libstdc++/3759: nonconforming use of unqualified std:: names

[David Abrahams]

The following reply was made to PR libstdc++/3759; it has been noted by GNATS.

From: "David Abrahams" <david.abrahams@rcn.com>
To: "Gabriel Dos Reis" <gdr@codesourcery.com>
Cc: <rittle@labs.mot.com>,
	"Gabriel Dos Reis" <gdr@codesourcery.com>,
	<gcc-bugs@gcc.gnu.org>,
	<gcc-gnats@gcc.gnu.org>
Subject: Re: libstdc++/3759: nonconforming use of unqualified std:: names
Date: Thu, 26 Jul 2001 08:36:13 -0400

 ----- Original Message -----
 From: "Gabriel Dos Reis" <gdr@codesourcery.com>
 
 > "David Abrahams" <david.abrahams@rcn.com> writes:
 >
 > The bottom line is that, we have an issue as to know which functions a
 > library can use unqualified.  As to now, there has been no definitive
 > resolution. Oh yes, one can modify insert() to please you, but that
 > doesn't answer the question of which function names the libray can use
 > in an unqualified manner.  The answer to that question is necessary
 > before going changing function name uses willy-nilly.
 
 I think we've reached an end of this debate. I understand that we disagree,
 but I the use of the term "willy-nilly" and the question of pleasing me
 personally is an unfair characterization. You understand the issue
 differently from me. I can respect that. I even think you misunderstand the
 issue, but I don't think you're being irresponsible, as the term
 "willy-nilly" would imply.
 
 Just to reiterate my position: regardless of the question of which names may
 be used unqualified, the library must meet its specification. The places
 where the question of the use of unqualified names are relevant are those
 where the library specification says that a particular function is used in
 the implementation of another function. Examples are in valarray and
 swap_ranges. I filed a bug report against libstdc++ concerned with the
 semantics of insert() and the large majority of other functions which are
 not documented as calling other standard algorithms precisely because the
 correctness of their implementation with respect to the standard is easy to
 determine. I did not touch on the issue of valarray and swap_ranges because
 the right thing to do, as Gaby implies, is not yet clear.
 
 And now, with respect, I withdraw from this discussion.
 
 Regards,
 Dave