TODO

# ./antiweb -hub /home/doug/antiweb/bin/junk-hub/
FATAL: run-logger: startup: aw_log directory isn't owned by logger-uid (23001)
... but hub is still running. should probably stop hub as well



Sec:

!! limit on # conns from same ip (per worker? per worker AND in the hub? relevant at all for ipv6?)
do fstats after opening the log files and make sure they are the same inodes as when we did the lstats
try to crash AW by truncating a file while it is being sent (with and without mmap())



AWP Tasks:

browsing to an .awp file should check that the cached dir exists in case it was deleted
all .awp files are re-compiled on start-up. should compare mtimes of aw_load.lisp vs the .awp file
      EXCEPT that ajax callbacks must be loaded always and only way to do that is compile the awp atm


Tasks:

check to see if CCL can use *, ?, and : in pathnames
normalise CCL prompt
-graceful-kill and -upgrade
supervising non-existing process should show error instead of doing nothing
trying to reload conf for unattached worker should show error instead of doing nothing
trying to load same worker twice should display error instead of doing nothing
(reset-all-timers) and (trigger-all-timers)
fast-files should cache gzipped content for applicable files if enabled
fast-files should handle etags if enabled
easy way to check if file/directory exists in :rewrite rules, like -f in apache rewrite

?? throttling
?? -kill should log
?? -kill should say "OK" - would need to be special cased cause process can't send back confirmation (it has been killed)
?? check owner/group of cache directory - (will fatal() on write anyways, why die if you have a bad cache you're NOT USING?)
?? open /dev/null descriptor for use even after chroot to stfu lisp compiler when running in -nodaemon mode
?? verbose stats for workers that lists all connected IPs, number of requests, keepalive duration, etc



CLISP issues:
  * can't read in all UTF-8 strings *** invalid byte #xCE in CHARSET:ASCII conversion
  * can't create or open files with * or ? characters
      http://osdir.com/ml/lisp.clisp.general/2003-04/msg00101.html
  * normalise clisp top-level setq behaviour over -attach with cmucl's behaviour
  * ?? make -noreadline default -- readline considered harmful: can fuck up s-expression pasting


API:

?? #"#"# should be "#"
?? USE_EPOLL -> AW_LINUX, USE_KQUEUE -> AW_BSD
?? merge state change with aw_event_update


Opt:

!! doubly linked conn lists to remove the 1 remaining O(N) operation

O_NOATIME on linux
special case: when there are exactly 2 events to run and one of them is a message to
   the hub (like pipelined log msgs), do the OTHER ONE first to increase responsiveness and log pipeline
stat "cache" (not really a cache just optimises duplicate stat() calls away--would help in 2 places so far)
write-to-conn-from-lisp-string could avoid an ioblock prealloc if there is enough space in current ioblock
eliminate worker-conn-table and inet-conn-table hash lookups on conn destruction
pre-malloc conn and ioblock structs as contiguous (but aligned) memory.

?? fd and mapped memory cache
?? only flush log messages to hub every N requests to increase pipeline
?? cgis should manually close all descriptors so we don't need to make every socket close on exec
?? inherit close on exec and non-blocking to avoid syscalls (BSD only?)
?? try readv()ing into more than just 2 ioblocks for HTTP conns - should almost never matter. check this