barrier/notes
crs 64232c7854 updated to new automake and refactored server stuff. the server
now speaks to the primary screen and secondary screens almost
everywhere the same way through an IClient interface;  only
special primary screen calls are accessed through a different
interface, the CPrimaryClient interface.  this simplifies the
server since it no longer needs to test whether the active screen
is the primary or a secondary in most cases.

the server no longer speaks directly to the primary screen;  all
that goes through the CPrimaryClient, which often just forwards
the call.  the primary screen no longer speaks directly to the
server either, again going through the CPrimaryClient via a
IPrimaryReceiver interface.

CServerProtocol classes have been replaced by CClientProxy
classes which are very similar.  the name makes more sense
though.
2002-07-09 21:22:31 +00:00

615 lines
20 KiB
Plaintext

server client
------ ------
[accept] <-- connect
challenge --> [encrypt]
[verify] <-- response (encrypted challenge, client name)
hangup if invalid
query info -->
<-- info (size)
---
nedit doesn't seem to be playing nice with the motif clipboard lock
it's not releasing it
may need to learn more about how the motif clipboard lock works
not sure if this a new problem with synergy or a problem with new nedit
---
enable heartbeat disconnection or remove heartbeat sending in client
should either ship with it fully enabled or fully disabled
possibly make it part of startup negotiation
---
getting a stuttering when leaving win32 server screen
---
merge platform dependent code into platform independent where possible
also try to simplify where possible
---
IServer and IPrimaryReceiver are really similar
can we merge them into one?
---
use automake
use doxygen
---
HTTP stuff
no way to save config using HTTP
should have a button or a page to force save of config
or just save config every time it's changed
should use authentication (at least basic)
provide way to kill/restart server via HTTP
provide way to query why locked to screen?
handy for debugging at least
---
negotiation:
use a generic negotiation message (name/value pair) or specific
messages (name/values)? later allows more parsing to be done by
CProtocolUtil but how can we skip unknown messages? would have
to have a method on CInputPacketStream to discard to end of
message and we'd need to know the stream is a CInputPacketStream.
how does negotiation proceed?
could have sequence of queries and replies:
server -> client -> server ... end
client -> server -> client ... end
or a sequence of messages:
server -> client ... end
client -> server ... end
probably go with latter because it's more efficient but make it
3-way:
client -> server ... end
server -> client ... end
client -> server ... end
first messages from client must be version and name. the server
can create the appropriate protocol object right away then.
example: clipboard formats:
# CNEG = negotiation command CNEG%s%s
# CNGE = end of negotiation command
# cbfa = clipboard, format, add (permitted format)
# cbia = clipboard, id, add (clipboard with id exists)
client -> server: CNEG "vers" "1.0"
client -> server: CNEG "name" "foobar"
client -> server: CNGE
server -> client: CNEG "cbfa" "text/plain/LF"
server -> client: CNEG "cbfa" "image/BMP"
server -> client: CNGE
client -> server: CNEG "cbia" "0"
client -> server: CNGE
server should just ask CProtocol (renamed from CServerProtocol) to
return a protocol. CProtocol should do negotiation, create the
appropriate protocol object, finish negotiation, and return the
new protocol object. unless connection is registered with server,
though, we can't save any screen info. perhaps CProtocol should
also return a screen info object or that should be part of the
protocol object (currently the protocol is part of the screen info).
if screen info is available to CProtocol then it should finish with
requesting the screen info, waiting for and then handling the reply.
the server can then just ask CProtocol for the protocol object then
add it to the connections.
maybe call the protocol types CClientProxy since that's how the
server uses it: as a stand-in for the client object. the interface
should closely reflect the CClient interface. do the reverse for
a server proxy for the client. maybe even have interface classes
for the client and server methods.
---
should add clipboard data type format negotiation
server sends permitted data type formats
client responds with known formats in permitted list
client must only send permitted formats
server can discard non-permitted formats
server must only send known, permitted formats to client
formats are names of the form [a-zA-Z0-9_/]+
server must be able to translate between all permitted formats
example: text formats:
text/plain/CRLF -- windows plain text
text/plain/LF -- unix plain text
text/plain/CR -- mac plain text
text/rtf -- rich text (does that require a particular newline?)
or image formats:
image/BMP
image/RAW
image/XPM
this will allow future versions to interoperate better. it also
allows compatibility with versions that are identical except for
the support clipboard formats, obviating bumping up the protocol
version number.
maybe config file can list format shared libraries to load. user
can then pick formats to support. but why would you ever want
fewer formats?
should the major burden of translation be on client or server?
probably client since it knows the formats preferred by the
platform and may be able to use platform utilities to convert.
server would then just support formats that minimize loss of
information.
desired formats to support:
text (LF)
text with character set
unicode (LF)
rich text (or some kind of formatting)
bitmap (BMP, PNG?)
sound (WAV)
note that formats should be added to the win32 clipboard in the
order of most descriptive to least because they're kept in order
and apps will generally use the first suitable match.
