Farid Zaripov wrote:
-----Original Message-----
From: Martin Sebor [mailto:[EMAIL PROTECTED]
Sent: Friday, August 10, 2007 9:30 PM
To: stdcxx-dev@incubator.apache.org
Subject: Re: svn commit: r564698 -
/incubator/stdcxx/trunk/include/rw/_mutex.h
* _mutex.h (__rw_atomic_preincrement) [_WIN32]: Cast
__x to int& instead of long&
to use InterlockedIncrement() optimized function
instead of generic one using
critical sections.
Really? The long overload should do the "right thing" when
long and int are the same size. If it doesn't it's our bug.
It does, but it's defined later than invoked from unsigned in
overload,
and compiler invokes the template<long> __rw_atomic_preincrement(...)
which uses critical section.
Ouch. Good catch! Does this solve the timeout problem in the atomic
tests on Windows?
Btw., this whole header is a mess. It really does need to be cleaned
up. Maybe when we get around to it, we should split up the platform
specific atomic declarations code into their own headers in addition
to simplifying the API and removing the ugly preprocessor logic (as
I mentioned in my response to Travis:
http://www.mail-archive.com/stdcxx-dev@incubator.apache.org/msg04251.html)
Martin
