Mat Marcus wrote:
On Thu, Apr 24, 2008 at 12:53 AM, James Talbut <James.Talbut@omg3d.com> wrote:
Hi,
There have been various discussions on here about the problems caused by _SECURE_SCL, but I haven't seen any concrete plans for changes to boost build to handle it.
At a minimum the build system should be changed to enable those users wanting _SECURE_SCL=0 to do so easily - ideally by providing an argument to bjam.
Is someone working on this?
I'd love to see someone pick up the ball on this one. For what it's worth, I've patched my local msvc.jam to add the feature:
feature.feature stl-security-theater : off on : propagated link-incompatible ;
flags msvc.compile C++FLAGS <stl-security-theater>off : /D_SECURE_SCL=0 ; flags msvc.compile C++FLAGS <stl-security-theater>on : /D_SCL_SECURE=1 ;
As a result of our analysis of the benefits vs. performance of the "feature", I've set _SECURE_SCL=0 off by default. I asked on the build list how to augment library name mangling to make it safer, but haven't received a reply yet, so I may have to study a bit to make it better. But I will probably just ship it as above with Adobe Source Libraries 1.037.
Hi Mat, Is it the secure SCL that is causing your problem, or more specifically the iterator debugging feature? I agree that MSVC's iterator debugging can cause huge performance problems, and I've had to disable it in the accumulator's tests to get them to complete in any reasonable amount of time. Here's the bjam magic I use for that: # bring in rules for testing import testing ; import feature ; feature.feature iterator_debugging : on off : composite propagated optional ; feature.compose <iterator_debugging>off : <define>_HAS_ITERATOR_DEBUGGING=0 ; project : requirements <library>/boost/test//boost_unit_test_framework <link>static <include>../../.. <toolset>msvc:<define>_SCL_SECURE_NO_DEPRECATE # MSVC's iterator debugging causes some tests to run forever. <toolset>msvc:<iterator_debugging>off <toolset>intel-win:<iterator_debugging>off ; Now, since <iterator-debugging> is a feature, boost.build ensures that the test framework (against which I link) is also built with this feature -- very important because this setting breaks binary compatibility. The resulting binaries are placed in paths whose named encode the value of this feature, so I can't inadvertently link to an incompatible lib. Does that help any? Maybe someone could add this feature to BBv2 proper. -- Eric Niebler Boost Consulting www.boost-consulting.com