We already have state fully stored within XKB, so instead of duplicating it,
just generate the values to send to clients when required.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
XkbInitKeyboardDeviceStruct is now the only valid keyboard
initialisation: all the details are hidden behind here. This now makes
it impossible to supply a core keymap at startup.
If dev->key is valid, dev->key->xkbInfo->desc is also valid.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
No more #ifdef XKB, because you can't disable the build, and no more
noXkbExtension either.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Note: properties don't need to be cleaned up, the DIX does it for us anyway.
Data that is stored in properties is cleaned up by the property system.
Handlers, etc. don't need to be unregistered while cleaning up, as they get
deleted when the device is removed anyway.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Signed-off-by: Simon Thum <simon.thum@gmx.de>
If the MD's lastSlave was a devices with custom axes ranges, then a
WarpPointer would position the cursor at the wrong location. A WarpPointer
request provides screen coordinates and these coordinates were scaled to the
device range before warping.
This patch consists of two parts:
1) in the WarpPointer handling, get the lastSlave and post the event through
this device.
2) assume that WarpPointer coordinates are always in screen coordinates and
scale them to device coordinates in GPE before continuing. Note that this
breaks device-coordinate based XWarpDevicePointer calls (for which the spec
isn't nailed down yet anyway) until a better solution is found.
X.Org Bug 19297 <http://bugs.freedesktop.org/show_bug.cgi?id=19297>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
positionSprite needs to scale to screen coordinates and in the process of
doing so alters dev->last.valuators[0:1]. Drop the real coordinates back after
finishing and before updating the motion history. This way, we don't push the
screen coordinates into the motion history.
X.Org Bug 19285 <http://bugs.freedesktop.org/show_bug.cgi?id=19285>
This commit moves the focus handling from events.c into enterleave.c and
implements a model similar to the core enter/leave model.
For a full description of the model, see:
http://lists.freedesktop.org/archives/xorg/2008-December/041740.html
This commit also gets rid of the focusinout array in the WindowRec, ditching
it in favour of a local array that keeps the current focus window for each
device.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Instead of keeping a flag on each window for the devices that are in this
window, keep a local array that holds the current pointer window for each
device. Benefit: searching for the first descendant of a pointer is a simple
run through the array.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The old model was implemented based on a misunderstanding of NotifyVirtual and
NotifyNonlinearVirtual events. It became complicated and was broken in some
places [1]. This patch wipes this model completely.
A much simplified implementation is provided instead. Rather than a top-down
approach ("we have a tree of windows, which ones need to get which event")
this one uses a step-by-step approach. For each window W between A and B
determine the pointer window P as perceived by this window and determine the
event type based on this information. This is in-line with the model described
by Owen Taylor [2].
[1] http://lists.freedesktop.org/archives/xorg/2008-December/041559.html
[2] http://lists.freedesktop.org/archives/xorg/2008-August/037606.html
Currently, if a device map differs from the core pointer map, then the
request may return MappingBusy, even though all the affected core
buttons are in the up state.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The builtin-fonts configure option was removed, as it at best should
have been a runtime option. Instead, now it always register all "font
path element" backends, and adds built-ins fonts at the end of the
default font path.
This should be a more reasonable solution, to "correct" the most
common Xorg FAQ (could not open default font 'fixed'), and also don't
break by default applications that use only the standard/historical
X Font rendering.
Devices are only activated once - right after they've been added to the
server. If a device failes activation, it's dead. There's no reason to
continue. Return the error code from ActivateDevice() without setting up
sprite information or even sending a event to the client.
Then - in the DDX - just remove the device again.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If a device hasn't been initialized, it doesn't have a cursor yet. So don't
set the cursor to the NullCursor, and don't try to DisableDevice either.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This still doesn't fix reset the MD's classes (a TODO that's been here for
ages), but at least we don't segfault anymore when detaching the last SD.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Fallout from aeff14d5f2. Yes, we don't malloc
anymore because we are inside a SIGIO and the memory is already there anyway.
But we still need to set the event length correctly, otherwise
mieqEnqueue/mieqProcessInputEvent don't know how much memory to copy.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We mustn't realloc as we are inside a signal handler. With
SetMinimumEventSize, this code should never be hit anyway, as the event list
should have the required memory before this code is hit.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
A grep on xorg/* revealed there's no consumer of this define.
