Re: [boost] [array] Misc comments

From: Gennaro Prota (gennaro_prota_at_[hidden])
Date: 2006-06-24 08:17:37

On Sat, 24 Jun 2006 12:34:43 +0100, "John Maddock"
<john_at_[hidden]> wrote:

>Gennaro Prota wrote:
>>
>>[...]
>
>Yep that can be simplied quite a bit, done in cvs.  I haven't added the 
>static_cast's: it obfuscates the code and serves no real purpose - no value 
>can actually ever be returned.

Thanks, I could have done that myself, but I always prefer to ask for
the original developer, first, just to be sure I'm not missing
anything.

>
>> NOTE: Going a bit further null_item() is not necessary, as one could
>> use a return *(new T) directly in failed_range_check(), though I
>> haven't thought of all the consequences in the case operator new is
>> overloaded.
>
>Not so sure about that.

Yeah, of course the type of the expression *(new T) would always be T,
but I'm not sure if it may cause code bloat in case the compiler is
not able to see that it is never reached. After a couple of nights of
sleep I find the static T solution preferable.

>
>> * why are the two operator[]() functions range-checked for the case
>> N==0? And what's the point of those BOOST_ASSERT( "out of range" )?
>
>No idea - Alistair?
>
>I've removed them for now since they serve no purpose that I can see: 
>they'll always pass won't they?

Tricky question. IMHO it is undefined behavior. When BOOST_ASSERT
falls back to the standard assert it really depends on the underlying
library. C90 does not require that assert works with expressions
having non-int type. In C99 it shall work for any scalar expression
(which anyway "out of range" is not; it has array type).

We could perhaps protect against this kinds of things by using a
static_cast in assert.hpp (probably documenting that)

--Gennaro.