Re: placement new and alignment
From: PaulOBear (PaulOBear_at_discussions.microsoft.com)
Date: 09/15/04
- Next message: Chrynne Schumptze: "__stdcall / CALLBACK"
- Previous message: Patty O'Dors: "Re: General COM question"
- In reply to: PaulOBear: "Re: placement new and alignment"
- Messages sorted by: [ date ] [ thread ]
Date: Wed, 15 Sep 2004 08:57:18 -0700
Ok, nevermind... Sigh... Do I feel sheepish. :) According to Deep C++, the
address passed in gets magically assigned to the this pointer in the
constructor of the class. Duh.
And if I just include <new>, it's all done for me already!! :^{=}
Thanks,
Paul
"PaulOBear" wrote:
> Hi Igor and Mycroft -
>
> I am having a related problem, and so perhaps this post should be in another
> thread. However, since you two seem to be working in the same muck as I am,
> I thought you might be able to help. :)
>
> The problem:
>
> I am an embedded systems developer on a PowerPC target, and I've been tasked
> with developing the UI on a Windows platform, since we don't have hardware
> (gee, *that's* never happened before! ;). When developing C++ apps in an
> embedded environment, it is often undesirable to use ordinary new and the
> heap. So we like to use placement new. In my experience, this is most
> typically performed by allocating space in a class to contain the object of
> interest:
>
> byte objectBuf[sizeof(SomeClass)];
>
> Then one merely placement new's the class into the buffer:
>
> SomeClass* pSomeClass = reinterpret_cast<SomeClass*>(new (&objectBuf)
> SomeClass());
>
> All the embedded compilers (that I have used) have such an operator new
> defined as part of the compiler. However, now I am trying to use VC++, and
> lo and behold, it's something that I need to provide.
>
> I presume that if I've allocated the space in the context of the code, then
> it will be aligned properly. Is that a correct assumption?
>
> I have read the documentation on placement new which comes with VC++ and
> tells one that one must define their own placement new operator, but it
> doesn't really go into the details of such a transformation (beyond having to
> register it).
>
> I imagine that I would need a declaration something like:
>
> ::operator new(size_t size, void* loc);
>
> And of course, the definition to match, and that's where I really am lost.
> Ok, so now I have a location to instantiate an object, but how do I do that
> w/o invoking a normal new and w/o invoking another placement new (recursively
> until the stack overwrites some vital organs. ;) .
>
> Thanks in Advance,
>
> Paul Ourada
>
> "Igor Tandetnik" wrote:
>
> > "Mycroft Holmes" <m.holmes@nospam.it> wrote in message
> > news:uF6rslEjEHA.2052@TK2MSFTNGP15.phx.gbl
> > >> Again, remember that the standard requires operator new to produce
> > >> storage suitably aligned for any type, not just some type. Thus,
> > >> calculating an alignment of a single type is not particularly useful.
> > >
> > > thsnks for the link, I'm going to read it immediately.
> > > Anyway I want to compute alignment for type T on demand: my program
> > > executes (well... I'm simplifying... ;)
> > >
> > > void* p = cache.gimmeSomeMemory( instance_of<T>() );
> > > return new(p) T;
> >
> > My (admittedly limited) understanding is that this is formally illegal -
> > though likely to work in all practical situations - unless
> > gimmeSomeMemory() returns memory aligned so as to satisfy requirements
> > on operator new (that is, properly aligned for any type). See
> >
> > http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-defects.html#114
> >
> > The standard does not explicitly say how the pointer passed to placement
> > new should be aligned, but it says that 1) placement new simply returns
> > the pointer passed to it and 2) all allocation functions (of which
> > placement new is one) must return a pointer suitably aligned for any
> > type. This seems to make it responsibility of the caller to ensure that
> > only a suitably aligned pointer is ever passed to placement new.
> >
> > > so type T is a single type and statically known and I need to ensure
> > > that gimmeSomeMemory returns a good pointer.
> >
> > Again, formally this does not seem to be sufficient, practically it most
> > likely is.
> > --
> > With best wishes,
> > Igor Tandetnik
> >
> > "For every complex problem, there is a solution that is simple, neat,
> > and wrong." H.L. Mencken
> >
> >
> >
- Next message: Chrynne Schumptze: "__stdcall / CALLBACK"
- Previous message: Patty O'Dors: "Re: General COM question"
- In reply to: PaulOBear: "Re: placement new and alignment"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
