empth_request_shutdown,loc_BlockMainThread,empth_init,
service_ctrl_handler) [_WIN32]: Remove the Windows Build UI and
replace with Ctrl-c signal trap. The shutdown event that was used for
STOPPING the service is now used for both foreground and background
(service) modes of operation. Also move the shutdown event from service
code to empth_t library.
empire thread debugging for Windows build.
(empth_select) [_WIN32]: Fix empth_select() so can be aborted.
The bug was that a command would not be aborted during an update.
The problem was select() was not interrupt by the setting of the
signalling wakeup event. Fix by replacing the select() with
WSAEventSelect().
(empth_exit) [_WIN32]:
Fix the shutdown sequence for Windows build to be running in
empth_t context. Add a loc_RunThisThread() in empth_exit()
before starting the shutdown() sequence.
_vsnprintf from the WIN32 library for changes to logerror().
(main) [_WIN32]: Add typedef for pid_t to match WIN32 definition
for getpid(). This is required for the create_pidfile() changes.
initial patch.
(create_pidfile, pidfname): New.
(start_server): Create pid file.
(finish_server): Remove pid file.
(main) [__linux__ && _EMPTH_POSIX]: Don't hack pid into argv[0].
Recent versions of Linux pthreads no longer show multiple PIDs, and
deities can get the PID fro the pid file anyway.
an odd case where signals were not catched when debugging.
(main, empth_start): Using sa_handler with SA_SIGINFO is wrong.
Simply don't set SA_SIGINFO; the additional signal information is not
used anyway.
(service_main,main,shutdown,empth_exit,service_stopped) [_WIN32]:
Add shutdown Event to allow the service to shutdown.
When SERVICE_STOPPED is passed to SetServiceStatus() all
threads are kills this prevents shutdown() from executing.
The shutdown Event is used to block the service_main thread until
the stopping of the service is started. After the shutdown() is
complete then the SetServiceStatus() is called.
* Acquire resources before daemonizing, fail in foreground
* Initialize threads and signals after daemonizing
* Make most file names relative to data directory
* emp_server and files make it their working directory
* emp_config() no longer screws up some file names
* Missing or incorrect econfig is now fatal
* Don't log to default log file when econfig changes it
from init_server() to start_server(), and service startup next to its
Unix equivalent disassoc(). This way, threading is always initialized
after daemonizing, which is simpler and more portable.
when it runs as daemon, because it relies on player_socket remaining
open across disassoc(). Change disassoc() to only close file
descriptors 0..2 instead of 0..9. By the way, it should really
redirect 0..2 to /dev/null instead.
(banfil, authfil, timestampfil): Make file names relative to DATADIR.
(main): emp_server and files change to the data directory. fairland
has to write newcap_script to the initial current directory, so it
keeps it and makes file names relative to DATADIR absolute.
(fixup_files): Nothing to do, remove. This fixes the following bug:
fixup_files() formed file names by appending ep->name instead of the
base name of ep->file to datadir. Thus, fixup_files() changed file
names even when datadir had the compiled-in value. When emp_config()
couldn't open econfig, it failed to call fixup_files(), and different
file names got used. Note that with this revision the file names
revert to the names used before fixup_files() was introduced. Closes
#1146080.
(print_config): Printing file names as comments is not really useful,
remove.
(otherfiles, ofiles): Unused, remove.
(emp_config, main): Previously, pconfig had to call emp_config() even
when it didn't want to read econfig, just to run fixup_files(). This
is no longer necessary. Simplify.
(emp_config): Use it when argument is NULL. This simplifies callers
except pconfig's main(), which now has to suppress reading the default
file. Also fix callers to check the return value. Previously, a typo
could easily start a grossly misconfigured program, with potentially
disastrous results.
player_accept() to player_init(). Acquiring resources during
initialization allows to exit unsuccessfully on errors. Previously,
the daemon detached itself (exiting successfully), then failed in the
background.
(player_socket): New.
