Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site alice.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!alice!td
From: td@alice.UUCP (Tom Duff)
Newsgroups: net.lang.c
Subject: Re: Re: Structure question (problem)
Message-ID: <3772@alice.UUCP>
Date: Thu, 30-May-85 12:09:26 EDT
Article-I.D.: alice.3772
Posted: Thu May 30 12:09:26 1985
Date-Received: Fri, 31-May-85 03:59:34 EDT
References: <10@ucbcad.UUCP>
Organization: Bell Labs, Murray Hill
Lines: 29

It is very difficult to formulate a reasonable rule for union
initialization.  The ANSI rule (initializers apply to the first member)
is just a kludge.  The `match the types' rule plays fast and loose
with C's type-matching rules.  Consider ucbcat!faustus's example
(here simplified):
union{
	int a;
	long b;
}c[]={
	57,
	(long)76365
};
ucbcat!faustus no doubt expects 57 to initialize a and 76365 to
initialize b.  But 57 is a legal initializer for b and (long)76365
is a legal initializer for a.
So the `match the types' rule is not a simple patch to the language,
but requires a completely different set of type-matching rules than
already exist.  This is, to say the least, unsatisfactory.

Furthermore, this scheme cannot even be made to work.  The following
example should indicate one reason:

union{
	int a;
	struct{
		int :16;
		int b;
	}c;
}d={5};