On 30.01.2014 16:35, Stefan Monnier wrote:
>> Warning (smie): Conflict: . </= .
>> Warning (smie): Conflict: . </= .
>
> The smie-precs->prec2 part of your grammar says that "." = "." (because
> of the new (assoc ".") you add there).
>
> But the BNF part of your grammar says "." < "." (i.e. "foo . bar . baz"
> is parsed as "foo . (bar . baz)" because of (id "." exp).

Fantastic, thanks for the explanation. With (assoc ".") changed to 
(right "."), the warning is gone.

I've also fixed the precedence of "." vs "? :" and modified a test case 
respectively. See the new patch attached.

Do you think it's not too late in the feature freeze to install a change 
like this, or should we wait until the trunk reopens?

We have a decent number of test cases, so it's not like the patch is 
likely to introduce major breakage.

> Apparently there is no more conflict in the BNF in this respect so the
> disambiguation constraint (assoc ".") that's passed to it is ignored.

Without that constraint there, SMIE showed warnings about ambiguities 
between ".", "," and "=". It somehow ignored the associativity, though.

The warnings with the previous patch also were gone when only the second 
one (assoc ".") were changed to (right "."). The first one, in the BNF 
disambiguation part, could be either.

Bozhidar, would you care to test the new patch?