---
hot keys
should have keyboard shortcuts to jump to screens
---
win32:
need icon
provide taskbar icon
provide control panel?
win32 dll:
should make this as small as possible
don't use standard libraries
use custom _DllMainCRTStartup(); just call DllMain() from it.
use /MERGE linker switch to merge sections
win32 service:
need message table for proper event reporting
clean up c:\winnt\synergy.sgc
desktop switcher program failure when running synergy as service
returns access denied from CreateDesktop()
don't know why
---
bug with half-duplex keys
win32 server sent num-lock to grace
grace converts to half-duplex so only synthesizes key down
now grace starts server and is locked to screen
grace should ignore half duplex keys when seeing if locked to screen
---
bug in updating screens
saw a goofy set of screens after update
i think the config was similar to:
A B
C
D E
---
should switch to user nobody (or something similar) if running as root
for security reasons
will make it impossible to overwrite /etc/synergy.conf
if that file is only root writable (as it should be)
---
Xsetup file may have to kill previous synergy
or synergy should exit if there's already a synergy running
---
not handling international characters
---
xscreensaver won't start if pointer on secondary screen
unless using MIT or SGI screen saver extensions
otherwise it cannot know when to stop the screen saver
cos it can't grab mouse and keyboard
---
all screensavers hidden when showing win32 screensaver lock window
win32 kills the screen saver and shows the lock window
synergy thinks the screen saver is dead (cos the process is)
deactivates secondary screen screen savers
user is constrained to locked display, though
also, we don't detect screen saver restarting if user doesn't unlock screen
---
not distinguishing between caps lock and shift lock
In Caps Lock mode, the alphabetic keys send their uppercase (shifted) char-
acter when pressed alone. For Caps Lock mode, pressing and releasing the
lock key turns on the mode and pressing and releasing the lock key again
turns off the mode. In Shift Lock mode, all keys on the main keypad send
their shifted character when pressed alone. For Shift Lock mode, pressing
and releasing the Lock key turns on the mode, and pressing and releasing
either the Lock or the Shift key turns off the mode.
---
currently sending all clipboards to all clients
some clients may not need all clipboards
add filtering mechanism
setup is probably part of clipboard (format) negotiation
---
not handling large motif selections
there must be a limit because properties can only be so big
don't know protocol for transfer, though
---
need timeout on CXWindowsClipboard::CReply objects
should flush replies that are too old
assumption is that requestor is broken
---
inconsistencies
exceptions
using getWhat() in some places to return a code, elsewhere returns message
references/pointers
should use pointers for any in/out parameter
but using references in some places
---
avoid fullscreen transparent window on win32
using CBT hook to discard activation/focus messages
but it's not being called when we lose keyboard input
also, we sometimes don't get the keyboard input to start with
---
keyboard hook problems (win32)
now passing keyboard events through in most cases
this fixes problem on isabel where toggle lights don't toggle
and num-lock behaved incorrectly (generated wrong keys).
seems the DefWindowProc() needed to process the keys
unfortunately, keys sometimes leak into wrong app. seems related
to times when we don't get the keyboard input. key events are
not delivered to app on primary until cursor returns to primary.
no idea how that's possible.
current have some code to check active window and reset it in
keyboard hook. that code doesn't work.
seem to be having problems with windows key now, too. looks like
a down is sent but not the corresponding up so secondary system
thinks the key is always down.
---
try to determine X11 keyboard quirks automatically
in particular, set the half-duplex flags
maybe KBD extension will help determine this
---
key events sent to console on win me (95/98?) are very slow
1/4 to 1/2 second delay before key up is processed
key up log message is also delayed
console causing a system-wide delay?