Quote Alan Coopersmith:
"The consumer was in past versions of the headers now located
in proto/x11proto - for instance, in X11R6.0's xc/include/Xproto.h,
all the event definitions were only available if NEED_EVENTS were
defined, and all the reply definitions required NEED_REPLIES.
Looks like Xproto.h dropped them by X11R6.3, which didn't have
the #ifdef's anymore, so these are truly ancient now."
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Signed-off-by: Adam Jackson <ajax@redhat.com>
Follow-up to 4971315296. countValuatorEvents was copied from GKVE where it
was obviously broken but nobody noticed. GPE had the correct version, but that
one got lost during de-duplication. Restoring the correct calculation - if we
have 6 valuators, we want 1 valuator event, not 2.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Save in a few special cases, _X_EXPORT should not be used in C source
files. Instead, it should be used in headers, and the proper C source
include that header. Some special cases are symbols that need to be
shared between modules, but not expected to be used by external drivers,
and symbols that are accessible via LoaderSymbol/dlopen.
This patch also adds conditionally some new sdk header files, depending
on extensions enabled. These files were added to match pattern for
other extensions/modules, that is, have the headers "deciding" symbol
visibility in the sdk. These headers are:
o Xext/panoramiXsrv.h, Xext/panoramiX.h
o fbpict.h (unconditionally)
o vidmodeproc.h
o mioverlay.h (unconditionally, used only by xaa)
o xfixes.h (unconditionally, symbols required by dri2)
LoaderSymbol and similar functions now don't have different prototypes,
in loaderProcs.h and xf86Module.h, so that both headers can be included,
without the need of defining IN_LOADER.
xf86NewInputDevice() device prototype readded to xf86Xinput.h, but
not exported (and with a comment about it).
Rather than assuming rules in the CoreKeyboardProc, init the default rules in
InitCoreDevices, then re-use them later.
In the xfree86 DDX, set the rules to "base" or "evdev", depending on whether
we'll load kbd or evdev.
If we create a new MD, use pc105,us as default and re-use the rules file used
previously.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
The device's button down state array was changed to use DOWN_LENGTH and thus
bitflags for each button in cfcb3da7.
Update the DBSN events to copy this bit-wise state.
Update xkb and Xi to check for the bit flag instead of the array value.
Reported by ajax.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
This is the biggest "visibility" patch. Instead of doing a "export"
symbol on demand, export everything in the sdk, so that if some module
fails due to an unresolved symbol, it is because it is using a symbol
not in the sdk.
Most exported symbols shouldn't really be made visible, neither
advertised in the sdk, as they are only used by a single shared object.
Symbols in the sdk (or referenced in sdk macros), but not defined
anywhere include:
XkbBuildCoreState()
XkbInitialMap
XkbXIUnsupported
XkbCheckActionVMods()
XkbSendCompatNotify()
XkbDDXFakePointerButton()
XkbDDXApplyConfig()
_XkbStrCaseCmp()
_XkbErrMessages[]
_XkbErrCode
_XkbErrLocation
_XkbErrData
XkbAccessXDetailText()
XkbNKNDetailMaskText()
XkbLookupGroupAndLevel()
XkbInitAtoms()
XkbGetOrderedDrawables()
XkbFreeOrderedDrawables()
XkbConvertXkbComponents()
XkbWriteXKBSemantics()
XkbWriteXKBLayout()
XkbWriteXKBKeymap()
XkbWriteXKBFile()
XkbWriteCFile()
XkbWriteXKMFile()
XkbWriteToServer()
XkbMergeFile()
XkmFindTOCEntry()
XkmReadFileSection()
XkmReadFileSectionName()
InitExtInput()
xf86CheckButton()
xf86SwitchCoreDevice()
RamDacSetGamma()
RamDacRestoreDACValues()
xf86Bpp
xf86ConfigPix24
xf86MouseCflags[]
xf86SupportedMouseTypes[]
xf86NumMouseTypes
xf86ChangeBusIndex()
xf86EntityEnter()
xf86EntityLeave()
xf86WrapperInit()
xf86RingBell()
xf86findOptionBoolean()
xf86debugListOptions()
LoadSubModuleLocal()
LoaderSymbolLocal()
getInt10Rec()
xf86CurrentScreen
xf86ReallocatePciResources()
xf86NewSerialNumber()
xf86RandRSetInitialMode()
fbCompositeSolidMask_nx1xn
fbCompositeSolidMask_nx8888x0565C
fbCompositeSolidMask_nx8888x8888C
fbCompositeSolidMask_nx8x0565
fbCompositeSolidMask_nx8x0888
