NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/std/c++/374

Internet Message Format | 1996-08-06 | 3.8 KB

Path: fido.asd.sgi.com!austern
From: [e-mail] (Jason Merrill)
Newsgroups: comp.std.c++
Subject: Re: Observations on templates
Date: 08 Feb 1996 14:30:35 PST
Organization: Cygnus Support, Mountain View, CA
Approved: [e-mail]
Message-ID: <[e-mail]>
References: <[e-mail]> <[e-mail]>
NNTP-Posting-Host: isolde.mti.sgi.com
X-Original-Date: 08 Feb 1996 13:30:55 -0800
In-Reply-To: "Eugene Radchenko"'s message of 8 Feb 1996 17:00:57 GMT
X-Newsreader: Gnus v5.0
X-Auth: PGPMoose V1.1 PGP comp.std.c++
    iQBVAwUBMRp5sky4NqrwXLNJAQEdUwIAvQplNB5+N+FIiihbwHnql1OGLHxKM9hY
    x/rrCqKTKtWFZlQOlyNRk+2Tv3/EEyMttjdE3cHMj21c2mcOePJYrA==
    =clXi
Originator: [e-mail]

>>>>> Eugene Radchenko <[e-mail]> writes:

> [e-mail] (Jason Merrill) writes
>>>>>>> Eugene Radchenko <[e-mail]> writes:

>>> 1) Clause [temp.res] verse 1 requires that all types dependent on the
>>> template parameter must be qualified by 'typename' keyword - supposedly to
>>> allow checking the template syntax. But it does not help anything as _each_
>>> instance will have to be checked anyway:

>> Requiring 'typename' allows syntactic analysis and some name binding
>> to be done at the point of definition of the template.  Older
>> implementations do all name binding at the point of instantiation,
>> which was thought to be confusing.

>> Each instance will indeed have to be checked, but only for semantic
>> correctness, not syntactic.

> And does such a simplification (which does not simplify anything as
> complete recheck is still needed) warrant the introduction of a new
> keyword (why could not at least the 'class' keyword be used like in
> template prefix?).

Because 'class' already has semantics in that position, and a new keyword
is really not a significant problem.  If it conflicts with your code, you
can easily work around it with a #define.

> Interestingly, WRT member functions (and especially operator->) the
> decision was roughly inverse (i.e. to perform checking only on the
> per-instance basis and only if given instance is required).

Nope; function templates, like class templates, are parsed immediately to
do name binding according to the new lookup rules.

>>> 2) Clause [temp.dep] verse 5 states that base class scope names hide the
>>> template parameter:

>> This is for orthogonality.  Think of the template header as creating a
>> special scope around the class; name lookups in class scope look in the
>> current class, then in base classes, then in the surrounding scope.

> That much I do understand. I only think that such an order is
> counter-intuitive.

Well, other people don't, apparently.

>>> 3) Clause [temp.arg.explicit] verses 2-3 requires that for explicit arg in
>>> member template to be present it must be qualified by 'template' keyword,
>>> e.g.
>>> X *p; p->template alloc<200>();
>>> But to use operator -> we need X class definition anyway - so we know that
>>> alloc() is a template.

>> It is needed in some situations where we do not have the X class definition
>> -- in template definitions, like with typename.

> And what are we supposed to do with function alloc() if we do not have the
> class definition and hence do not know anything about this function?

We only need the 'template' keyword there so we know what '<' means.  If
'alloc' is a template, it means "begin template arg list".  Otherwise, it
means "less than", and it doesn't matter if 'alloc' is a function or a
field or an enum or whatever; that can wait until semantic analysis at
instantiation time.

Jason
---
[ comp.std.c++ is moderated.  Submission address: [e-mail].
  Contact address: [e-mail].  The moderation policy is
  in http://reality.sgi.com/employees/austern_mti/std-c++/policy.html. ]