Re: [boost] Boost inspection notification (2006-09-13/RC_1_34_0) *X*

From: David Abrahams (dave_at_[hidden])
Date: 2006-09-14 15:20:57

"Peter Dimov" <pdimov_at_[hidden]> writes:

> Gennaro Prota wrote:
>> On Thu, 14 Sep 2006 07:00:26 -0400, David Abrahams
>> <dave_at_[hidden]> wrote:
>>
>>> Gennaro Prota <gennaro_prota_at_[hidden]> writes:
>>>
>>>> On Wed, 13 Sep 2006 22:00:09 +0200, Thorsten Ottosen
>>>> <thorsten.ottosen_at_[hidden]> wrote:
>>>>
>>>>> What is the problem with an unnamed namespace in a header anyway?
>>>>> Is it illegal according to the standard?
>>>>
>>>> Thorsten, this is not addressed to you, but I find shameful that so
>>>> many boosters don't know this C++ 101.
>>>
>>> Now, now, Genny.  I don't recall anyone ever talking about the
>>> problem until I raised it a year or two ago.
>>
>> And here's the ego I was talking about. You may feel like the first
>> man who brought the light on all us, but that light is common advice
>> for any comp.lang.c++.moderated regular.
>
> Not quite... the general advice against putting unnamed namespaces in 
> headers is one thing, explaining how the constant variable (unnamed)::_1 
> technically causes ODR violations is another.
>
> In practice, Bind is getting away with this kind of a technical ODR 
> violation on every supported compiler.

Yes, and let me support that assertion by saying it's probably going
to continue to work on every C++ compiler we ever encounter.  Maybe
some fancy lint tool would be able to detect it.  It's really only a
technical violation of the standard.  But that said, I don't like the
idea of, entrapment even if the crime commited is victimless... or
something.  I hope y'all see my point.

> But my question wasn't about ODR violations or egos. I asked what is the 
> proper course of action for a library developer whose library has been 
> flagged by the report. And it seems that nobody knows. I can remove the 
> unnamed namespace (at some cost) but this is a potentially destabilizing 
> change and not suited for 1.34, against which the report is being run. 

I agree.  I think you should put the report silencing directive in the
RC_1_34_0 branch only, with a note saying that we will address it in
the next release.  How does that sound to you?

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com