init() it while it's already running. fixed an uninitialized
pointer bug in CServer and some cleanup-on-error code in
CMSWindowsPrimaryScreen. also added timeout to read() on
IInputStream and a heartbeat sent by clients so the server
can disconnect clients that are dead but never reset the TCP
connection. previously the server would keep these dead
clients around forever and if the user was locked on the
client screen for some reason then the server would have to
be rebooted (or the server would have to be killed via a
remote login).
and back to secondary when it stops. also now redirecting
keyboard input to root window when screen saver starts; this
allows the user to type in the lock dialog and also effectively
discards any input used to deactivate the screen saver.
installing and uninstalling the new error handler, causing
errors before the lock to be caught and errors during the lock
to not be caught. had to add Display* as argument to c'tor.
and untested. also some known problems: not detecting an
xscreensaver started after us and not detecting built-in
screen saver activation (not sure if we can without using
ugly extensions).
to the thread list: the child thread would add itself to the
list which means there could be a time interval in the parent
where the child thread exists but isn't on the list. the
parent now does the adding and removing.
motion events relative to the previous mouse position. for
example, if two mouse events arrive, the first at x+1,y and
the second at x+2,y, we used to compute deltas of 1,0 and 2,0
instead of 1,0 and 1,0. that's fixed. also worked around a
bug (probably) in windows that caused a motion event after a
SetCursorPos() to be lost or reported one pixel off from the
correct position. now using mouse_event() which doesn't
have that problem. also fixed calculation of normalized
coordinates for mouse_event() when there are multiple
displays.
also no longer warping mouse to 0,0 when setting server screen
info. that was causing the mouse to jump if the server screen
had itself as its left or top neighbor (directly or indirectly)
once a screen could be its own neighbor.
include a screen that was already connected under an alias
then that screen wouldn't be disconnected and removed from
the screen list until the screen voluntarily disconnected.
at that time removeConnection() would assert because the
screen name would not be found. now using the canonical
name in the protocol object as well as CServer. this
allows setConfig() to always detect removed screens and
disconnect them.
the address and port to listen on or connect to. changed the
default port and put it in ProtocolTypes.h. the HTTP port is
now no longer opened unless the --http argument is supplied
or the config file includes it.
restarting and daemonizing on unix. daemon sends log messages
to syslog. unix now reads config file from file named on
command line; if no command line arg then uses effective
user's config file and if that's not there it finally tries
/etc/synergy.conf. if there are no screens configured then
one is added for the primary screen. broke some startup
stuff on win32.
also now timing out if X primary screen can't grab the mouse
and keyboard. the server will just give up trying to switch
screens. the grabs will fail is some other app has a grab
and won't release it. note that kdm grabs the keyboard for
the duration that the login window is displayed, effectively
disabling synergy.
the apps still exit but they do it in a mostly controlled
manner. in particular, the server threads except the one
processing primary screen events will terminate gracefully.
this will be important should the server ever allow HTTP
clients to rewrite the configuration file.
note that X makes it effectively impossible to continue once
the X server disconnects. even if it didn't it would be
difficult for synergy to recover. users will have to add
synergy to the X display manager's startup script if they
expect the server to be restarted. alternatively, we could
add code to fork synergy at startup; the child would do
the normal work while the parent would simply wait for the
child to exit and restart it.
on clients that cause us to use too much memory. also put
methods in CHTTPRequest to get/set headers and changed the
data structure used to store them. fixed a couple of other
miscellaneous bugs in CHTTPProtocol.cpp.
CServer to the primary screen when the configuration changes so it
can make necessary adjustments (the win32 primary screen must tell
the hook dll about the new jump zones).
changed includes of some std c++ library files to go through
our own include files. these wrap the include with stuff to
keep vc++ quiet when compiling at warning level 4, which is
what it does now. it also works around missing <istream> and
<ostream> on g++2.96.
added missing std:: where necessary. g++ doesn't really support
namespaces so it lets references without the namespace slip
through.
added workaround or fix. not sure if istringstream::str(string)
should reset eofbit. it does on g++ but does not on vc++.
added clear() after str() so it works either way.
added low-level keyboard hook to win32. if available (it's only
available on NT SP3 and up) it allows us to catch and handle
alt+tab, alt+esc, ctrl+esc, and windows key hot keys. i think
that leaves only ctrl+alt+del and accessibility functions
uncaught on those systems.
fixed two blown assertions. first, if user tried to switch to
a client that had connected but hadn't yet sent the first info
message it would assert on the zero size screen. second, if
the primary screen was handling a mouse motion on behalf of a
secondary screen when that secondary screen disconnected then
an assert would blow because the primary screen would call
onMouseMoveSecondary() but m_protocol on the active screen is
NULL because disconnecting the active secondary screen caused
the mouse to jump to the primary screen.
crs