Re: [boost] Lightweight Mutex and others

From: Peter Dimov (pdimov_at_[hidden])
Date: 2007-05-28 18:17:50

Sohail Somani wrote:
> On Mon, 2007-05-28 at 23:37 +0300, Peter Dimov wrote:
>> Sohail Somani wrote:
>>> On Mon, 2007-05-28 at 20:27 +0300, Peter Dimov wrote:
>>>> Robert Ramey wrote:
>>>>> I'm looking at some code in boost/detail - in particular
>>>>> Lightweight mutex.
>>>>> Is there any chance that this as well as other useful modules
>>>>> can get promoted to boost "first class" status.  That is
>>>>> would include thier own documentation(especially) and tests.
>>>>
>>>> Speaking of lightweight_mutex, it was only intended as a stopgap
>>>> measure until the 'real' boost::mutex becomes header-only and
>>>> 'lightweight'. This should happen any day now. :-)
>>>
>>> I'm taking this to imply that the conversion is already occurring.
>>> Could the header-only-ness please be made an option?
>>
>> I don't see how.
>>
>> What do you expect to gain from the optional header-only-ness?
>
> Hmm. Well not #include'ing <windows.h> is one.

Neither lightweight_mutex.hpp nor a future header-only boost/mutex.hpp 
include (or will include) <windows.h>.

> I don't mean to start this whole header-only thing again, *but* would be
> nice if it weren't forced.

I'm not sure that I understand your usage of 'forced'. The problem that 
motivated the creation of lightweight_mutex is that boost::mutex forces a 
dependency on the Boost.Threads library. This is unacceptable for many 
libraries, both header-only and otherwise. You can't solve that with making 
a boost::mutex that is optionally header-only, because it will probably lead 
to ODR violations when library A uses its header-only variant and program B 
uses both the out of line variant of boost::mutex and library A.