllImportedGrammars() { if (importedGrammars == null) { return null; } LinkedHashMap delegates = new LinkedHashMap(); for (Grammar d : importedGrammars) { delegates.put(d.fileName, d); List ds = d.getAllImportedGrammars(); if (ds != null) { for (Grammar imported : ds) { delegates.put(imported.fileName, imported); } } } return new ArrayList(delegates.values()); } public List getImportedGrammars() { return importedGrammars; } public LexerGrammar getImplicitLexer() { return implicitLexer; } /** convenience method for Tool.loadGrammar() */ public static Grammar load(String fileName) { Tool antlr = new Tool(); return antlr.loadGrammar(fileName); } /** Return list of imported grammars from root down to our parent. * Order is [root, ..., this.parent]. (us not included). */ public List getGrammarAncestors() { Grammar root = getOutermostGrammar(); if ( this==root ) return null; List grammars = new ArrayList(); // walk backwards to root, collecting grammars Grammar p = this.parent; while ( p!=null ) { grammars.add(0, p); // add to head so in order later p = p.parent; } return grammars; } /** Return the grammar that imported us and our parents. Return this * if we're root. */ public Grammar getOutermostGrammar() { if ( parent==null ) return this; return parent.getOutermostGrammar(); } public boolean isAbstract() { return Boolean.parseBoolean(getOptionString("abstract")); } /** Get the name of the generated recognizer; may or may not be same * as grammar name. * Recognizer is TParser and TLexer from T if combined, else * just use T regardless of grammar type. */ public String getRecognizerName() { String suffix = ""; List grammarsFromRootToMe = getOutermostGrammar().getGrammarAncestors(); String qualifiedName = name; if ( grammarsFromRootToMe!=null ) { StringBuilder buf = new StringBuilder(); for (Grammar g : grammarsFromRootToMe) { buf.append(g.name); buf.append('_'); } if (isAbstract()) { buf.append("Abstract"); } buf.append(name); qualifiedName = buf.toString(); } else if (isAbstract()) { qualifiedName = "Abstract" + name; } if ( isCombined() || (isLexer() && implicitLexer!=null) ) { suffix = Grammar.getGrammarTypeToFileNameSuffix(getType()); } return qualifiedName+suffix; } public String getStringLiteralLexerRuleName(String lit) { return AUTO_GENERATED_TOKEN_NAME_PREFIX + stringLiteralRuleNumber++; } /** Return grammar directly imported by this grammar */ public Grammar getImportedGrammar(String name) { for (Grammar g : importedGrammars) { if ( g.name.equals(name) ) return g; } return null; } public int getTokenType(String token) { Integer I; if ( token.charAt(0)=='\'') { I = stringLiteralToTypeMap.get(token); } else { // must be a label like ID I = tokenNameToTypeMap.get(token); } int i = (I!=null)? I : Token.INVALID_TYPE; //tool.log("grammar", "grammar type "+type+" "+tokenName+"->"+i); return i; } public String getTokenName(String literal) { Grammar grammar = this; while (grammar != null) { if (grammar.stringLiteralToTypeMap.containsKey(literal)) return grammar.getTokenName(grammar.stringLiteralToTypeMap.get(literal)); grammar = grammar.parent; } return null; } /** Given a token type, get a meaningful name for it such as the ID * or string literal. If this is a lexer and the ttype is in the * char vocabulary, compute an ANTLR-valid (possibly escaped) char literal. */ public String getTokenDisplayName(int ttype) { // inside any target's char range and is lexer grammar? if ( isLexer() && ttype >= Lexer.MIN_CHAR_VALUE && ttype <= Lexer.MAX_CHAR_VALUE ) { return CharSupport.getANTLRCharLiteralForChar(ttype); } if ( ttype==Token.EOF ) { return "EOF"; } if ( ttype==Token.INVALID_TYPE ) { return INVALID_TOKEN_NAME; } if (ttype >= 0 && ttype < typeToStringLiteralList.size() && typeToStringLiteralList.get(ttype) != null) { return typeToStringLiteralList.get(ttype); } if (ttype >= 0 && ttype < typeToTokenList.size() && typeToTokenList.get(ttype) != null) { return typeToTokenList.get(ttype); } return String.valueOf(ttype); } /** * Gets the name by which a token can be referenced in the generated code. * For tokens defined in a {@code tokens{}} block or via a lexer rule, this * is the declared name of the token. For token types generated by the use * of a string literal within a parser rule of a combined grammar, this is * the automatically generated token type which includes the * {@link #AUTO_GENERATED_TOKEN_NAME_PREFIX} prefix. For types which are not * associated with a defined token, this method returns * {@link #INVALID_TOKEN_NAME}. * * @param ttype The token type. * @return The name of the token with the specified type. */ @NotNull public String getTokenName(int ttype) { // inside any target's char range and is lexer grammar? if ( isLexer() && ttype >= Lexer.MIN_CHAR_VALUE && ttype <= Lexer.MAX_CHAR_VALUE ) { return CharSupport.getANTLRCharLiteralForChar(ttype); } if ( ttype==Token.EOF ) { return "EOF"; } if (ttype >= 0 && ttype < typeToTokenList.size() && typeToTokenList.get(ttype) != null) { return typeToTokenList.get(ttype); } return INVALID_TOKEN_NAME; } /** * Gets the constant channel value for a user-defined channel. * *

