Antlr pattern matching and lexer modes

I am trying to compile a pattern for html grammar. The code below shows how to parse a string containing htmlAttributeRule:

String code = "href=\"val\"";
CharStream chars = CharStreams.fromString(code);
Lexer lexer = new HTMLLexer(chars);
lexer.pushMode(HTMLLexer.TAG);
TokenStream tokens = new CommonTokenStream(lexer);
HTMLParser parser = new HTMLParser(tokens);
parser.htmlAttribute();

But when i'm trying to:

ParseTreePatternMatcher matcher = new ParseTreePatternMatcher(lexer, parser);
matcher.compile(code, HTMLParser.RULE_htmlAttribute);

it fails with error:

line 1:0 no viable alternative at input 'href="val"'

org.antlr.v4.runtime.NoViableAltException
at org.antlr.v4.runtime.atn.ParserATNSimulator.noViableAlt(ParserATNSimulator.java:2026)
at org.antlr.v4.runtime.atn.ParserATNSimulator.execATN(ParserATNSimulator.java:467)
at org.antlr.v4.runtime.atn.ParserATNSimulator.adaptivePredict(ParserATNSimulator.java:393)
at org.antlr.v4.runtime.ParserInterpreter.visitDecisionState(ParserInterpreter.java:316)
at org.antlr.v4.runtime.ParserInterpreter.visitState(ParserInterpreter.java:223)
at org.antlr.v4.runtime.ParserInterpreter.parse(ParserInterpreter.java:194)
at org.antlr.v4.runtime.tree.pattern.ParseTreePatternMatcher.compile(ParseTreePatternMatcher.java:205)

When i tried to:

List<? extends Token> tokenList = matcher.tokenize(code);

The result contained a single token, the same as when using the lexer with DEFAULT_MODE. Is there some way to fix this?

The problem was the following code from ParseTreePatternMatcher::tokenize:

TextChunk textChunk = (TextChunk)chunk;
ANTLRInputStream in = new ANTLRInputStream(textChunk.getText());
lexer.setInputStream(in);
Token t = lexer.nextToken();

Lexer::setInputStream clears _modeStack and sets _mode to 0. One possible solution is to extend ParseTreePatternMatcher, override method tokenize and insert lexer.pushMode(lexerMode) after lexer.setInputStream(in):

TextChunk textChunk = (TextChunk)chunk;
ANTLRInputStream in = new ANTLRInputStream(textChunk.getText());
lexer.setInputStream(in);
lexer.pushMode(lexerMode);
Token t = lexer.nextToken();

But method tokenize uses Chunk and TextChunk which cannot be accesses from outsize package, so we are obligated to define the extension class in the same package as ParseTreePatternMatcher.

Another solution i'm considering is to modify byte code of the method using ASM.