NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/lang/c/19329

Internet Message Format | 1993-01-06 | 2.0 KB

Path: sparky!uunet!mcsun!sun4nl!and!jos
From: [e-mail] (Jos Horsmeier)
Newsgroups: comp.lang.c
Subject: Re: Moving from Pascal to C, Help please!!!!!!
Message-ID: <[e-mail]>
Date: 6 Jan 93 16:56:14 GMT
References: <[e-mail]> <[e-mail]> <[e-mail]>
Organization: AND Software BV Rotterdam
Lines: 39

In article <[e-mail]> [e-mail] (Dave Eisen) writes:
|In article <[e-mail]> [e-mail] (Jos Horsmeier) writes:

|>Then you've just been plain lucky and so were the authors of NRC.

|My understanding was that Numerical recipes in C used the trick:
|
|#define   Array      (array - 1)
|
|and then indexed the array as Array[1] to Array[n-1].
|
|Yes, array - 1 is illegal. Even when used as Array[1] which
|is the same as *(array - 1 + 1). But even if a system uses
|segments in such a way that array - 1 is not a valid address,
|it is very likely that array - 1 + 1 gives a pointer that
|is represented exactly the same way as array is. Even if pointer
|arithmetic is done using only the offset.

That's what I thought so too, 'till someone pointed out to me that
the pointer (array-1) points _outside_ the object itself (and it
doesn't point one byte beyond the object, which is an exception
to the rule.) All pointer arithmetic, using pointers _not_ pointing
to a valid object cause undefined behavior. Maybe my example where
segmented memory was used wasn't a good example. But consider an
architecture where special pointer registers are used and whenever
such a register is loaded with a new value, a validity check is
done. Loading a register with a value (array-1) can cause a core
dump or whatever ...

|Don't take any of this as an endorsement of this "technique". It
|isn't. I just said that it was not enough to keep me from buying
|the book. It is not something I would ever do in my own code; I
|try to stay out of the areas of undefined behavior.

A very sensible strive! ;-)

kind regards,

Jos aka [e-mail]