* This method only returns channel values for user-defined channels. All * other channels, including the predefined channels * {@link Token#DEFAULT_CHANNEL} and {@link Token#HIDDEN_CHANNEL} along with * any channel defined in code (e.g. in a {@code @members{}} block), are * ignored.

* * @param channel The channel name. * @return The channel value, if {@code channel} is the name of a known * user-defined token channel; otherwise, -1. */ public int getChannelValue(String channel) { Integer I = channelNameToValueMap.get(channel); int i = (I != null) ? I : -1; return i; } /** * Gets an array of rule names for rules defined or imported by the * grammar. The array index is the rule index, and the value is the name of * the rule with the corresponding {@link Rule#index}. * *

If no rule is defined with an index for an element of the resulting * array, the value of that element is {@link #INVALID_RULE_NAME}.

* * @return The names of all rules defined in the grammar. */ public String[] getRuleNames() { String[] result = new String[rules.size()]; Arrays.fill(result, INVALID_RULE_NAME); for (Rule rule : rules.values()) { result[rule.index] = rule.name; } return result; } /** * Gets an array of token names for tokens defined or imported by the * grammar. The array index is the token type, and the value is the result * of {@link #getTokenName} for the corresponding token type. * * @see #getTokenName * @return The token names of all tokens defined in the grammar. */ public String[] getTokenNames() { int numTokens = getMaxTokenType(); String[] tokenNames = new String[numTokens+1]; for (int i = 0; i < tokenNames.length; i++) { tokenNames[i] = getTokenName(i); } return tokenNames; } /** * Gets an array of display names for tokens defined or imported by the * grammar. The array index is the token type, and the value is the result * of {@link #getTokenDisplayName} for the corresponding token type. * * @see #getTokenDisplayName * @return The display names of all tokens defined in the grammar. */ public String[] getTokenDisplayNames() { int numTokens = getMaxTokenType(); String[] tokenNames = new String[numTokens+1]; for (int i = 0; i < tokenNames.length; i++) { tokenNames[i] = getTokenDisplayName(i); } return tokenNames; } /** * Gets the literal names assigned to tokens in the grammar. */ @NotNull public String[] getTokenLiteralNames() { int numTokens = getMaxTokenType(); String[] literalNames = new String[numTokens+1]; for (int i = 0; i < Math.min(literalNames.length, typeToStringLiteralList.size()); i++) { literalNames[i] = typeToStringLiteralList.get(i); } for (Map.Entry entry : stringLiteralToTypeMap.entrySet()) { if (entry.getValue() >= 0 && entry.getValue() < literalNames.length && literalNames[entry.getValue()] == null) { literalNames[entry.getValue()] = entry.getKey(); } } return literalNames; } /** * Gets the symbolic names assigned to tokens in the grammar. */ @NotNull public String[] getTokenSymbolicNames() { int numTokens = getMaxTokenType(); String[] symbolicNames = new String[numTokens+1]; for (int i = 0; i < Math.min(symbolicNames.length, typeToTokenList.size()); i++) { if (typeToTokenList.get(i) == null || typeToTokenList.get(i).startsWith(AUTO_GENERATED_TOKEN_NAME_PREFIX)) { continue; } symbolicNames[i] = typeToTokenList.get(i); } return symbolicNames; } /** * Gets a {@link Vocabulary} instance describing the vocabulary used by the * grammar. */ @NotNull public Vocabulary getVocabulary() { return new VocabularyImpl(getTokenLiteralNames(), getTokenSymbolicNames()); } /** Given an arbitrarily complex SemanticContext, walk the "tree" and get display string. * Pull predicates from grammar text. */ public String getSemanticContextDisplayString(SemanticContext semctx) { if ( semctx instanceof SemanticContext.Predicate ) { return getPredicateDisplayString((SemanticContext.Predicate)semctx); } if ( semctx instanceof SemanticContext.AND ) { SemanticContext.AND and = (SemanticContext.AND)semctx; return joinPredicateOperands(and, " and "); } if ( semctx instanceof SemanticContext.OR ) { SemanticContext.OR or = (SemanticContext.OR)semctx; return joinPredicateOperands(or, " or "); } return semctx.toString(); } public String joinPredicateOperands(SemanticContext.Operator op, String separator) { StringBuilder buf = new StringBuilder(); for (SemanticContext operand : op.getOperands()) { if (buf.length() > 0) { buf.append(separator); } buf.append(getSemanticContextDisplayString(operand)); } return buf.toString(); } public LinkedHashMap getIndexToPredicateMap() { LinkedHashMap indexToPredMap = new LinkedHashMap(); for (Rule r : rules.values()) { for (ActionAST a : r.actions) { if (a instanceof PredAST) { PredAST p = (PredAST) a; indexToPredMap.put(sempreds.get(p), p); } } } return indexToPredMap; } public String getPredicateDisplayString(SemanticContext.Predicate pred) { if ( indexToPredMap==null ) { indexToPredMap = getIndexToPredicateMap(); } ActionAST actionAST = indexToPredMap.get(pred.predIndex); return actionAST.getText(); } /** What is the max char value possible for this grammar's target? Use * unicode max if no target defined. */ public int getMaxCharValue() { return org.antlr.v4.runtime.Lexer.MAX_CHAR_VALUE; // if ( generator!=null ) { // return generator.target.getMaxCharValue(generator); // } // else { // return Label.MAX_CHAR_VALUE; // } } /** Return a set of all possible token or char types for this grammar */ public IntSet getTokenTypes() { if ( isLexer() ) { return getAllCharValues(); } return IntervalSet.of(Token.MIN_USER_TOKEN_TYPE, getMaxTokenType()); } /** Return min to max char as defined by the target. * If no target, use max unicode char value. */ public IntSet getAllCharValues() { return IntervalSet.of(Lexer.MIN_CHAR_VALUE, getMaxCharValue()); } /** How many token types have been allocated so far? */ public int getMaxTokenType() { return typeToTokenList.size() - 1; // don't count 0 (invalid) } /** Return a new unique integer in the token type space */ public int getNewTokenType() { maxTokenType++; return maxTokenType; } /** Return a new unique integer in the channel value space. */ public int getNewChannelNumber() { maxChannelType++; return maxChannelType; } public void importTokensFromTokensFile() { String vocab = getOptionString("tokenVocab"); if ( vocab!=null ) { TokenVocabParser vparser = new TokenVocabParser(this); Map tokens = vparser.load(); tool.log("grammar", "tokens=" + tokens); for (String t : tokens.keySet()) { if ( t.charAt(0)=='\'' ) defineStringLiteral(t, tokens.get(t)); else defineTokenName(t, tokens.get(t)); } } } public void importVocab(Grammar importG) { for (String tokenName: importG.tokenNameToTypeMap.keySet()) { defineTokenName(tokenName, importG.tokenNameToTypeMap.get(tokenName)); } for (String tokenName: importG.stringLiteralToTypeMap.keySet()) { defineStringLiteral(tokenName, importG.stringLiteralToTypeMap.get(tokenName)); } for (Map.Entry channel : importG.channelNameToValueMap.entrySet()) { defineChannelName(channel.getKey(), channel.getValue()); } // this.tokenNameToTypeMap.putAll( importG.tokenNameToTypeMap ); // this.stringLiteralToTypeMap.putAll( importG.stringLiteralToTypeMap ); int max = Math.max(this.typeToTokenList.size(), importG.typeToTokenList.size()); Utils.setSize(typeToTokenList, max); for (int ttype=0; ttype=typeToStringLiteralList.size() ) { Utils.setSize(typeToStringLiteralList, ttype+1); } typeToStringLiteralList.set(ttype, lit); setTokenForType(ttype, lit); return ttype; } return Token.INVALID_TYPE; } public int defineTokenAlias(String name, String lit) { int ttype = defineTokenName(name); stringLiteralToTypeMap.put(lit, ttype); setTokenForType(ttype, name); return ttype; } public void setTokenForType(int ttype, String text) { if (ttype == Token.EOF) { // ignore EOF, it will be reported as an error separately return; } if ( ttype>=typeToTokenList.size() ) { Utils.setSize(typeToTokenList, ttype+1); } String prevToken = typeToTokenList.get(ttype); if ( prevToken==null || prevToken.charAt(0)=='\'' ) { // only record if nothing there before or if thing before was a literal typeToTokenList.set(ttype, text); } } /** * Define a token channel with a specified name. * *

