'tree' as name of a parser rule causes syntax error.

[mateuszpawlik]

A simple grammar ('g.g4'):

grammar g;
tree : 'X';

causes the following error

java -jar antlr-4.7-complete.jar g.g4 
error(50): g.g4::: syntax error: mismatched character ':' expecting 'g'
error(50): g.g4:2:7: syntax error: ''X'' came as a complete surprise to me

Removing the space between 'tree' and ':' or changing 'tree' to other rule name helps.

I've found that 'tree' was a keyword in antlr v3. But the reference book of v4 doesn't list it any more.

If 'tree' is a reserved keyword, the error message is confusing.

[moy]

It expects a g, and when you give that g (tree g:), it asks for an r, then a a, and so on until you type tree grammar completely, in which case it throws syntax error: 'tree grammar' came as a complete surprise to me. It seems ANTLR is trying to special-case tree grammar as a kind of two-words keyword. Weird.

Workarounds:

• Don't use tree as a rule name.

• Stick the : to tree as in tree:, and the error disappears.

[moy]

It seems ANTLR is trying to special-case tree grammar as a kind of two-words keyword.

Indeed:

antlr4/tool/src/org/antlr/v4/parse/ANTLRLexer.g

Line 449 in 432022f

TREE_GRAMMAR : 'tree' WSNLCHARS* 'grammar' ;

Actually, if I read correctly, this rule is only used to throw a relevant error message when the user writes tree grammar (not supported in ANTLRv4):

antlr4/tool/src/org/antlr/v4/parse/ANTLRParser.g

Lines 200 to 210 in 432022f

	if ( $tg!=null ) throw new v3TreeGrammarException(tg);
	if ( $t!=null ) ((GrammarRootAST)$tree).grammarType = $t.type;
	else ((GrammarRootAST)$tree).grammarType=COMBINED;
	}
	: ( t=LEXER g=GRAMMAR -> GRAMMAR<GrammarRootAST>[$g, "LEXER_GRAMMAR", getTokenStream()]
	| // A standalone parser specification
	t=PARSER g=GRAMMAR -> GRAMMAR<GrammarRootAST>[$g, "PARSER_GRAMMAR", getTokenStream()]
	// A combined lexer and parser specification
	| g=GRAMMAR -> GRAMMAR<GrammarRootAST>[$g, "COMBINED_GRAMMAR", getTokenStream()]
	| tg=TREE_GRAMMAR

Still, I don't understand why using a prefix (tree ) of a token raises an error.