Parens

Dee Girl deegirl at noreply.com
Sat May 10 09:26:14 PDT 2008


Bruce Adams Wrote:

> On Sat, 10 May 2008 06:51:21 +0100, Dee Girl <deegirl at noreply.com> wrote:
> 
> > Janice Caron Wrote:
> >
> >> On 10/05/2008, Dee Girl <deegirl at noreply.com> wrote:
> >> >  void main(string[] args)
> >> >  {
> >> >     auto c = new const C;
> >> >  }
> >> >
> >> >  Does not compile. But if I change to const(C) it does compile. It is  
> >> a bit unusual that parens make such a big difference. Thank you, Dee  
> >> Girl
> >>
> >> When it comes to parens-version versus non-parens-version, it's not
> >> really a problem if one of the two doesn't compile. It would be a
> >> bigger problem if both compile, but behave differently.
> >>
> >> But I agree with you. "new const C" should behave like "new const(C)".
> >>
> >> However, what you probably should have written is:
> >>
> >>     const c = new C;
> >>
> >> because "auto" is what you use in the absence of any other attribute.
> >> In this case, it suffices to create a new C, and then assign it to a
> >> const-thereafter variable.
> >
> > I respectfully disagree, Janice-san. The const subject is of big  
> > interest to me and I read many times on the website about it. I think in  
> > general it is better to write new const(C) but not new C and then use it  
> > to assign a const. It is similar to invariant(C) and it is impossible to  
> > construct a new C and then assign it to invariant(C) object. In Andres  
> > document from ACCU he shows how invariant constructor must know that  
> > object being constructed is invariant. And the same a const constructor  
> > could know that the object being constructed is const and use that  
> > information. Thank you, Dee Girl
> 
> It could know and use that but is that really a good idea?
> I can imagine a class which had an invariant constructor that set up some  
> sort
> of mutex to prevent modification whereas its mutable counterpart did not.
> I would hate to try and unravel code like that while debugging.
> Its the sort of thing only the compier should need to know about.

There could be a simpler case. A const C can have more sharing than an not const C. If the constructor does not know this will be const then it must assume this will be mutated.

And there is one other case. If a const(C) is constructed the invariant constructor can work because invariant is a subtype of const.

> That being the case the equals is just syntactic sugar like in C++.
> 
> const c = new C;
> 
> is identically equivalent to
> 
> const c(new C);
> 
> and:
> 
> const c(new const c);

Is that true? I tried to compile these codes and it does not work. Thank you, Dee Girl



More information about the Digitalmars-d mailing list