Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!microsoft!jimad
From: jimad@microsoft.UUCP (Jim ADCOCK)
Newsgroups: comp.lang.c++
Subject: Re: Pointer conversion macro brainteaser
Message-ID: <56042@microsoft.UUCP>
Date: 24 Jul 90 18:32:51 GMT
References: <43645@cornell.UUCP>
Reply-To: jimad@microsoft.UUCP (Jim ADCOCK)
Distribution: comp
Organization: Microsoft Corp., Redmond WA
Lines: 39

In article <43645@cornell.UUCP> peierls@cs.cornell.edu (Tim Peierls) writes:
|The question is, for all you spec hackers, under what conditions is
|this macro guaranteed by E&S to work?  By "conditions" I mean nasty
|things like implementation defined behavior, and by "work" I mean conform
|to the description below in some useful if not rigorous way.
|
|#define bp2dp(B,D,p)  ((p)?((D*)(((char*)(p))+1-((size_t)(B*)(D*)(void*)1))):0)

If a coding trick uses any constructs considered by E&S to be "implementation
dependent," then any "guarantees" from E&S are off.  I don't have an E&S in 
front of me at the moment, but here's a short list of a few of the
implementation dependencies I believe I see in this macro:

* cast of small integer to void*
* cast back of void* to something different than originally cast to void*
* cast of B* to size_t
* cast of B* to char*
* use of odd address for structure pointer
* cast of char* to D*

|It works for cfront 2.0 and probably everywhere else.  

I disagree with both the first part and the second part of this statement.

When I tried this macro on a cfront 2.0 derivative with B a virtual base
class of D, the code compiled silently, then crashed at runtime.  The 
reason it crashed was that in the case of a virtual base class, a D instance
contains a pointer to the virtual base class B instance.  Casting from D* to
B* requires an actual evaluation of the pointer -- which resides in the 
hypothetical structure residing at address "1"

Since the cfront implementation of virtual base classes seems reasonable
[assuming one feels virtual base classes are reasonable :-] it follows
that this macro will probably fail everywhere else too.

|A reasonable translator or compiler should have no problem turning the entire
|expression into a compare and a subtract.  

Again, not true if a class uses a virtual base.