Path: csiph.com!fu-berlin.de!uni-berlin.de!individual.net!not-for-mail
From: Bo Persson <bop@gmb.dk>
Newsgroups: comp.lang.c++
Subject: Re: This will work or not?
Date: Wed, 31 Aug 2016 18:53:00 +0200
Lines: 72
Message-ID: <e2ogbcFb2uiU1@mid.individual.net>
References: <20160831145918.45a0c396@maxa-pc> <nq6ljf$65e$1@dont-email.me>
Mime-Version: 1.0
Content-Type: text/plain; charset=windows-1252; format=flowed
Content-Transfer-Encoding: 7bit
X-Trace: individual.net pdYQmixtbetL2bDOpyeaqQ3Ziri8dFX1OeVV80abYVlMvFeozy
Cancel-Lock: sha1:Kddsp568+e1azTaPnLKU7vPi40c=
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:45.0) Gecko/20100101 Thunderbird/45.3.0
In-Reply-To: <nq6ljf$65e$1@dont-email.me>
Xref: csiph.com comp.lang.c++:44639

On 2016-08-31 15:20, David Brown wrote:
> On 31/08/16 14:59, Melzzzzz wrote:
>> Code in question is like this:
>>
>> struct Problem{
>> 	uint32_t a;
>> 	uint32_t b;
>> 	uint64_t c;
>> 	uint32_t d;
>> 	bool check_it() {
>> 		for (uint32_t* i = &a;i != &d; ++i)
>> 			if (*i>0) return true;
>> 		return false;
>> 	}
>>
>> };
>>
>> I am not sure if this will work, so if someone has clear explanation
>> why it will work (or not) , I would be grateful.
>>
>
> I don't think you are guaranteed that a, b, c, and d follow each other
> in the structure layout (though it is very likely to be the case).  With
> a pointer to a single item &a, incrementing the pointer is undefined
> behaviour - as is accessing *i.  And accessing the uint64_t member using
> an uint32_t* pointer contradicts aliasing restrictions.
>
> All in all, the compiler is free to assume that this code is full of
> undefined behaviour, therefore you don't really care what it does -
> therefore it can all be optimised to "return true", "return false", or
> even "launch nasal daemons".  (This last option is legal for the
> compiler, but thankfully unlikely in practice.)
>
> You can make it legal with:
>
> struct Problem {
> 	union {
> 		struct {
> 			uint32_t a;
> 			uint32_t b;
> 			uint64_t c;
> 			uint32_t d;
> 		}
> 		uint32_t as_array[5];
> 	}
> 	bool check_it() {
> 		for (uint32_t* i = &as_array[0]; i != &as_array[4]; ++i)
> 			if (*i > 0) return true;
> 		return false;
> 	}
> }
>
> (Any issues regarding anonymous unions/structs are left as an exercise
> for the reader :-)
>

This might work in C, but a problem is that the C++ standard says:

9.3 Unions
"In a union, a non-static data member is active if its name refers to an 
object whose lifetime has begun and has not ended (3.8). At most one of 
the non-static data members of an object of union type can be active at 
any time, that is, the value of at most one of the non-static data 
members can be stored in a union at any time. "

So, if the struct with a, b, c, and d is active, as_array is not 
"stored" and cannot be read. You can't read an object that isn't there!



     Bo Persson