On Sun, Jul 15, 2012 at 3:35 PM, Benjamin Thaut <c...@benjamin-thaut.de>wrote:
> Am 15.07.2012 13:24, schrieb Gor Gyolchanyan: > >> On Sun, Jul 15, 2012 at 1:41 PM, Benjamin Thaut <c...@benjamin-thaut.de >> >> <mailto:c...@benjamin-thaut.de**>> wrote: >> >> Am 15.07.2012 02:02, schrieb Timon Gehr: >> >> On 07/15/2012 12:55 AM, Benjamin Thaut wrote: >> >> Am 14.07.2012 19:30, schrieb Gor Gyolchanyan: >> >> On Sat, Jul 14, 2012 at 9:23 PM, Benjamin Thaut >> <c...@benjamin-thaut.de <mailto:c...@benjamin-thaut.de**> >> <mailto:c...@benjamin-thaut.de >> >> <mailto:c...@benjamin-thaut.de**>__>> wrote: >> >> Am 14.07.2012 19:21, schrieb kenji hara: >> >> 2012/7/15 Benjamin Thaut <c...@benjamin-thaut.de >> <mailto:c...@benjamin-thaut.de**> >> <mailto:c...@benjamin-thaut.de >> <mailto:c...@benjamin-thaut.de**>__>>: >> >> >> The only problem about this is: >> >> class Fruit >> { >> class Seed { >> void SetFruit(Fruit fruit) >> { >> this.outer = fruit; >> >> >> Setting to pseudo variable 'outer' should be rejected in >> compilation. >> Please report it to bugzilla. >> >> } >> } >> } >> >> class Apple: Fruit >> { >> void AppleOnlyMethod(){ ... } >> >> class AppleSeed: Fruit.Seed { >> void DoSomething() >> { >> AppleOnlyMethod(); >> } >> } >> >> auto GetNewSeed() { return new AppleSeed(); } >> } >> >> auto apple = new Apple(); >> auto seed = apple.GetNewSeed(); >> seed.SetFruit(new Fruit()); >> seed.DoSomething(); //what happens here? >> >> Kind Regards >> Benjamin Thaut >> >> >> Kenji Hara >> >> >> I will not report this, beacuse it will break my custom >> new operator >> (template) for inner classes ;-) >> >> Kind Regards >> Benjamin Thaut >> >> >> That's most unwise, because if it's not supposed to be >> like that it will >> get fixed anyway, so you better start replacing your >> custom new >> operator. >> >> -- >> Bye, >> Gor Gyolchanyan. >> >> >> Replacing my custom new operator exactly by what? >> Overloading the build >> in new is deprecated... >> >> >> class C{ >> class D{ >> >> } >> } >> void main(){ >> auto c = new C; >> auto buf = new void[__traits(__**classInstanceSize, C.D)]; >> >> (cast(byte[])buf)[] = typeid(C.D).init[]; >> auto d = cast(C.D)buf.ptr; >> static if(is(typeof(d.__ctor()))) d.__ctor(); >> enum offset=d.outer.offsetof; >> static assert(offset%C.sizeof==0); >> (cast(C[])buf)[offset/C.__**sizeof]=c; >> >> assert(d.outer is c); >> } >> >> >> Yes of course I can assign the reference by computing the address >> and then using that to assign it. But the point here is, that this >> is currently the only bug that is hepling me in what I'm doing, and >> I rather want the 7 other bugs I reported so far to be fixed, rather >> then the one that helps me. >> >> Kind Regards >> Benjamin Thaut >> >> >> Not being able to assign is not about some sort of authoritative >> forbidding. It's about not breaking a working mechanism. You can't cast >> a function pointer into a class object not because the type system says >> so (which it does), but because that'll result in an undefined behavior, >> which will result in an immediate crash in the best scenario. Go ahead >> and change the outer via those hacks. But then don't get surprised when >> your program crashes for no reason (because the compiler assumed it not >> to change and ended up being wrong). >> >> -- >> Bye, >> Gor Gyolchanyan. >> > > Before your change suggestion it was absolutley save to assign to outer. > Thats the whole point why I posted this here. I very well know that the bug > will get fixed some time. But before your change suggestion it was not even > a bug. > > Kind Regards > Benjamin Thaut > > A bug remains a bug whether it was discovered or not. If it wasn't - it would never get discovered. Even if the bug didn't get discovered for a long time, making designs around bugs is a very bad idea precisely for this reason. But it's not entirely your fault, because there's no reliable reference to D at the moment, so it's very hard to determine what is and is not supposed to work. That's one of the problems, that drive new commers away: lack of a reference material. -- Bye, Gor Gyolchanyan.
