Zero-copy C++ iostreams, soon? Maybe in a gcc-3.4.x. But only on NetBSD and OpenBSD, at first, 'cause only they have UVM, so far. (Linux might get it, in 2.8, in a couple of years.) Libstdc++ already has (probably in gcc-3.4) cycle-for-cycle speed parity between streambufs and glibc stdio. (All hail Paolo Carlini!) Native, default large-file support in the iostream interface is more urgent than zero-copy (which doesn't affect the ABI), and ought to appear in gcc-3.4, even if not universally implemented underneath, at first. The gcc-3.4.x series libstdc++ will start impressive and get better from there, and thousands of existing programs will magically get faster.
32-bit CRCs, still? Why do people still use 32-bit CRCs in new protocols? There's a very nice CRC-64 implementation in PostgreSQL, free for the taking.
robocoder: Trust me, you don't want to use C++ Xalan/Xerces. It's irresponsible for Apache to continue distributing them. Gnome's libxml2/libxslt work ever so much better, and are licensed compatibly. (Thanks, DV!)
Apt-get installed Dasher on my laptop tonight. mjg59 rocks.
I/ENTP: Me too. It's pretty common among programmers. The Js tend to become scientists instead, because they tend to jump to conclusions too early to be good software architects. A good programmer has to be comfortable with not knowing until the right time.