Re: [boost] boost::mpl::for_each and value_initialized

Subject: Re: [boost] boost::mpl::for_each and value_initialized
From: Niels Dekker - address until 2010-10-10 (niels_address_until_2010-10-10_at_[hidden])
Date: 2010-01-13 12:11:20

Peter Foelsche wrote:
> in boost_1_41_0 there is still an unconditional call to memset in
> value_initialized()

Fernando Cacciola and I originally added the memset call within 
value_initialized() for MSVC only, because MSVC's implementation of 
value-initialization was known to incorrectly leave some bytes 
uninitialized, in some very relevant use cases. 
https://svn.boost.org/trac/boost/changeset/39157/trunk/boost/utility/value_init.hpp 
Thereby we also added extra unit tests, which were passed by MSVC by then, 
because of the memset. But unfortunately they failed on other compilers 
(GCC, Borland/Codegear and Sun), because of similar value-initialization 
issues. That's when we thought it was wise to just have the memset call 
there for all compilers: 
https://svn.boost.org/trac/boost/changeset/41942/trunk/boost/utility/value_init.hpp

Do I understand correctly that you want the memset call to be skipped for 
those compiler versions that have implemented value-initialization entirely 
correctly? Do you know exactly which compiler versions do implement 
value-initialization correctly?

Personally I do appreciate that the behavior of value_initialized is now 
exactly the same for all compilers. But I do understand you're concerned 
about the performance penalty of the memset call, for those compilers that 
have implemented value-initialization correctly. If so, can you please 
create a ticket, preferably including a patch?

Kind regards,

  Niels

--
Niels Dekker
http://www.xs4all.nl/~nd/dekkerware
Scientific programmer at LKEB, Leiden University Medical Center