| /[rdesktop]/jpeg/rdesktop/trunk/xwin.c |
Parent Directory
| 
Revision Log
| Links to HEAD: | (view) (annotate) |
| Sticky Revision: |
Date: Sun, 19 Jul 2009 14:00:30 -0400 From: Daniel Jarboe <daniel.jarboe(at)gmail.com> To: Dobrica Pavlinusic <dpavlin(at)rot13.org> Subject: Re: rdesktop diff On Sun, Jul 19, 2009 at 8:01 AM, Dobrica Pavlinusic <dpavlin(at)rot13.org>wrote: > > I'm looking forward for your patch, and will keep you informed what I > did with it :-) This diff was an earlier one with some debug printfs in so you can see how the tuning knobs work. I was only keeping an eye on a young computer user's activities so ignored small screen changes and wrote out the stills to lossy jpegs with libjpeg. If you plan on saving screens no matter if there are changes or not then I'd throw away all the g_bitmap_data_last_write stuff. diff is against rdesktop-1.6 with whatever patches are in ubuntu 9.04 jaunty. Added libjpeg62-dev in addition to the regular rdesktop build dependencies. Have fun and thanks for letting me know what you end up with. ~ Daniel
branch for integration of Daniel Jarboe <daniel.jarboe(at)gmail.com> patches for jpeg
Moving rdesktop trunk to correct location.
Ran indent-all
brush cache for > 2 color brushes
update the copyright year
Indented
Fix bug #1886199 (tracked down by nicolyon@users.sourceforge.net).
Moved call to seamless_restack_test from ui_init to end of ui_create_window. This seems to fix problems the WTS disconnecting the session due to "DATA ENCRYPTION" errors. At this point, it's unknown why this patch fixes the problem, but ui_create_window is a good place for seamless_restack_test() in any case.
Decreased timeout in sw_wait_configurenotify to one second, to reduce frusrtation with old metacity versions. Need to use gettimeofday.
Re-worked the support for SeamlessRDP window stacking: * Since many window managers cannot properly restack a window between two other windows, we need to check for this at startup. * A new utility function, sw_wait_configurenotify, is used to wait for the WM to process our restacking request. * We are using XReconfigureWMWindow instead of XRestackWindows and XRaiseWindow, to easier meet the demands of ICCCM. Restacking between is only done if the WM is not broken, though. * The error handler does not ignore BadMatch from ConfigureWindow requests any longer. I haven't found any WM that gives BadMatch for XReconfigureWMWindow. * The test cases has been updated to test more stacking cases. A somewhat related bug fix wrt focus handling is also included, which prevents FOCUS messages when reverting focus from a destroyed window
SeamlessRDP bugfix: The first window to XRestackWindows should be the reference window (the "behind" window); not the window to restack.
Applied patch: [ 1715358 ] seamlessrdp window close
Applied rdesktop part of patch: [ 1715359 ] seamlessrdp topmost windows
32 bit color
Implement support for icons in SeamlessRDP.
Send physical mouse buttons over RDP rather than logical ones. Fixes bug 1693905. Patch from Ilya Konstantinov (slightly modified)
[PATCH] rdesktop segfault with libx11-6 1.0.3-7 by Mark Heily <mark@heily.com>
Call ewmh_set_wm_name for main window as well, not just SeamlessRDP windows. This allows specifying an UTF-8 title with the -T argument.
prefix BOOL with RD_
allow inclusion of Xproto.h by renaming their BOOL to XPROTO_BOOL
copyright year update
RD_ prefix and win32 compatibility for core files
Abstract select() handling in rdpsnd so that backends can do their thing more correctly.
Prevent segfaults in out of memory conditions by checking the pointer returned from XGetImage.
Rewrite the queue management a bit so that blocks are not completed until they have finished playing. This also makes the queue system mandatory for all backends.
allow multiple sound-drivers to be compiled in simultaneously and make the runtime selectable
Indent fixes
Fixed a minor seamless bug in DRAW_ELLIPSE: Was copying from backingstore before updated.
set screen-number in XGetVisualInfo (patch 1417414 by Jeremy Smith)
Add a destroy group command to SeamlessRDP for when entire groups of windows get killed off in one go.
