Exercises on Bison

Error Recovery

There is one single important feature of Yacc parsers that we have not revealed here: the use of the error token to recover from errors. The general idea is that when the parser finds an error, it tries to find the nearest rule which claims to be ready to return to normal processing after having thrown away embarrassing symbols. For instance:

          exp: '(' error ')'
          

But including recovery means you must pretend everything went right. In particular, an exp is expected to have a value, this rule must provide a valid $$. In our case, the absence of a rule means proceeding with a random integer, but in other applications it may be a wandering pointer!

Equip yleval with proper error recovery. Voir Writing rules for error recovery, for detailed explanations on error and yyerrok.

LR(2) Grammars

Consider the following excerpt of the grammar of Bison grammars:

          grammar: rule
                 | grammar rule
          rule: symbol ':' ride-hand-side
          right-hand-side: /* Nothing */
                         | right-hand-side symbol
          symbol: 's'
          

Convince yourself that Bison cannot accept this grammar either by (i) feeding it to bison and understanding the conflict, (ii) drawing the LR(1) automaton and understanding the conflict, (iii) looking at the strategy you, human, use to read the text s : s s s : s : s.

Yes, the bottom line is that, ironically, Yacc and Bison cannot use themselves! (More rigorously, they cannot handle the natural grammar describing their syntax.)

Once you have understood the problem, (i) imagine you were the designer of Yacc and propose the grammar for a better syntax, (ii) stop dreaming, you are not Steven C. Johnson, so you can't change this grammar: instead, imagine how the scanner could help the parser to distinguish a symbol at the right hand side of a rule from the one at the left hand side.