fbCompositeSolidMask_nx8x8888
fbCompositeSrc_0565x0565
fbCompositeSrc_8888x0565
fbCompositeSrc_8888x0888
fbCompositeSrc_8888x8888
fbCompositeSrcAdd_1000x1000
fbCompositeSrcAdd_8000x8000
fbCompositeSrcAdd_8888x8888
fbGeneration
fbIn
fbOver
fbOver24
fbOverlayGeneration
fbRasterizeEdges
fbRestoreAreas
fbSaveAreas
composeFunctions
VBEBuildVbeModeList()
VBECalcVbeModeIndex()
TIramdac3030CalculateMNPForClock()
shadowBufPtr
shadowFindBuf()
miRRGetScreenInfo()
RRSetScreenConfig()
RRModePruneUnused()
PixmanImageFromPicture()
extern int miPointerGetMotionEvents()
miClipPicture()
miRasterizeTriangle()
fbPush1toN()
fbInitializeBackingStore()
ddxBeforeReset()
SetupSprite()
InitSprite()
DGADeliverEvent()
SPECIAL CASES
o defined as _X_INTERNAL
xf86NewInputDevice()
o defined as static
fbGCPrivateKey
fbOverlayScreenPrivateKey
fbScreenPrivateKey
fbWinPrivateKey
o defined in libXfont.so, but declared in xorg/dixfont.h
GetGlyphs()
QueryGlyphExtents()
QueryTextExtents()
ParseGlyphCachingMode()
InitGlyphCaching()
SetGlyphCachingMode()
libXfont has stubs for these symbols, so, when compiling with hidden
symbols by default, these symbols must be visible in the X Server, or
the stubs in libXfont will be used.
I merged the wrong patch. See correct patch at:
http://lists.freedesktop.org/archives/xorg/2008-November/040540.html
Not activating the device before attempting to enable it would leave the
sprite unset, crashing the server when enabling the real devices.
This reverts commit e078901a4e.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
This prevents a protocol visible side-effect (XVisibilityEvent) on
XCompositeRedirectWindow() followed by a XCompositeUnredirectWindow().
The problem shows up in gnome-screensaver with compiz and "unredirect
fullscreen windows" enable. A VisibilityNotify event is generated (first
with obscured and than with unobscured) when the window swithces from
redirected to unredirected.
https://bugs.freedesktop.org/show_bug.cgi?id=18133http://launchpad.net/bugs/278112
As proposed by Owen Taylor [1], the enter-leave event model needs to adjust
the events sent to each window depending on the presence of pointers in a
window, or in a subwindow.
The new model can be summarised as:
- if the pointer moves into or out of a window that has a pointer in a child
window, the events are modified to appear as if the pointer was moved out of
or into this child window.
- if the pointer moves into or out of a window that has a pointer in a parent
window, the events are modified to appear as if the pointer was moved out of
or into this parent window.
Note that this model requires CoreEnterLeaveEvent and DeviceEnterLeaveEvent to
be split and treated separately.
[1] http://lists.freedesktop.org/archives/xorg/2008-August/037606.html
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
FirstPointerChild: Return the first child that has a pointer within its
boundaries.
FirstPointerAncestor: return the first ancestor with a child within its
boundaries.
These are required for the updated enter/leave model.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Device events always need to be delivered, core events only in some cases.
Let's keep them completely separate so we can adjust core event delivery.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Adam Jackson <ajax@redhat.com>
We need them for each window, every time a window is allocated. Storing them
in a devPrivate is the wrong thing to do.
This also removes the unused ENTER_LEAVE_SEMAPHORE_ISSET macro.
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
grab == devgrab anyway, this is a leftover from the time when we had two
different grabs per device (core and XI grab).
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Really, this was a bad idea. It's not security, the UI features that would
have been cool (e.g. clicking through windows) aren't implemented anyway, and
there's nothing you can't achieve just by using plain XI anyway.
Requires inputproto 1.9.99.6.
EnableDevice and DisableDevice both change the property too.
And enabled must be set to FALSE in AddInputDevice, the device is not enabled
yet.
X.Org Bug 18111 <https://bugs.freedesktop.org/show_bug.cgi?id=18111>
dev->enabled is a Bool. Bool is two bytes.
