Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!cornell!batcomputer!gould!steinmetz!uunet!modcomp!joe
From: joe@modcomp.UUCP
Newsgroups: comp.lang.c
Subject: Re: static list initialization
Message-ID: <8000008@modcomp>
Date: 16 Oct 88 15:13:00 GMT
References: <196@donk.UUCP>
Lines: 26
Nf-ID: #R:donk.UUCP:196:modcomp:8000008:000:1204
Nf-From: modcomp.UUCP!joe    Oct 16 11:13:00 1988


> Here's a fragment of code I find myself staring at glumly from time to time.
> It sets up the nucleus of a 2-way list, into which extra malloc'd nodes will
> be inserted.
>
>     struct whatever {
>         struct whatever *next;            /* next node */
>         struct whatever *prev;            /* previous node */
>         int    eresting_stuff;            /* blah, blah, ... */
>     };
>     extern struct whatever tail;
>     static struct whatever head = { &tail, (struct whatever *)0 };
>     static struct whatever tail = { (struct whatever *)0, &head };
>
> Without the 'extern', the compiler can't initialize 'head'.  But with it, I
> get a 'non-standard extension' warning, although everything works just fine.

This kind of initialization can only be safely done at run time.  Even if
you found a compiler that would accept such an unethical sequence of code,
(two variables named "tail" -- wow!), you should not use it -- it would be
unportable, even to a different C compiler on the same machine/OS.  However,
I see how being able to forward reference structures, just as easily as we
can forward reference structure pointers today, could be appealing.

Joe Korty
uunet!modcomp!joe