From mboxrd@z Thu Jan  1 00:00:00 1970
From: Andrew Cagney <ac131313@cygnus.com>
To: Fergus Henderson <fjh@cs.mu.oz.au>
Cc: gcc-patches@gcc.gnu.org, gdb-patches@sources.redhat.com, binutils@sources.redhat.com
Subject: Re: [rfa] Change lbasename() to return ``const char *''
Date: Wed, 08 Aug 2001 06:35:00 -0000
Message-id: <3B71401D.2010409@cygnus.com>
References: <3B709080.7040508@cygnus.com> <20010808220910.A8080@murlibobo.cs.mu.OZ.AU>
X-SW-Source: 2001-08/msg00080.html

> On 07-Aug-2001, Andrew Cagney <ac131313@cygnus.com> wrote:
> 
>> The attached patch changes the liberty function lbasename() so that it 
>> returns ``const char *''.
> 
> 
> Why?  That will lead to the need for casts in the callers.  Why not just
> put the cast in lbasename(), like is done for strchr() in the C standard?


(This was discussed once before as an RFC :-).

Why not?  At present lbasename() throws away information.  The thing to 
remember is that lbasename() is ment to be kind of like basename() and 
other (botched?) ISO/POSIX library funtions, only useful :-)

If someone really wants to blat the returned buffer let that be made 
_explicit_ at the callers end.  Better still, use:

	xstrdup (lbasename (...))

which gives well defined semantics.


> lbasename() is const-polymorphic, i.e. the return value should be
> a const pointer iff the argument is a const pointer.
> C doesn't have any good way to express the type of const-polymorphic routines.
> Unless you duplicate the code in two functions, one for const pointers and one
> for non-const pointers, some casting will be required.


I've been through all the callers and I didn't find a single places 
where a cast was needed.

enjoy,
	Andrew