On Mon, Feb 3, 2020 at 1:29 AM aotto <aotto1968@xxxxxxxxxxx> wrote: > Hi, > > this is a test: https://godbolt.org/z/8PpX8W > why is method_base(&aO->obj, 1); so much worse than method_base(aO, 1); it's the same resulting code... method_base(aO, 1); mov rax, QWORD PTR [rbp-8] mov esi, 1 mov rdi, rax call method_base method_base(&aO->obj, 1); mov rax, QWORD PTR [rbp-8] mov esi, 1 mov rdi, rax call method_base and there's no forced cast.... > > > On 03.02.20 10:07, J Decker wrote: > > > > On Mon, Feb 3, 2020 at 12:23 AM aotto <aotto1968@xxxxxxxxxxx> wrote: > >> Hi, >> >> there is an easy test, if Stefan is right and I'm wrong. >> >> 1. cast an INVALID Pointer to "aObj" >> > how would friend prevent an invalid pointer pass? > > >> 2. call "method_base(&aObj->obj,...)" >> >> -> if stefan is right than "method_base" is called with an INVALID aObj >> as Pointer >> -> if I am right than your software will CRASH :-) >> > given 1) sure, true in every case, doesn't make stephen wrong. > >> >> that is the reason why I require the "frienddef". >> >> >> mfg AO >> >> >> >> On 03.02.20 09:16, aotto wrote: >> > Hi, >> > >> > Can someone confirm that *aObj->obj does NOT create any extray "code" >> > and does NOT do any >> > POINTER resolution? -> if so than "gcc" already uses the "friend" >> > logic I require :-) >> > > > https://godbolt.org/ this is a good place to collaborate on shared > code... it shows the generated assembly... > https://godbolt.org/z/X5mRYg is something resembling your case; > especially the function call of g()... I think probably an optimized > compile would generate the same for the assignments to tmp also... > > So you just want to shortcut that the any first struct in a containing > pointer is valid as a parameter, and an implicit cast to the inner type > should be done? The internal browsing of what members are in a type is not > very convenient, and would require quite a few lines to search... > > struct a { > struct b { > struct c { > int d; > }C; > }B; > }A; > > void g( struct c* pC ) {} > void h( struct b* pB) {} > > void f(void){ > g( &A ); // instead of g( &A.B.C ); > h( &A ); // instead of h( &A.B ); > } > > > > > > >> > → I don't think that "gcc" do this already >> >> Hi, the following scenario has a "definition hole" in the "C" language >> >> code example: >> >> ------------------------- >> struct base { >> ... >> }; >> >> struct A { >> struct base obj; >> ... >> } aObj; >> >> struct B { >> struct base obj; >> ... >> } bObj; >> >> void method_base (struct base * hdl, ...); >> >> method_base(&aObj, ...) >> method_base(&bObj, ...) > > ------------------------ >> >> >> -----Ursprüngliche Nachricht----- >> Von: aotto <aotto1968@xxxxxxxxxxx> >> Gesendet: Montag, 3. Februar 2020 09:02 >> An: Stefan Franke <stefan@xxxxxxxxx> >> Betreff: Re: AW: Feature request for "friendship" of pointers in "C" >> >> This is not the same! >> >> 1. usually the aObj is a Pointer: >> -> &aObj.obj => &(aObj->obj) >> 2. this DOUBLE indirect access creates definitely a CRASH if >> aObj is invalid or freed >> 3. I use a internal a Pointer typecheck… where I can detect if >> "my" pointer is valid or freed… but a "&(aObj->obj)" you >> don't survive. >> >> mfg >> >> * Please reply at the bottom. >> >> Ad 1: there is no need for '(' ')' if it's a pointer >> >> &aObj->obj >> >> >> Ad 2: this is not a double indirect access, the resulting address is the >> same as aObj as long obj is the first element of your struct, only the >> type changes >> >> (void *)aObj == (void*)&aObj->obj >> >> Ad 3: since the address is the same, your check will work. >> >> Your issues aren't related to gcc - you might consider improving your C >> skills. >> >> Cheers >> >> Stefan >> >> >