The default color depth is now the depth of the root window.
Compile fixes for GCC 2.96. Fixes 1444824.
Make sure we notify other rdesktops that we exited so that they can update their clipboard status.
Added SeamlessRDP support: Merged seamlessrdp-branch
* If fullscreen mode, set g_using_full_workarea = True; * When using -g workarea, if workarea fetch fails, do not set g_using_full_workarea. Fall back to size of entire screen, instead of 800x600.
Applied rdesktop-visual-host-be-fix.patch from #1390148, which fixes a problem when rdesktop is run on a machine with different byte order than the Xserver.
Applied patch #1390148 from Ilya Konstantinov: Refactoring of color depth code.
Corrected spelling error.
In translate24to32, only read 3 bytes from data buffer. If we are reading 4 bytes, we are reading beyond the buffer. Fixes valgrind warnings.
Applied patch #1363476: -D option doesn't work on 64bit system.
Another try at the fix for short 0xff and 0xfe commands: * On short commands, print a warning as well as a debug string. * For 0xfe, allow one byte commands. In this case, do not increment x/y; assume that the third byte is zero.
cache_get_text never returns NULL, so check if entry->data is NULL instead.
Handle too short 0xff and 0xfe commands in ui_draw_text. This fix should remove both a segfault and a "this shouldn't be happening" abortion, which I can trigger with vertical text in OpenOffice 2.0.
ellipse fix for non copy rop
Disable moving in single-app mode when window is occupying entire workarea.
Moved mouse button handling to separate function, handle_button_event. The single app mode has been enhanced: The minimazation now works better.
Protection against recursive sequences
Implemented support for keyboard "sequences", which makes it possible to send multiple scancodes to the RDP server in response to one X11 keyboard event.
Tell the server to stop sending window updates when the rdesktop window is unmapped
Handle RDP recv queue even in RDP send queue is full. Patch from James Cameron <james.cameron@hp.com> This closes Debian Bug #246461.
Window-placement patch by <gregmhughes@comcast.net> small modifications by me, to only set the PPosition sizehint, if the position was specified on the command-line
do polyline in a single ui operation: ui_polyline
Fix text2 order parsing. The brush is parsed but not yet used.
Add support for ellipse and polygon orders
bump version to 1.4.0 change year in files add missing docu for updated commandline-flags update changelog
fix missing while. This garbled data when displaying on a BE host running at 32bpp, when rdesktop is running on a LE host connecting with 16bpp to the TS
compile fix for older gcc
More optimisations
Indent fixes
add support for specifying the window-position with the -g parameter
Big serial- and disk-redirection update from Andreas Flick <Andreas.Flick@unicon-ka.de>
null_pointer_data must actually be 24 bytes. Fixes another Valgrind warning.
The current version of ui_create_cursor requires that null_pointer_data is nine octets. This patch fixes valgrind errors.
gui should really have higher priority then sound...
oss sound fixes
added persistant bitmap chaching from Jeroen Meijer, slightly modified
added ui_begin/end_update and rdp_loop for ports
clean up and comment the repeat macros
warnings when screen size of depth change and backstore resizeing when screen size changes
Indent fixes
fix for colourspace-expansion
Indent fixes.
fix compiler warning/error "a cast does not yield a lvalue"
Indent fixes.
speed up screenblt if screen is unobscured
only do stipple ops on backingstore when possible
unroll the bits in translate8to16/32
new: ui_resize_window() and related, which is used when resizing while shadowing. And fallback for color when connecting to a session with fewer colors than you have set in your session. Jeroen Meijer jeroen@oldambt7.com
Indent fixes
rdp2vnc fixes from Johannes Schindelin <Johannes.Schindelin@gmx.de>
return data only if 15 or 16 bbp in translate_image
if rdp server and client bpp match and architectures match, don't translate image
only check g_xserver_be once in translate8toXX
XEmbed-patch by Arend van Beelen jr. <arend@auton.nl>
Sometimes, the boxcx value is something really large, like 32691. This makes XCopyArea fail with Xvnc. If boxcx is this large, reset to g_width - boxx. (The reason why boxcx is strange is that os->boxright is strange. I have no idea what's causing this, though.)
