Re: Conflict-markers for shift-reduce conflicts similar to Lezer

[Akim Demaille]

> Le 21 janv. 2020 à 01:04, Adrian Vogelsgesang <address@hidden> a écrit :
> 
> Hi,

Hi Adrian,

> I came across https://lezer.codemirror.net last week and I think they have a 
> really good way of dealing with conflicts:
> In Lezer’s grammars, all conflicts must be marked with “ambiguity markers”. 
> The concept is described on https://lezer.codemirror.net/docs/guide/, 
> subsection “Allowing ambiguity”.

First time I hear about lezer.  I don't know if their approach is unique, but 
it's the first time I see it.

> I think this approach is superior to bison’s “%expect and %expect-rr 
> modifiers on individual rules”

I don't think they address the problem of s/r conflicts.  Their doc shows the 
case of r/r conflicts only.

I had a similar idea, but somewhat different: using conflict identifiers on the 
%expect-rr annotation of rules.  I was about to post it, but it seems I forgot 
to do it :(  No-one had express interest in these annotations anyway 
(https://lists.gnu.org/r/bison-patches/2018-11/msg00055.html).

> because:
> * the ambiguity markers specify at what exact position inside the rule the 
> conflict occurs
> * using the ambiguity markers, it is exactly documented which positions 
> conflict with which other rules, instead of only the number of conflicts

I agree counts are not a satisfying option.  Yet I think that putting the 
markers inside the rules themselves is too intrusive.  It's much more accurate, 
granted, but I felt that passing names to %expect-rr was enough.

> What do you think about this feature? Would something like this also be 
> valuable for bison?
> How hard would it be to implement similar functionality in bison? (I have no 
> clue about bison’s C-parts, yet…)

I have not evaluated how effort that would be.  Especially the case where 
position matters.