2010/12/3 Justin Chase <justin.m.ch...@gmail.com> > From there the precedence of all the operators is determined by the > ordering but can have nested lower precedence rules. This same principle > applies for binary arithmetic expressions and can also be used for things > like expression chaining (e.g. x[0].y(z)() ) > > But if you were to introduce indirect left recursion into this it would > either cause this pattern to no longer work or be so complex I'm not sure > you would be able to reason about unexpected results very well. I would be > interested to hear evidence otherwise, or to hear about features that are > enabled by iLR that you cannot solve without. > > That's a pretty interesting question. I guess you could always turn LR or iLR rules into sets of non-LR rules, and still match left-precedence right with some side-work, so I don't think there is anywhere you really NEED iLR or even LR. It's just a lot more convenient in some cases.
>From what I've found out concerning PEGs, I still have difficulties believing what you stated, that a combination of the ordering of choices and (i)LR can define a solid precedence system. Look for instance at this grammar : EXP <- EXP '*' EXP | EXP '+' EXP | NUM NUM <- [0-9]+ Following the two PEGs axiom of greedy match AND ordered choice, I haven't found a way to get '*' and '+' precedence to match the intuition in every possible cases. It may exist, but it's not implied by most PEG-support-for-LR algorithms I saw. For instance : 1+1*1+1 should parse (((1)+(1*1))+1), and 1*1+1*1 should parse ((1*1)+(1*1)) Of course if you have an algorithm that does so, I absolutely need and want to look at it :). Cheers, Alex PS : Laurence Tratt did some work about associativity in LR-PEG, it's surely worth taking a look at it. I don't think precedence is managed though (I'm asking him right away). http://tratt.net/laurie/research/publications/ On Fri, Dec 3, 2010 at 3:18 AM, Alex Repain <alex.rep...@gmail.com> wrote: > >> >> >> 2010/12/3 Gordon Tisher <gor...@balafon.net> >> >>> Hi Justin, >>> >> >> >>> LR is useful for handling operator precedence, for example. I'm afraid >>> that I have no evidence that indirect left-recursion does not result in >>> increased complexity -- being able to handle the increased complexity >>> was the attraction for me in implementing the algorithm. >>> >>> >> Is there some study work about how LR is supposed to help setting operator >> precedences ? >> >> I have worked around with left-recursion support for PEGs, and I still >> have doubts that PEG+LR really offers 'native' support for operator >> precedence... >> >> Cheers, >> Alex >> >> >> >>> Cheers, >>> >>> Gordon >>> >>> -- >>> Gordon Tisher >>> http://balafon.net >>> >>> >>> _______________________________________________ >>> PEG mailing list >>> PEG@lists.csail.mit.edu >>> https://lists.csail.mit.edu/mailman/listinfo/peg >>> >> >> > > > -- > Justin Chase > http://www.justinmchase.com >
_______________________________________________ PEG mailing list PEG@lists.csail.mit.edu https://lists.csail.mit.edu/mailman/listinfo/peg