Indent fixes
optimize split_colour15 and 16
Make internal backing-store the default && small fix for it
fix select errors due to exploding fdnum n
fixes for HP-UX
tmp disable select timeout - we segfault
reworked the asyncio some and completed it. parallel and serial uses it now. disk.c is a likely candidate for the future, but there are some problems with disk.c that we will have to look into first. parallel and serial works better, they do not hang the session, but are still yerky at large chunks... possibly we could split the chunks, but the terminal server does not seem to like partial transfers, i've tried some variations.. :) fns->write() could be split in smaller pieces, but what should the thresholds be.
Ran indent-all.sh
fix g_null_pointer memleak fixes for compiles on NetBSD & FreeBSD
fix g_gc memleak
Try to use a TrueColor visual, even if the default-visual is PseudoColor. This enables high-colour modes on displays with a PseudoColor default visual and eliminates colourmap-flashing at 8 bpp there.
ui_draw_text optimize and small fix
When mixing -f and -g, the last option should take precedence.
compile when not all baudrates are defined on a system
Re-worked numlock handling: Keeping remote numlock state in sync with local state
Indent fixes
Fix crash bug when toggling fullscreen-mode with fancy cursors. Do no longer destroy the global null_cursor, when destroying the current window.
Need unistd.h for select.
Fix compile warnings on HP-UX.
translate24to{24,32} were not using make/split-colour which makes
them unuseable on BE RGB-Displays of this depth...
This should fix it.
Indent fixes
Fix translate16to16 along the lines of all the other translate functions.
A more general fix for the last problem - fixes 15-bit X as well.
Hopefully make it work with both RGB and BGR X servers.
endianess fixes for 8 bit fix for crash when g_bpp==24
Convert bpp printf into DEBUG_RDP5 (perhaps not the ideal category but it will do for now).
private colour map code only works for 8 bpp, disable otherwise (from neoware)
Applied null cursor patch from jeroen@oldambt7.com
Make sure that g_dsp_busy is initialized to False
added switch (-A) to disable rdpsnd
Support for specifying geometry in terms of percent of whole screen
Fixed indentation
xkeymap_init must be before using g_enable_compose.
Corrected SET_FOREGROUND/SET_BACKGROUND in Hatch handling routine. This should be done according to patch 688390, and feels right, although I basically have no idea of what I'm doing :-)
big endian fixes
Make some variables static.
Fix crash bug when dsp_fd = -1.
Preliminary sound support (PCM only). Based on code from GuoJunBo <guojunbo@ict.ac.cn>
big edian support for 32 bpp xwindows, still need 16 and 24
Kbd fix: Shift was always reset after press
restore_remote_modifiers takes ev_time argument.
Indent fixes
more g_ prefix for global vars
kbd: Restoring changed remote modifiers after keypress. Fixes problem with shift & altgr 'stuck' down after typing, for example, A
Fixed memory leak with colormap allocation.
more g_ prefix for global vars
more g_ prefix for global vars
Indent fixes
Commit of work in progress on channels (so that other people can hack on RDPSND), in particular: * channel layer takes care of virtual channel header * split X dependent parts out of CLIPRDR, simplified IPC implementation * initial RDPDR implementation
Listen for clipboard-related events and handle them. Listen for IPC-related events and handle them. Changes after running indent-all.sh
changes so it compiles with g++(mainly for ports)
support for hatch patterns in patblt
Fix an endianness problem.
Support for moving window by dragging applications title bar. To be used in conjunction with -s, -S and -D.
Prevent restoring by double click in single app mode.
Prevent access to sys menu in single app mode
New option for running single applications maximized
Private funcs should be static.
