Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!tut.cis.ohio-state.edu!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!uflorida!haven!adm!cmcl2!stealth.acf.nyu.edu!brnstnd
From: brnstnd@stealth.acf.nyu.edu
Newsgroups: comp.lang.misc
Subject: Unions improve aliasing analysis; EQUIVALENCEs hurt it
Message-ID: <1990:Apr1119:39:5690@stealth.acf.nyu.edu>
Date: 11 Apr 90 19:39:56 GMT
References: <8960014@hpfcso.HP.COM> <14334@lambda.UUCP>
Reply-To: brnstnd@stealth.acf.nyu.edu (Dan Bernstein)
Distribution: usa
Organization: IR
Lines: 20
X-Original-Subject: Re: Cheap implementations of languages
X-Original-Original-Subject: Re: Pointers and poor implementations

In article <14334@lambda.UUCP> jlg@lambda.UUCP (Jim Giles) writes:
  [ various stuff, including another implication that C's unions are ]
  [ as painful to analyze/implement as Fortrans' EQUIVALENCE statement ]

Sorry, Jim, you're wrong on this one too. It's only valid to use a union
element if that element was assigned to most recently. (This brings up a
few pet peeves of mine, but never mind.) I believe a compliant compiler
can implement union as struct---though I wouldn't want to use such a
compiler if I cared about memory or if I had to look at the internal
representation of data. In contrast, EQUIVALENCE is a real pain to
analyze.

To put it differently, you can regard ``union'' as the assertion ``I'll
never use element X unless my last assignment to elements of this union
was to X. Same for Y. Same for Z.'' This lets compilers save on memory
and improves---not hurts---aliasing analysis. EQUIVALENCE makes no such
guarantees; hence C is easier to optimize than Fortran in this respect.
Shall we start flogging this horse too?

---Dan