* If a channel with the specified name already exists, the previously * assigned channel value is returned.

* * @param name The channel name. * @return The constant channel value assigned to the channel. */ public int defineChannelName(String name) { Integer prev = channelNameToValueMap.get(name); if (prev == null) { return defineChannelName(name, getNewChannelNumber()); } return prev; } /** * Define a token channel with a specified name. * *

* If a channel with the specified name already exists, the previously * assigned channel value is not altered.

* * @param name The channel name. * @return The constant channel value assigned to the channel. */ public int defineChannelName(String name, int value) { Integer prev = channelNameToValueMap.get(name); if (prev != null) { return prev; } channelNameToValueMap.put(name, value); setChannelNameForValue(value, name); maxChannelType = Math.max(maxChannelType, value); return value; } /** * Sets the channel name associated with a particular channel value. * *

* If a name has already been assigned to the channel with constant value * {@code channelValue}, this method does nothing.

* * @param channelValue The constant value for the channel. * @param name The channel name. */ public void setChannelNameForValue(int channelValue, String name) { if (channelValue >= channelValueToNameList.size()) { Utils.setSize(channelValueToNameList, channelValue + 1); } String prevChannel = channelValueToNameList.get(channelValue); if (prevChannel == null) { channelValueToNameList.set(channelValue, name); } } // no isolated attr at grammar action level @Override public Attribute resolveToAttribute(String x, ActionAST node) { return null; } // no $x.y makes sense here @Override public Attribute resolveToAttribute(String x, String y, ActionAST node) { return null; } @Override public boolean resolvesToLabel(String x, ActionAST node) { return false; } @Override public boolean resolvesToListLabel(String x, ActionAST node) { return false; } @Override public boolean resolvesToToken(String x, ActionAST node) { return false; } @Override public boolean resolvesToAttributeDict(String x, ActionAST node) { return false; } /** Given a grammar type, what should be the default action scope? * If I say @members in a COMBINED grammar, for example, the * default scope should be "parser". */ public String getDefaultActionScope() { switch ( getType() ) { case ANTLRParser.LEXER : return "lexer"; case ANTLRParser.PARSER : case ANTLRParser.COMBINED : return "parser"; } return null; } public int getType() { if ( ast!=null ) return ast.grammarType; return 0; } public org.antlr.runtime.TokenStream getTokenStream() { if ( ast!=null ) return ast.tokenStream; return null; } public boolean isLexer() { return getType()==ANTLRParser.LEXER; } public boolean isParser() { return getType()==ANTLRParser.PARSER; } public boolean isCombined() { return getType()==ANTLRParser.COMBINED; } /** Is id a valid token name? Does id start with an uppercase letter? */ public static boolean isTokenName(String id) { return Character.isUpperCase(id.charAt(0)); } public String getTypeString() { if ( ast==null ) return null; return ANTLRParser.tokenNames[getType()].toLowerCase(); } public static String getGrammarTypeToFileNameSuffix(int type) { switch ( type ) { case ANTLRParser.LEXER : return "Lexer"; case ANTLRParser.PARSER : return "Parser"; // if combined grammar, gen Parser and Lexer will be done later // TODO: we are separate now right? case ANTLRParser.COMBINED : return "Parser"; default : return ""; } } public String getOptionString(String key) { return ast.getOptionString(key); } /** Given ^(TOKEN_REF ^(OPTIONS ^(ELEMENT_OPTIONS (= assoc right)))) * set option assoc=right in TOKEN_REF. */ public static void setNodeOptions(GrammarAST node, GrammarAST options) { if ( options==null ) return; GrammarASTWithOptions t = (GrammarASTWithOptions)node; if ( t.getChildCount()==0 || options.getChildCount()==0 ) return; for (Object o : options.getChildren()) { GrammarAST c = (GrammarAST)o; if ( c.getType()==ANTLRParser.ASSIGN ) { t.setOption(c.getChild(0).getText(), (GrammarAST)c.getChild(1)); } else { t.setOption(c.getText(), null); // no arg such as ID } } } /** Return list of (TOKEN_NAME node, 'literal' node) pairs */ public static List> getStringLiteralAliasesFromLexerRules(GrammarRootAST ast) { String[] patterns = { "(RULE %name:TOKEN_REF (BLOCK (ALT %lit:STRING_LITERAL)))", "(RULE %name:TOKEN_REF (BLOCK (ALT %lit:STRING_LITERAL ACTION)))", "(RULE %name:TOKEN_REF (BLOCK (ALT %lit:STRING_LITERAL SEMPRED)))", "(RULE %name:TOKEN_REF (BLOCK (LEXER_ALT_ACTION (ALT %lit:STRING_LITERAL) .)))", "(RULE %name:TOKEN_REF (BLOCK (LEXER_ALT_ACTION (ALT %lit:STRING_LITERAL) . .)))", "(RULE %name:TOKEN_REF (BLOCK (LEXER_ALT_ACTION (ALT %lit:STRING_LITERAL) (LEXER_ACTION_CALL . .))))", "(RULE %name:TOKEN_REF (BLOCK (LEXER_ALT_ACTION (ALT %lit:STRING_LITERAL) . (LEXER_ACTION_CALL . .))))", "(RULE %name:TOKEN_REF (BLOCK (LEXER_ALT_ACTION (ALT %lit:STRING_LITERAL) (LEXER_ACTION_CALL . .) .)))", // TODO: allow doc comment in there }; GrammarASTAdaptor adaptor = new GrammarASTAdaptor(ast.token.getInputStream()); org.antlr.runtime.tree.TreeWizard wiz = new org.antlr.runtime.tree.TreeWizard(adaptor,ANTLRParser.tokenNames); List> lexerRuleToStringLiteral = new ArrayList>(); List ruleNodes = ast.getNodesWithType(ANTLRParser.RULE); if ( ruleNodes==null || ruleNodes.isEmpty() ) return null; for (GrammarAST r : ruleNodes) { //tool.log("grammar", r.toStringTree()); // System.out.println("chk: "+r.toStringTree()); org.antlr.runtime.tree.Tree name = r.getChild(0); if ( name.getType()==ANTLRParser.TOKEN_REF ) { // check rule against patterns boolean isLitRule; for (String pattern : patterns) { isLitRule = defAlias(r, pattern, wiz, lexerRuleToStringLiteral); if ( isLitRule ) break; } // if ( !isLitRule ) System.out.println("no pattern matched"); } } return lexerRuleToStringLiteral; } protected static boolean defAlias(GrammarAST r, String pattern, org.antlr.runtime.tree.TreeWizard wiz, List> lexerRuleToStringLiteral) { HashMap nodes = new HashMap(); if ( wiz.parse(r, pattern, nodes) ) { GrammarAST litNode = (GrammarAST)nodes.get("lit"); GrammarAST nameNode = (GrammarAST)nodes.get("name"); Tuple2 pair = Tuple.create(nameNode, litNode); lexerRuleToStringLiteral.add(pair); return true; } return false; } public Set getStringLiterals() { final Set strings = new LinkedHashSet(); GrammarTreeVisitor collector = new GrammarTreeVisitor() { @Override public void stringRef(TerminalAST ref) { strings.add(ref.getText()); } @Override public ErrorManager getErrorManager() { return tool.errMgr; } }; collector.visitGrammar(ast); return strings; } public void setLookaheadDFA(int decision, DFA lookaheadDFA) { decisionDFAs.put(decision, lookaheadDFA); } public static Map getStateToGrammarRegionMap(GrammarRootAST ast, IntervalSet grammarTokenTypes) { Map stateToGrammarRegionMap = new HashMap(); if ( ast==null ) return stateToGrammarRegionMap; List nodes = ast.getNodesWithType(grammarTokenTypes); for (GrammarAST n : nodes) { if (n.atnState != null) { Interval tokenRegion = Interval.of(n.getTokenStartIndex(), n.getTokenStopIndex()); org.antlr.runtime.tree.Tree ruleNode = null; // RULEs, BLOCKs of transformed recursive rules