From: j DOT aldrich6 AT genie DOT com
Message-Id: <199604160421.AA016598460@relay1.geis.com>
Date: Tue, 16 Apr 96 04:10:00 UTC 0000
To: djgpp AT delorie DOT com
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Subject: Re: SIZEOF disparity...?

Reply to message 3101999    from JES AT PRESTO DOT ME on 04/11/96  5:19PM


>Like everyone else, I wanted to use a structure as a file header, but
>I knew gcc was going to try to align the members.  Trying to be smart
>(and trying to avoid explicitly packing the elements), I placed the
>long element first, assuming it wouldn't need extra padding.  Gcc
>outsmarted me though, and padded the size of the structure to a
>multiple of four; I suppose to preserve alignment if it were used in
>an array (which it wasn't).  So,
>
>   read(fd, buf, sizeof(struct header_t))
>
>was reading two bytes too far even though none of the structure
>elements needed padding.
>
>Sigh.  Structures as file headers: just say 'no'.

That's not true at all.  In addition to being able to pack individual
elements of structures, gcc also allows you to pack the _entire_
structure.  Look in the gcc docs for the __attribute__ directive.
To force the compiler to pack the structure to the absolute minimum
size, justdo this:

struct foo
{
    long blah;
    char c;
    short q[3];
#if defined(GNUC)
} __attribute__((packed));
#else
};
#endif

This will make the structure exactly 4+1+6=11 bytes long, where it
would otherwise be (4+0) + (1+3) + (6+2) = 16 bytes.
The #if directives ensure portability by only invoking __attribute__
if you are in fact running a GNU compiler.  AFAIK, any compiler
that doesn't allow you to explicitly pack structures should do so
by default (such as Turbo C), so the above will create a completely
portable structure definition.

John