Path: utzoo!attcan!uunet!wuarchive!zaphod.mps.ohio-state.edu!mips!ardent!jra!jss
From: jss@jra.ardent.com
Newsgroups: comp.lang.c++
Subject: Re: ostream_with_assign versus specialization of cout?
Message-ID: <9806@ardent.UUCP>
Date: 21 Dec 89 20:29:38 GMT
References: <1989Dec19.175438.12843@odi.com>
Sender: news@ardent.UUCP
Reply-To: jss@jra.ardent.com ()
Organization: Ardent Computer Corp., Sunnyvale, CA
Lines: 34

In article <1989Dec19.175438.12843@odi.com> benson@odi.com () writes:
>I've got a derived class of ostream that internally sets the unitbuf
>flag. I tried
>
>   cout = one_of_my_streams;
>
>The operator= goes and reinitializes the stream, clearing the 
>crucial flag (unitbuf). If operator= is going to do this,
>it seems that cout should have been a pointer. As it is,
>one can't make cout be another type, it seems.

I tend to agree.  The predefined classes ought to have been
pointers rather than streams, however I didn't feel that
iostreams could make that radical a departure from streams.
I originally tried having pointers and #defining the predefined

	#define cout (*cout_ptr) 

but it eventually became clear to me that macros here were
a bad idea. (If nothing else they create confusing error messages
for beginners.)

My own coding style uses a global ostream* and assigns
&cout to the global when appropriate rather than assigning 
streams to cout. 

For a long time I resisted having assignment of streams at all 
in iostream.  I finally allowed it primarily for backward 
compatibility.  (Although it was a nightmare to implement)
But it was never clear to me what the semantics of the 
assignment should be.  

Jerry Schwarz
jss@ardent.com