BOOL on the other hand is a protocol type and always 1 byte. So copy the value
into the one-byte type before passing it into XIChangeDeviceProperty.
Found by Michel Dänzer.
X.Org Bug 18111 <http://bugs.freedesktop.org/show_bug.cgi?id=18111>
The current code exposes to inconsistent updates, i.e. if handler N succeeds
but handler N+1 fails in setting the property, an error is returned to the
client although parts of the server now behave as if the property change
succeeded.
This patch adds a "checkonly" parameter to the SetProperty handler. The
handlers are then called twice, once with checkonly set to TRUE.
On the checkonly run, handlers _MUST_ return error codes if the property
cannot be applied. Handlers are not permitted to actually apply the changes.
On the second run, handlers are permitted to apply property changes.
Errors codes returned on the second run are ignored.
For two axes [a, b] and [x, y] (inclusive), the formula to scale point P(ab)
to (x,y) is:
(P - a)/(b - a) * (y - x) + x
And the whole end result rounded of course to get the integer we need.
It's not especially obvious, and unpleasantly overloaded for the Xnest
case. Typically this gives you a server that looks for its auth data in
the authority file you were using for the running X session, which
generally doesn't have an entry for the display you just started.
All the major dm's, and startx, pass -auth explicitly, so this shouldn't
cause too much upheaval.
A property can only be deleted if any of the following is true:
- if a property is deletable and all handlers return Success.
- if a property is non-deleteable and the all handlers return Success AND the
delete request does not come from a client (i.e. driver or the server).
A client can never delete a non-deletable property.
Now that the code has been fixed so that Unmap means unmap and not "don't
remap", 'remap' was confusing to have in the function names/parameters, so
change it to simple 'map'.
Signed-off-by: Keith Packard <keithp@keithp.com>
Since ReparentWindow() does a unmap/map pair for windows that are already
mapped, for saveset windows with SaveSetUnmap, we must unmap the window
before calling ReparentWindow() to avoid the generation of MapRequest
events, and so forth.
For master devices, the ptraccel code could segfault on free since we'd be
dereferencing random memory. Callocing the valuatorClassRec is the easy fix.
This removes yet another xalloc() each server generation. Also, I
couldn't find the corresponding xfree() so I guess that used to be a
memory leak there.
Two corrections
1. the "detail" field has NotifyVirtual, etc., not the "mode" field. This was
a clear bug.
2. don't set/unset the flags for NotifyGrab or NotifyUngrab. Clients are
expected to deal with multiple enter/leave events per window if the mode is
not NotifyNormal.
Testable with TCL menu boxes (such as used in gitk):
tk_optionMenu .menu globVar Val1 Val2 Val3 ValJunk
pack .menu
Thanks to Michel Dänzer for pointing this out.
Unplug a mouse, then warp the pointer and the warp pointer code will try
to update the position of the last slave device associated with the
master. That pointer will be stale and the X server will crash.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This was to account for cases where you had video and print screens in
the same server. Lunacy. Leave the slot in ScreenInfo, but rename it,
and stop looking at it.
Add strncasecmp (as we're now using it) in case someone doesn't have it,
and also change strncasecmp args to be const, in accordance with
everything else.
In the map stored in each keyboard device, the first line refers to
minimum keycode, i.e., the 0th line refers to keycode 8. When not
using XKB the wrong test caused some keys to be interpreted as
locks ('m' for instance). The had to be pressed twice to generate
both KeyPress and KeyRelease events.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We may need more than one handler to deal with a property (e.g. one in the
driver, one in the DIX), so get the handlers into a linked list and call them
one-by-one. This is of course slightly less entertaining than the hilarious
WRAP/UNWRAP game we play in other parts of the server.
XIRegisterPropertyHandler/XIUnregisterPropertyHandler are the interface
drivers/the DIX should use to attach themselves to the device.
XIDeleteAllDeviceProperties destroys everything, including the handlers.
Basically just copied from randr properties, with minor changes only.
Each device supports arbitrary properties that can be modified by clients.
Modifications to the properties are passed to the driver (if applicable) and
can then affect the configuration of the device.
Note that device properties are limited to a specific device. A property set
on a slave device does not migrate to the master.
Basically just copied from randr properties, with minor changes only.
Each device supports arbitrary properties that can be modified by clients.
