precedence!

As it says, it's state 29. Hello, state 29. First the matching rules are listed. Anything after the `expr:' or `|' is part of the rule that's matched, and the lone `.' indicates where it matches.. for example:

• expr . '.' funccall -- we've just seen an expr, and there's a full stop followed by a function call ahead
• expr . '*' expr -- we just saw an expr, and we're seeing an asterisk and another expr afterward.
• '-' expr . -- we just saw a negation operator and an expr (what lies ahead is irrelevant)

Next come two stanzas; the first lists shifts, the second lists reduces (I explained shifts and reduces briefly above). The first token is the what comes immediately next (after the .), and the rest explains how it'll handle the action. So, if we see a ^ (which currently is an exponentiation operator, though this is incorrect in Ruby), it'll shift. This is because exponentiation is right-associating: a^b^c is a^(b^c). If we reduced after seeing an expr and before we see a ^, we'd end up getting a^b after a while, and then a^b would be the reduced expr, and that'd become (a^b)^c. Bad! Note that none of the other left-associating rules are explicitly mentioned; since they all left-associate, they're handled by the $default rule to reduce.

The evil here is that '.' is also shifting. That means instead of reducing expr into a funccall or whatever we like, we continue to shift expr.something. We keep shifting until we reduce, and that'll be when we see something stupid like (b).c, and that reduces nicely into an expr of its own, and then expr expr will reduce into funccall! Mayhem!

The important thing to notice here is that it's also detected that we could reduce '.'. By default, bison handles shift/reduce conflicts by shifting, thus capturing the greatest "area", and associating elements as closely as possible. It shows that the reduce is not being used by surrounding it in brackets. Let's have a quick survey of the other rules:
Here's state 40. We can see that, depending on what happens next, it shifts. Remember kids, that expr on the left is only the most recent expr, since we didn't reduce earlier! That is to say, a(b) was never reduced into one expr, so it still exists as two; a, (b), and thus (b) is the expr before the '.' in these rules.

To back up a little, I don't think state 29 is the cause of all my problems, but similar states elsewhere which don't reduce first are a headache.