---
adjust thread priorities on win32
maybe remove changes altogether
currently isabel starts single-stepping mouse when dragging window
sometimes it goes normal speed but it's mostly very slow
even a very high priority doesn't seem to help
haven't tried real-time priority
a very high priority on client fixes delay when typeing into console
is it the output console causing the slowness?
mouse motion compression seems to handle slow dragging on grace
pretty sure nothing else was wrong on grace
maybe same problem and solution on win32
---
results:
X primary
ctrl+alt+keypad_add, ctrl+alt_keypad_minus -> eaten by primary X server
ctrl+alt+backspace probably eaten by primary X server if so configured
ctrl+alt+delete probably eaten locally
probably underneath primary X server too
Win32 desktop -> X laptop
eaten global hot keys:
ctrl+alt+delete -> task manager (or logon or reboot)
alt+[shift+]tab -> cycle through tasks
alt+[shift+]esc -> cycle windows
ctrl+esc -> start menu
windows+R -> run dialog
windows+M -> minimize all
windows+shift+M -> undo minimize all
windows+F1 -> help
windows+E -> explorer
windows+F -> find files
windows+ctrl+F -> find computer
windows+tab -> cycle through taskbar buttons
windows+break -> system properties
not eaten:
alt+space -> system menu
shift+F10 -> context menu
app key -> context menu
misc:
accessibility shortcuts: not eaten but dialogs appear anyway
X laptop -> Win32 desktop
check:
shift+tab -> okay
alt+[shift+]tab -> okay
alt+[shift+]esc -> okay
ctrl+esc -> start menu -> okay
ctrl+esc, esc, shift+F10 -> toolbar context menu -> okay
F10 -> activate menu in explorer -> okay
shift+F10 -> context menu -> okay
alt+space -> context menu -> okay
alt+hyphen -> MDI window menu -> okay
alt+letter -> open corresponding menu -> okay
alt+F6 -> switch between windows in single program -> okay
shift+insert CD-ROM -> bypass auto-run
check accessibility shortcuts -> does not work
check accessibility features
sticky keys -> no
filter keys -> no
toggle keys -> no
mouse keys -> no
high contrast -> no
X desktop -> win32
check:
shift+tab
windows+R -> run dialog
windows+M -> minimize all
windows+shift+M -> minimize all
windows+F1 -> help
windows+E -> explorer
windows+F -> find files
windows+ctrl+F -> find computer
windows+tab -> cycle through taskbar buttons
windows+break -> system properties
app key -> context menu
Win32 -> Win32
alt+F6 -> eaten (isabel to audrey, but not audrey to isabel)
check accessibility shortcuts work on secondary
check accessibility features work on secondary
---
Accessibility Shortcuts
Key Result
-------------------------------------------------------------------
Tap SHIFT 5 times Toggles StickyKeys on and off.
Press down and hold the right Toggles FilterKeys on and off.
SHIFT key for 8 seconds
Press down and hold the NUM LOCK Toggles ToggleKeys on and off.
key for 5 seconds
Left ALT+left SHIFT+NUM LOCK Toggles MouseKeys on and off.
Left ALT+left SHIFT+PRINT SCREEN Toggles High Contrast on and off.
---
disable ctrl+alt+del info:
SystemParametersInfo(SPI_SETSCREENSAVERRUNNING, TRUE, &dummy, 0)
fools system into thinking screen saver is running
that disables ctrl+alt+del, alt+tab, ctrl+esc
dunno if keystrokes are queued
may be limited to win 95/98/me
win nt sp 3 and above:
low-level keyboard hook is called before ctrl+esc, alt+tab, alt+esc
use that to capture and disable
seems that low-level hook is notified of ctrl+alt+del but can't stop it
win nt sp 2 and below
ctrl+esc can be disabled by replacing task manager (dunno how)
RegisterHotKey() can be used to catch alt+tab and alt+esc
only while app is running, of course
win nt
keyboard filter driver can capture keys (see win NT DDK)
http://216.239.51.100/search?q=cache:q-f03UHhFMMC:www.thescarms.com/VBasic/StopReBoot.asp+alt%2Btab+disable&hl=en
some info on changing keyboard scan code mapping
can be used to disable keys but looks like reboot is required
---
Q179905
suggestion for bypassing hook code when called by debugger.
may prevent system from getting locked up when debugging.
---
win32 rel notes:
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Windows\NoInteractiveServices
can prevent an interactive service from running interactively.
how can we tell if we're not running interactively?
