Re: [Boost-users] [endian] Endian types in union

From: Dave Dribin (dave-ml_at_[hidden])
Date: 2006-07-20 18:19:59

On Jul 20, 2006, at 4:40 PM, me22 wrote:
> I don't think this is a worry. varargs aren't typesafe and can
> certainly be avoided.

printf-like interfaces are still used quite a bit.  The HFS+ code I  
was writing was actually a device driver.  The kernel debugging print  
functions had printf-like interfaces, which is the only reason I know  
this is an issue.

> The .get is much better.  operator() is for calling functions, while
> .get is already used in many places (smart pointers, for example) for
> that purpose.

Makes sense.  I can live with .get(), and if it's consistent with  
other libraries, then even better.

> However, neither is needed. (int)x or boost::implicit_cast<int>(x)
> will both work fine.  Of course, this is assuming that the value in x
> fits properly in an int.  Better would be typeof_x::value_type, but
> then you can't be sure that your format string matches, which is the
> same point as varargs being evil in the first place.

The issue with using either cast is I have to remember the correct  
type to use in the cast.  By using a .get(), I don't have to worry  
about it, and it's less repeated information.  And as bonus, a  
compiler that recognizes format strings will warn me if I use an  
incorrect format string:

     big8_t y = big8_t::init(5);
     printf("%d\n", y.get());

gcc give's me this warning:

   warning: format '%d' expects type 'int', but argument 2 has type  
'int_least64_t'

If I just instinctively cast to int, I'd silently lose information.   
The warning gives me a chance to use '%lld'.

> Much better to implement it like this:
>          T get() const { return *this; }
> and not duplicate knowledge.

Agreed.  Much better, thanks.

BTW, I started a thread on Boost-dev, as well, since that may be the  
more appropriate list.

-Dave