Re: About %destructor is c++ mode

[Min Wang]

HI

@Hans.  thanks

so if every rule has an explicit action, the issue could be avoided, right?


min

On Fri, Aug 19, 2016 at 4:15 AM, Hans Åberg <address@hidden> wrote:

>
> > On 19 Aug 2016, at 05:50, Min Wang <address@hidden> wrote:
>
> > on another thread at:
> > http://lists.gnu.org/archive/html/bug-bison/2016-08/msg00006.html
>
> I am on that list, too.
>
> > it seems I could use std::unique_ptr as the semantic value , but need one
> > hack
> >
> > -      return *new (yyas_<T> ()) T (t);
> > +      return *new (yyas_<T> ()) T (std::move((T&)t));
>
> The issue is that in the original C parser, the implicit $$ = $1 that
> occurs when not writing an explicit action is always executed first before
> any other eventual action. In C this is fine, and not so so expensive as
> there are no copy constructors. This implementation was carried over at
> least in the original C++ parser.
>
> When you then use the std::unique_ptr in $1 it will first implicitly move
> over to $$, and if trying to access $1 explicitly in an action, the pointer
> is dead.
>
> So the hack depends on what the C++ implementation currently is, which may
> change.
>
>
>


-- 
http://www.comrite.com