Path: utzoo!censor!geac!jtsv16!uunet!seismo!ukma!tut.cis.ohio-state.edu!bloom-beacon!EXPO.LCS.MIT.EDU!jim
From: jim@EXPO.LCS.MIT.EDU
Newsgroups: comp.windows.x
Subject: Re: Xt and SparcStation 1
Message-ID: <8910192213.AA02401@kanga.lcs.mit.edu>
Date: 19 Oct 89 22:13:17 GMT
References: <8910191235.AA08289@orzo.ndl.com>
Organization: X Consortium, MIT Laboratory for Computer Science
Lines: 44


    Has anyone been successful in getting the Xt toolkit up and running on
    a SparcStation?  I've run into a problem where the optimizer seems to
    be generating bogus code.  By compiling it debuggable, I can get around
    the initial problem which is in ParseRepeat,

The problem rears its ugly head in all of the procedures that return a
parameter that was never referenced.  Instead of returning the string,
the compiler seems to be returning whatever what the last value computed.

There are two solutions that we've used here:

	(1)  Don't compile TMparse.c with the optimizer (this is why there
	     was some weirdness in the Xt Imakefile).

	(2)  Stick a static String variable into TMparse.c and replace the
	     "return str" in various routines to call a magic macro:

#ifdef sparc
/*
 * The silly optimizer in SunOS 4.0.3 and below generates bogus code that
 * causes the value of the most recently used variable to be returned instead
 * of the value passed in.
 */
static String silly_optimizer_kludge;
#define BROKEN_OPTIMIZER_HACK(val) silly_optimizer_kludge = (val)
#else
#define BROKEN_OPTIMIZER_HACK(val) val
#endif


So, for example, ParseImmed ends up looking like the following:

static String ParseImmed(str, closure, event,error)
    register String str;
    register Opaque closure;
    register EventPtr event;
    Boolean* error;
{
    event->event.eventCode = (unsigned long)closure;
    event->event.eventCodeMask = (unsigned long)~0L;

    return BROKEN_OPTIMIZER_HACK(str);
}