NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/comp/lang/c/16885

Internet Message Format | 1992-11-20 | 3.0 KB

Path: sparky!uunet!news.claremont.edu!bridge2!sgiblab!swrinde!gatech!concert!sas!mozart.unx.sas.com!sasghm
From: [e-mail] (Gary Merrill)
Newsgroups: comp.lang.c
Subject: Re: Is this ANSI?
Message-ID: <[e-mail]>
Date: 20 Nov 92 16:15:37 GMT
References: <[e-mail]> <[e-mail]>
Sender: [e-mail] (Noter of Newsworthy Events)
Organization: SAS Institute Inc.
Lines: 64
Originator: [e-mail]
Nntp-Posting-Host: theseus.unx.sas.com


In article <[e-mail]>, [e-mail] (Andrew Koenig) writes:
|> In article <[e-mail]> srhqla!quest!kdq writes:
|>
|> >     Here is (a simplified version of) the struct that the compiler
|> > calls an error:
|>
|> > struct FM
|> > {
|> >     short    data_len;
|> >     char    data[];        /*  This line is the kicker  */
|> > };
|>
|> >     As I recall, ANSI says this is legal, with sizeof( struct FM) is 2
|> > bytes, and offsetof( struct FM, data) is 2.
|>
|> I don't think you recall correctly.  If nothing else, there is no
|> requirement that sizeof(short) <= 2.  Moreover, since every object,
|> including members of structures, must have nonzero size, the line
|> in question would have to allocate at least one element.  I would
|> therefore expect that if I hunted through my copy of the Standard,
|> I would find that this fragment is indeed illegal.

It sure is illegal.  Some compilers (our mainframe compiler, for example)
allow this as an extension.  It's not a good idea to code this even if
it is permitted.

|> I'm not going to bother to hunt, though, because I think your question
|> conceals a more important question that I can answer much more easily.
|> The point is to ask why you're trying to do this in the first place.
|>
|> That is, why bother with what looks like an array with no elements?
|> What use is it?  The only answer I can think of is that you're intending
|> to allocate `extra' memory and then subscript off the end of the array --
|> and the standard definitely does not allow that technique, period.

I think "does not allow that technique" is a little strong.  What you
*can* do is

    struct FM
    {
       short data_len;
       char  data[1];
        };

    struct FM *fmptr;

    fmptr = malloc( sizeof(struct FM) + ??? );

where the ??? is the number of bytes of your data minus 1.  Some may
regard this as "impure", but it is not forbidden by the standard and
is eminently portable.  Why would you want to do this?  Well, it
can really reduce the number of malloc() calls and hence the
function call overhead in general.  I have achieved quite measurable
performance improvements on some systems with this technique.

It's not so hot if you ever want to *change* the size of the data
array, free it, etc.  If you are going to do ugly stuff, you need
to think about it carefully and weigh the benefits against possible
future costs.
--
Gary H. Merrill  [Principal Systems Developer, C Compiler Development]
SAS Institute Inc. / SAS Campus Dr. / Cary, NC  27513 / [phone]
[e-mail] ... !mcnc!sas!sasghm