[boost] [config] New defect macro: BOOST_NO_COMPLETE_VALUE_INITIALIZATION

Subject: [boost] [config] New defect macro: BOOST_NO_COMPLETE_VALUE_INITIALIZATION
From: Niels Dekker - address until 2010-10-10 (niels_address_until_2010-10-10_at_[hidden])
Date: 2010-04-09 09:25:50

FYI,

John Maddock has just added the tests to trunk/libs/config that I wrote for 
a new defect macro: BOOST_NO_COMPLETE_VALUE_INITIALIZATION
  https://svn.boost.org/trac/boost/changeset/61153

And some documentation:
  https://svn.boost.org/trac/boost/changeset/61154

A test failure indicates that the specific compiler doesn't have completely 
implemented value-initialization. In the coming few days, the test is likely 
to cause regression failures, because I still need to define the defect 
macro for those specific compilers (at boost/config/compiler).

Although I already expect certain compilers to fail (MSVC <= 2010, 
CodeGear/Borland <= 2010, GCC < 4.4, and various versions of Sun) I chose 
not to define the defect macro /beforehand/, because the test output 
provides some extra information about the particular compiler issue. 
Moreover, once the defect macro is defined, the test framework no longer 
indicates whether the compiler defect is really there.

I'll keep an eye on the regression page, 
http://www.boost.org/development/tests/trunk/developer/config_.html and fix 
them one by one, but please don't hesitate to remind me if there's a 
regression failure regarding the new defect macro that I overlooked.

The defect macro will allow fixing ticket #3869, "Unconditional call to 
memset in value_initialized()", reported by Aleksey Gurtovoy: 
https://svn.boost.org/trac/boost/ticket/3869 by adding an #ifdef to 
utility/value_init.hpp:

  #ifdef BOOST_NO_COMPLETE_VALUE_INITIALIZATION
       std::memset(&x, 0, sizeof(x));
  #endif

Kind regards,

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