Re: push parser

[Paul Eggert]

Bob Rossi <address@hidden> writes:

>> Why the #ifdef on YYLTYPE_IS_TRIVIAL?  YYLTYPE_IS_TRIVIAL tells
>> us whether yyltype is trivial (i.e., does not need attention by
>> C++ storage managers), not whether it exists.
>
> OK, I need to determine if YYLTYPE is exists. What's the best way 
> to do that?

b4_locations_if.

>>        status = yypushparse (ctx, ch, yylval, yylloc);
> ...
> I'm not sure that I like passing all the values into yypushparse though.
> The problem is, not everyone wants to pass all the values in. We would
> force them to pass values in that they don't care about which could
> be confusing.

They always need to pass in ctx, ch, right?  And the ", yylloc"
will be required if b4_locations_if says you're using locations.
So the only questionable argument will be yylval.

One way to address this issue would be to have two entry points:

       status = yypushparse (ctx, ch, yylloc);

for tokens that have no semantic value, and

       status = yypushparseval (ctx, ch, yylval, yylloc);

for tokens that do.

Another possibility, which will avoid a copy in some cases if semantic
values are large, is to pass a pointer:

       status = yypushparse (ctx, ch, &yylval, yylloc);

where you pass a NULL pointer if the token has no semantic value.  If
the copying issue is of concern, it may also make sense to pass yylloc's
address too:

       status = yypushparse (ctx, ch, &yylval, &yylloc);

I don't know whether the copying concern is enough to affect
performance, though.