Re: $1 of 'whatever' has no declared type

[Philip Herron]

> Joel E. Denny wrote:
> > On Sat, 21 Nov 2009, brad wrote:
> >
> >  
> > > I have a Bison grammar that I've been working on for a while now, 
> > > and I never
> > > had this problem until I recently upgraded Ubuntu. Now when I try to 
> > > run the
> > > grammar through Bison I get loads of "$1 of 'whatever' has no 
> > > declared type"
> > > errors. Has something recently changed in Bison that breaks old 
> > > grammars?
> > >
> > > You can see my grammar here:
> > > http://gk3tools.svn.sourceforge.net/viewvc/gk3tools/trunk/sheep/sheepParser.y?revision=232&view=markup 
> > >
> > >     
> >
> > I haven't taken the time to try out your grammar, but I took a quick 
> > glance and spotted a problem that's been reported before.
> >
> > You have this:
> >
> >   #define YYSTYPE SheepCodeTreeNode*
> >
> > When you set YYSTYPE to a single type as above, then it doesn't make 
> > sense to declare semantic types, which are fields of a union 
> > YYSTYPE.  However, you have this:
> >
> >   %token <id> IDENTIFIER <id> LOCALIDENTIFIER <intVal> INTEGER 
> > <floatVal>          FLOAT <stringVal> STRING
> >
> > Bison 2.3 ignored this.  Bison 2.3a and later logically assume you 
> > want a union when you use a <...> construct.  Moreover, POSIX 
> > requires the corrected behavior as discussed in the NEWS entry for 
> > Bison 2.3a.
> >
> > You should remove the redundant <...> constructs.
> >
> >   
Hey guys,

Chukcing in my $0.02, but i have always found using a:

%union{
char *id;
...
}

Then:

%token <id> IDENTIFIER

or

%type <id> rule_names

So then you don't have to go through your non-terminals or productions 
assigning types on the fly because it should already be resolved :).

But really its the same idea just different way, though doing #define 
YYSTYPE is only really ok if you only have one rule return type, but it 
kind of sucks when you want rules to return different types. But i guess 
it depends what you want :)

--Phil