We added some tokenizing errors in Parser. But it is never exposed to users when the lexer says I have m_error. We should move tokenizing errors from Parser to Lexer.