Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!elan!tom
From: tom@elan.elan.com (Tom Smith)
Newsgroups: comp.lang.c++
Subject: Re: default argument bug in 1.2!
Message-ID: <590@elan.elan.com>
Date: 24 Jul 89 19:42:28 GMT
References: <6409@columbia.edu>
Organization: Elan Computer Group, Inc., Mountain View, CA
Lines: 50

From article <6409@columbia.edu>, by kearns2@read.columbia.edu (Steve Kearns):
> 
> Is there a workaround for the following bug?
> 
> When I declare a member function to have a default constant argument,
> such as :
> 
> class bar {
> 		foo(int joe, int bill = -1);
> };
> 
> 
> Then the definition of foo usually contains a test like the following:
> 
>    if (bill == -1) 
> 		{ ...do this...}
>    else 
>         { ...do that...}
> 
> The bug is that 1.2 seems to optimize out the if (bill == ..) test
> and include, in this case, only the first branch of the "if".  
> It seems to treat the "int bill = -1" declaration as if I was initializing
> a local variable.  
> 
> Does anyone know any workarounds?
> -steve

I observed this in Oasys' distribution of the Glockenspiel port of
AT&T's 1.2.1 (version "d", I think).  It was not only default values
of -1 in particular, but any negative number.  In addition, the amount
of code removed was different for different negative values in the
same function!

This works, for all values:
    static const int NO_BILL = -1;
    ...
    void foo(int joe, int bill = NO_BILL);
    ...
    // in bar::foo
    if (bill == -1)
	{ ...do this... }
Note that you can still compare against -1 explicitly; it is only the function
template that needs to change.

Question: Why would a code generator treat negative default arguments
differently from positive ones?  Takes your breath away, doesn't it...

Hope this helps.
    Thomas Smith
    tom@elan.com, {ames, hplabs, uunet}!elan