Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
> I just looked at the bison manual at gnu.org and also at `info bison` on
> my local machine, and couldn't see any reference to semicolons being
> optional at the end of a rule. Under the heading "Syntax of Grammar
> Rules" it says this:
> A Bison grammar rule has the following general form:
> RESULT: COMPONENTS...;
> Making it optional without putting that in the manual is just awful.
Yeah. I wonder if they removed that info in 1.34 and failed to
put it back in 1.875?
Anyway, I'm of the opinion that omitting the semi is poor style
and our tools should insist on it even if Bison does not. Thus,
I think the correct fix is for the scripts to complain about a
missing semi, not cope.
My initial look at parse.pl last night left me feeling pretty
disheartened about its robustness in general --- for example,
it looks like { } /* or */ inside a string literal or Bison
character token would break it completely, because it wouldn't
distinguish those cases from the same things outside a string.
It's just luck we haven't broken it yet (or, perhaps, we have
and nobody exercised the relevant productions yet?).
Probably, somebody who's a better Perl programmer than me
ought to take point on improving that.
regards, tom lane