point to original token interval switch ( n.getType() ) { case ANTLRParser.RULE : ruleNode = n; break; case ANTLRParser.BLOCK : case ANTLRParser.CLOSURE : ruleNode = n.getAncestor(ANTLRParser.RULE); break; } if ( ruleNode instanceof RuleAST ) { String ruleName = ((RuleAST) ruleNode).getRuleName(); Rule r = ast.g.getRule(ruleName); if ( r instanceof LeftRecursiveRule ) { RuleAST originalAST = ((LeftRecursiveRule) r).getOriginalAST(); tokenRegion = Interval.of(originalAST.getTokenStartIndex(), originalAST.getTokenStopIndex()); } } stateToGrammarRegionMap.put(n.atnState.stateNumber, tokenRegion); } } return stateToGrammarRegionMap; } /** Given an ATN state number, return the token index range within the grammar from which that ATN state was derived. */ public Interval getStateToGrammarRegion(int atnStateNumber) { if ( stateToGrammarRegionMap==null ) { stateToGrammarRegionMap = getStateToGrammarRegionMap(ast, null); // map all nodes with non-null atn state ptr } if ( stateToGrammarRegionMap==null ) return Interval.INVALID; return stateToGrammarRegionMap.get(atnStateNumber); } public LexerInterpreter createLexerInterpreter(CharStream input) { if (this.isParser()) { throw new IllegalStateException("A lexer interpreter can only be created for a lexer or combined grammar."); } if (this.isCombined()) { return implicitLexer.createLexerInterpreter(input); } char[] serializedAtn = ATNSerializer.getSerializedAsChars(atn, Arrays.asList(getRuleNames())); ATN deserialized = new ATNDeserializer().deserialize(serializedAtn); List allChannels = new ArrayList(); allChannels.add("DEFAULT_TOKEN_CHANNEL"); allChannels.add("HIDDEN"); allChannels.addAll(channelValueToNameList); return new LexerInterpreter(fileName, getVocabulary(), Arrays.asList(getRuleNames()), allChannels, ((LexerGrammar)this).modes.keySet(), deserialized, input); } /** @since 4.5.1 */ public GrammarParserInterpreter createGrammarParserInterpreter(TokenStream tokenStream) { if (this.isLexer()) { throw new IllegalStateException("A parser interpreter can only be created for a parser or combined grammar."); } char[] serializedAtn = ATNSerializer.getSerializedAsChars(atn, Arrays.asList(getRuleNames())); ATN deserialized = new ATNDeserializer().deserialize(serializedAtn); return new GrammarParserInterpreter(this, deserialized, tokenStream); } public ParserInterpreter createParserInterpreter(TokenStream tokenStream) { if (this.isLexer()) { throw new IllegalStateException("A parser interpreter can only be created for a parser or combined grammar."); } char[] serializedAtn = ATNSerializer.getSerializedAsChars(atn, Arrays.asList(getRuleNames())); ATN deserialized = new ATNDeserializer().deserialize(serializedAtn); return new ParserInterpreter(fileName, getVocabulary(), Arrays.asList(getRuleNames()), deserialized, tokenStream); } protected static class AltLabelVisitor extends GrammarTreeVisitor { private final Map>> labeledAlternatives = new LinkedHashMap>>(); private final List unlabeledAlternatives = new ArrayList(); public AltLabelVisitor(org.antlr.runtime.tree.TreeNodeStream input) { super(input); } public Map>> getLabeledAlternatives() { return labeledAlternatives; } public List getUnlabeledAlternatives() { return unlabeledAlternatives; } @Override public void discoverOuterAlt(AltAST alt) { if (alt.altLabel != null) { List> list = labeledAlternatives.get(alt.altLabel.getText()); if (list == null) { list = new ArrayList>(); labeledAlternatives.put(alt.altLabel.getText(), list); } list.add(Tuple.create(currentOuterAltNumber, alt)); } else { unlabeledAlternatives.add(alt); } } } }




© 2015 - 2024 Weber Informatics LLC | Privacy Policy