Indent fixes
15 and 24 bit server support
support for > 16 bit clients with 16 bit server
16bit
unsigned long != uint32 on a 64-bit machine.
Fix a few compile warnings.
Add some signed types and use them in the places we have silly casts now.
Add a warning function, change some errors to warnings, improve a couple of error messages (not sure whether scewed was meant to be skewed or screwed :))
Rename FILL_RECTANGLE_FAST to FILL_RECTANGLE_BACKSTORE to clarify what it does
focus fullscreen window if not focused when mouse moves, it seems some display managers don't focus new windows, this will provides a way to fucus it
make ui_draw_text a bit safer when cache_get_text returns NULL
minor draw_text optimize
owncolmap reintroduced, basically the same code, but with a switch -C. didn't use the -v switch as in the patches. I believe it is confusing and should be reserved for -version information. Other than that I can say that the code does not work as it does in the patches... haven't found what's wrong with it. bit shifting?
ui_draw_text optimization from unified patches
Applied WM-kill patch from Matthew Riechers (slightly modified
Small indent fix.
Made -g workarea a bit less fatal when the WM does not support _NET_WORKAREA: defaulting to 800x600.
exit()ing after "this shouldn't be happening".
Support for extended WM hint _NET_WORKAREA, via -g workarea
Support for hiding WM decorations
Indent fixes.
can't grabkeyboard on focusin unless mouse is in window, can't grabkeyboard on enternotiy unless window is focused, cache them, other grabkeyboard fixes
Need to make absolutely sure rdesktop ungrabs the keyboard - do it on LeaveNotify as well.
Focus upon receiving EnterNotify in fullscreen mode (in lieu of window manager)
Show display name in "Failed to open display" message.
Arguments to get_key_state the wrong way around.
Need to wait for visibility before calling XSetInputFocus (pointed out by Jay).
Update copyright dates on all files that have changed. Bump version to 1.2-cvs.
Oops typo... should have tested it first :)
Optimise get_key_state by caching the modifier map and the state (this can probably still be improved in the future). Only call time(NULL) for the events where it's needed.
Fix for flashing title bar problem.
Indent fixes.
Move keyboard grabbing to FocusIn/FocusOut to fix "rdesktop gets screensaver password" problem. Hopefully this won't bring back the phantom flashing titlebar...
foo function() -> foo function(void) to conform to ANSI C.
Make no translation messages DEBUG_KBD. Clean up includes, fix a few compile warnings, avoid rdesktop.h depending on X (at the expense of KeySym -> uint32 - maybe in the future we need a better proto scheme).
Reinstate a cleaned-up version of the old xwin_toggle_fullscreen - I indeed can't get it working reliably with all window managers without creating a new window. Move remaining non-window dependant code into ui_init/ui_deinit so that xwin_toggle_fullscreen can do ui_destroy_window/ui_create_window.
Owncolmap removed in favour of more intelligent color allocation.
Small line-break fix for DEBUG_KBD.
xwin_reset_keys -> reset_modifier_keys. Only when necessary
Small indent fix.
Now that we have ui_init move rounding of geometry there (it's X-specific).
Possible fix for "flashing title bar" problem (don't grab in response to FocusIn/FocusOut - seems to cause an event loop).
Fix rdesktop not resizing properly in xwin_toggle_fullscreen with some window managers.
Rewrite of toggle_fullscreen so it modifies the attributes of the existing window instead of creating a new one.
Move non-window-related initialisation to ui_init. All functions after ui_init can now assume display and other globals are valid.
Fixes Excel scroll-of-death problem: earlier select-loop failed to call xwin_process_events when the queue was non-zero but the X11 socket was empty.
Support for Windows keys (via Ctrl-Esc)
Inhibited keys are defined in keymap files; not hardcoded
Fixes NumLock problems; using RDP_INPUT_SYNCHRONIZE
Make sure width is a multiple of 4 (previous version in ui_create_window was getting done too late, after connection).
