On Wed, Jun 30, 2004 at 04:17:47PM +1000, Rob Weir wrote: > http://www.mvps.org/vcfaq/lang/1.htm. I do believe it has been fixed in > recentish versions of VC++, though.
I tested it: it is fixed, along with many of the STL problems. Microsoft actually purchased their STL implementation for VC5-VC7 from Dinkumware (http://www.dinkumware.com/), and for most of the >VC6 time frame it was always possible to purchase a more recent incarnation of same from Dinkumware; or use Roguewave, or STLport. Microsoft I think either replaced STL in VC7.1 with the most recent dinkumware or else rewrote it from scratch; can't remember which. The competition between STL providers is about locking performance in high-end multithreaded apps; the Microsoft-provided STL still sucks for that. (As does the Microsoft heap: for SQL Server Microsoft uses an in-house heap which has very good characteristics, you can use a thing called Low-Fragementation Heap in Server 2003; and there is the commercial SmartHeap). The guy who wrote the super-cool heap SQL Server uses wants to give it away as an MSDN download; nobody really disagreed with it, it was just a matter of supporting it and laziness. He blew my mind describing it to me: his heap is not "fair" -- you are never guaranteed to get the lock at all. It is only "statistically fair" -- which means on average you'll get the lock faster. There's a thing called Lock Train Wrecks. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
