Hannah Schroeter wrote:
On Thu, Oct 30, 2008 at 03:24:06AM +0100, Uwe Stuehler wrote:
I also disabled BOOST_TEST_USE_ALT_STACK to allow me to run
multi-threaded tests, although maybe a better fix exists. The test
setup failed, because of the way our sigaltstack() works.
I don't think a better fix exists (until maybe OpenBSD's thread
implementation switches to rthreads). Boost just should not assume
sigaltstack works with multithreaded. The standards allow OpenBSD's
behavior, see for example
http://www.opengroup.org/onlinepubs/000095399/functions/sigaltstack.html
the last paragraph in DESCRIPTION.
Yes, I was aware that OpenBSD's sigaltstack() is conformant and that
Boost.Test's assumptions about how sigaltstack() works is wrong, but I
am not aware of the consequences of disabling BOOST_TEST_USE_ALT_STACK.
It likely doesn't matter much, I was just slightly worried because I
thought about using the test execution monitor in my program for UNIX
signals support (which no other Boost library seems to provide, and
which is explicitly not supported in standard C++).
Attached is a new diff against -current.
Compiling (grind grind), still without trying a fix for the
PFRAG.shared problem or multi-packaging bjam/boost.build.
I've also recently experienced that this port doesn't compile when it
(the new version) is already installed. However I can only look into
it later if at all.
That was solved by passing --ignore-site-config to bjam. It is already
fixed in my second diff.
Cheers!