even if not interactive we can use MessageBox() to alert user
use MB_SERVICE_NOTIFICATION flag
must use a separate thread in initialization if called from handler
must close dialog if CTRL_LOGOFF_EVENT is received
see SetConsoleCtrlHandler() and CLog (which uses it)
maybe possible to get around NoInteractiveServices
modify DACLs on windows station/desktops to impersonate logged on user
see "Interacting with the User by a Win32 Service"
---
unix release notes:
restarting fails to connect until a user logs in:
Xlib: no protocol specified
this is caused by login manager
all clients are rejected, except those started by manager itself
workaround is to have synergy started by login manager
should then use --no-restart, though that's not necessary
affects client and server
cannot switch screens on login screen:
xdm,kdm grab keyboard for duration of login screen
synergy cannot switch unless it can grab keyboard
gdm doesn't appear to grab keyboard or mouse for duration
affects server, only
=== feedback ===
send all toggle states instead of (individual) toggle ups/downs (gspencer)
Actually, Teleffect already does that too, but what I meant was that
each time the state of a "toggle" key (caps lock, scroll lock, num lock)
changes on the master, I think it should force the client to take on
that state, instead of just toggling on the client. That way when Word
(on the client) helpfullly turns off the caps lock, it doesn't get
toggled back on when I turn off the caps lock on the master. Ideally,
it would just notice that Word turned off the caps lock on the client
and turn it off on the master (and all other clients) too, but I think
that's probably too complicated to implement.
double clicking NT binaries fails (liz)
add note to README or use separate binaries for services. client
would also need a dialog to query options. server would probably
need one too. use dialogs if there are no command line args.
how to exit (liz)
no good way to exit on windows
win32 console disappears too quickly (liz)
maybe show a message box if exiting to due an error
better support for reloading configuration (liz)
use SIGHUP on unix
use an app message on win32
provide access through a control panel or taskbar icon
non-functional on ctrl+alt+del screen in win2k (kurt)
i suppose it must be when win2k is the client. mouse input wasn't
working. keyboard probably didn't work either.
gspencer:
OK, I've narrowed it down a little -- it seems to happen if you have any
key down (I tried the shift key and the letter 's' key) when you cross
over from master to client, which would make me suspect the "capture"
code, since if you have a key down you aren't supposed to be able to
cross over...
gspencer:
bouncy mouse at edge of screen (win32 server). almost certainly
related to changes that fixed mouse delta calculation.
== fixed? ===
dragging windows is too slow
grace (as client) exhibits this
dunno if it happens with audrey2 as the client
liz says her windows laptop does it too
also closing/iconifying has a delay
dragging multiple files also sluggish
liz seems to still be seeing sluggish cursor updates
---
automake stuff
rebuild:
* 'autoheader' - creates config.h.in
* 'touch NEWS README AUTHORS ChangeLog'
* 'touch stamp-h'
* aclocal - adds aclocal.m4 to directory. Defines some m4 macros used by the auto tools.
* 'autoconf '- creates configure from configure.in
* 'automake' - Creates Makefile.in from Makefile.am
* './configure' - creates Makefile from Makefile.in
* 'make'
funky functions:
snprintf -- ick!
some return negative value -- no problem
some return truncated length -- uh oh
some return buffer length without truncation -- huh?
some actually overrun the buffer -- must not build on those!
use AC_LIBOBJ(function) for missing/broken functions
adds function.c to build
must add @LIBOBJS@ to *_LIBADD
or AC_REPLACE_FUNCS(function ...)
automatically calls AC_LIBOBJ for missing functions
use AC_CHECK_FUNC or AC_CHECK_FUNCS to detect missing functions
use AC_CHECK_LIB first if function not in std C lib
maybe AC_CHECK_TYPE for size_t and ssize_t
replace sys/time.h with:
#if TIME_WITH_SYS_TIME
# include <sys/time.h>
# include <time.h>
#else
# if HAVE_SYS_TIME_H
# include <sys/time.h>
# else
# include <time.h>
# endif
#endif
what does `make install' do?
check in derived files?
aclocal.m4
config.h.in
configure
used library functions:
accept
bind
chdir
close
connect
dup
fork
fprintf
getenv
gethostbyaddr
gethostbyname
gethostname
getpeername
getprotobyname
getprotobynumber
getservbyname
getservbyport
getsockname
getsockopt
gettimeofday
inet_addr
inet_ntoa
ioctl
listen
memcpy
memmove
memset
nanosleep
open
openlog
poll
read
recv
recvfrom
send
sendto
setsid
setsockopt
shutdown
sigaction
sigaddset
sigemptyset
socket
sprintf
strcat
strcmp
strerror
strrchr
syslog
umask
vsnprintf
write
included files
#include <arpa/inet.h>
#include <fcntl.h>
#include <pthread.h>
#include <pwd.h>
#include <shlobj.h>
#include <signal.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <syslog.h>
#include <unistd.h>
these put stuff in std::. replace with .h versions or use std::?
3 #include <cctype>
3 #include <cerrno>
1 #include <clocale>
3 #include <cstdio>
4 #include <cstdlib>
12 #include <cstring>
1 #include <ctime>
can we use AC_REPLACE_FUNCS and only build the missing function files
in a single directory. do we have to have a separate configure.in
script to do that?