redux: Blocking issues to 2.0.45?

["William A. Rowe, Jr." <wr...@rowe-clan.net>]

Ok, these are whittling away.  APR is hashing out the very last obstacles to
0.9.2 and those should be done in the next day or few.

I plan to lay down the -rc1 tag of httpd-2.0.45 just as soon at 0.9.2-rc1 is ready.
I have only three patches myself, first to change .dbgmark to .dbr (because
we conflicted in the 8.3 namespace on win32.)  Second is a set of fixes to
mod_isapi that I will email to the community of ISAPI troublemakers tomorrow 
as a new binary to .45, after they are committed to 2.1.0 (so everyone can have 
a look-see, and we assure no regressions if I backport to .45.)  Finally, and
most important is a little hack to the winnt_mpm child.c logic to deal with
a broken listen socket (v.s. an incompleted accept socket.)  Because the
AcceptEx() takes both sockets, we were naively trusting that any socket
error was an accept socket error.  Not so, and we need to unwind and pull
off a server restart when the listeners get borked.

Many, many users are complaining of that last bug, because of (alternately)
DHCP and socket filter fubars (e.g. firewall and antivirus software.)  The real
problem is that our newest code just goes loopy, and slurps up 80% CPU
on the now-worthless child process.

So that's my personal short list.  I see many good patches proposed for
backport in STATUS... and hope those get in early this week for inclusion 
in .45.  As always, the goal is "the best version of the software available."

On that tangent, it was really reassuring to see Jeff's note that we broke
head ;-)  I'm happy to see experiments move forward, yet the pitfalls are
discovered on the non-release branch!

Bill