Alt modifier for fullscreen toggle
Indent fixes
toggle full screen
focus out reset keys so some keys do not stick down
clear window upon startup so cached data is not viewed on screen from up19-7-5
Got rid of var-arg macros
Changed max line length to 100
added ui_init for pre connect inits like fullscreen
patblt reset orgin from up19-7-4
Indentation fixes
Applied patch from Bob Bell for -K option
Applied patch from Bob Bell (#2)
Applied patch from Bob Bell (#1), which fixes compilation warning messages.
Add support for enabling/disabling local compose handling. Compose support is now disabled per default
Added "localstate" support to keymapping: Send local modifier state. This fixes problems with Shift-Home etc
Merged new keysym-based keyboard handling (from alt. CVS repos)
Fixed indentation with indent
upside down pattern in patblt
using XEvent instead of XkbEvent. better tracking of modifiers. ( TODO: 1. sync at start up, 2. fallback for missing or improperly configured xkb module. )
Revert to old behaviour of connecting before creating the UI. This is to stop the UI generating input before the connection is complete (which causes connection failures). The one difficulty is that ui_select is now called before ui_create_window, which may cause problems for other UIs - we may still need to change this in the future.
adds support for accessx, makes it possible for people with disabilities to use rdesktop. Relies on XKB. Sync problem with changed modifiers on focus in not fixed yet...
There is an extended key flag that we should be setting for extended keys. (Originally fixed by Ben McKeegan <Ben.McKeegan@fitz.cam.ac.uk>) Ran make proto.
This fixes TEXT2(cache) procesing with corrupted screen with long lines and vertical text.
Bugfix for ERROR: get font messages, particularly with vertical text.
Fix for rdesktop hanging when it loses the X server connection. (reported by Per Kristian Hove <Per.Hove@math.ntnu.no>)
Committing my keymap work - unlike the version in the unified patches, this one uses external keymap files (which are my preference).
OSF1/Alpha build fixes.
Be very careful with if and macros...
Called translate_colour one too many times.
Add a ui_select to xwin.c to reduce latency. Remove extraneous error messages - only report at lowest level. Endianness and IRIX compile fixes.
Software backing store support.
Portability fixes, including elimination of variable argument macros. Rudimentary configure script. Miscellaneous cleanups.
Major cleanups, particularly in X code.
Merges from pl19-6-5.
Changed indentation style (-psl).
ran indent (-bli0 -i8 -cli8 -npcs -npsl)
Fix for big-endian X-servers: rely on server to handle byte/bit order.
Added a number of command line options including autologon. Inverted sense of -m (the default is now to send mouse move events). Preparing for release of 1.0.0.
Respect x offset in font information.
Fixes (?) for font problems with Microsoft Office Added URL to website http://www.rdesktop.org/ Releasing 0.9.0-alpha2
Added specific handler for triblt opcode 0x69 (used by Microsoft Word), and fixed a typo in the default handler. Fixed implementation of desktop cache code, which was sometimes overlapping saves.
Portability fixes for Solaris and OSF1.
Major commit of work from laptop - done in various free moments. Implemented encryption layer and some basic licensing negotiation. Reorganised code somewhat. While this is not quite as clean, it is a lot faster - our parser speed was becoming a bottle-neck.
Committing some awesome progress I made while overseas - this commit really embodies a huge number of changes. We are now able to talk quite fluently to a French NT Terminal Server - in normal usage only minor font issues remain (handling of TEXT2 order is not perfect). The next major hurdle is encryption, and it will be quite a big hurdle - there seems to be some quite nasty session key stuff.
Miscellaneous updates: implemented some more protocol features including colour maps. Started on a new bitmap decompression engine which is not completely working yet - however I am going back on the road so I am committing now.
Started hacking on an X-Windows (Xlib) interface. Currently pops up a window and displays bitmaps it sees side by side. Next step is to go back to the protocol and interpret the surrounding data stream.
This form allows you to request diffs between any two revisions of this file. For each of the two "sides" of the diff, enter a numeric revision.
| ViewVC Help | |
| Powered by ViewVC 1.1.26 |