Modifications to the properties are passed to the driver (if applicable) and
can then affect the configuration of the device.
Note that device properties are limited to a specific device. A property set
on a slave device does not migrate to the master.
This fixes a severe issue - when the client died the event mask didn't get
unregistered and a future event would dereference dangling pointers. By
storing the event masks in the resource system we can free them when the
client dies.
Using id = 0 only worked pre-MPX since XInput didn't allow XOpenDevice for the
core devices (0 and 1). Now we can now legally register for events so we may
overwrite our device-independent classes with the ones selected for the VCP.
So, increase the EMASKSIZE to MAX_DEVICES + 1 and use MAX_DEVICES as the ID
when we don't have a device.
Mixing usage where some parts of the code treated this field as a bitmask
and other parts as an array of card8 was wrong, and as the wire protocol
wanted bitmasks, it was less invasive to switch the newer counting code use
booleans.
Master devices track slave buttons by waiting for all slave buttons to be
released before delivering the release event to the client.
This also removes the state merging code in DeepCopyDeviceClasses -- that
code was changing master device state without delivering any events,
violating protocol invariants. The result will be that existing slave
button state which does not match the master will not be visible through the
master device. Fixing this would require that we synthesize events in this
function, which seems like a bad idea. Note that keyboards have the same
issue.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Peter Hutterer <peter@cs.unisa.edu.au>
Dereferencing into dev->valuator could crash the server, although it looks
like I could only reproduce this by having a keyboard send an event after it
was created and the WM was still replaying. Or so.
device->button->down used to be a 32-byte bitmask with one bit for each
button. This has changed into a 256-byte array, with one byte assigned for
each button. Some of the callers were still using this array as a bitmask
however, this is fixed with this patch.
Thanks to Keith Packard for pointing this out. See also:
http://lists.freedesktop.org/archives/xorg/2008-June/036202.html
Button events were mapped once in GetPointerEvents and then again in
UpdateDeviceState. While it might make sense to just fix up UpdateDeviceState,
it turns out to be better to leave the raw button number in the event because
DGA reports raw device events without button translation, and so when it calls
UpdateDeviceState, the button down counts get scrambled and buttons get stuck
down.
See also:
http://lists.freedesktop.org/archives/xorg/2008-June/036201.html
Signed-off-by: Peter Hutterer <peter@cs.unisa.edu.au>
RealizeCursor should be called when the cursor is allocated. However, when the
root cursor is allocated, no devices exist yet, and thus RealizeCursor is
never called. This may lead to segfaults lateron in DDXes like Xnest that
actually need to do something for each cursor, and lateron rely on that
DDX-specific data for each cursor has been initialized.
We only have one set of default rules options in xkb. When the second keyboard
is brought up with Xkb options specified, these new options overwrite the old.
In future server generations, the rules used for the VCK are a mixture of the
default ones and ones previously specified for other keyboards. Simply
resetting the xkb default rules to NULL avoids this issue.
Reproducable by setting XkbLayout "de" and XkbVariant "nodeadkeys". In the
second server generation, the VCK has "us(nodeadkeys)". This again produces a
SIGABRT when the first key is hit.
I could not figure out why the SIGABRT happens. This patch is avoiding the
issue rather than fixing it.
The core protocol requires absolute values and it's a bit hard to get them if
we only have relative ones in the history. Switch the motion history to
absolute, and if we really need the relative values, we can probably generated
them from the abs. ones in the future.
Add each event to the master's MH as well as to the SDs. In the MD, store
min/max and the actual value. When retrieving the MH, rescale all coordinates
to the current coordinate range and only post those valuators that are
currently active on the device.
Since we can't predict how many valuators may be in a future SD attached to an
MD, we need to preallocate a history buffer that is large enough to keep
MAX_VALUATORS coordinates per event.
In addition, the history buffer needs to memorize the coordinate ranges at the
time, thus requiring MDs to store (min_val, max_val, current_val, time)
instead of (current_val, time) for each motion history entry.
This commit only fixes the allocation.
With the MD/SD device hierarchy we need control over the generation of the
motion history as well as the conversion later before posting it to the
client. So let's not let the drivers change it.
No x.org driver currently uses it anyway, linuxwacom doesn't either so dumping
it seems safe enough.
master->last.valuator[x] for x > 2 is undefined. For all other devices, it's
the respective device's last valuators.
If the lastSlave did not have a valuator that is to be updated now, it is
reset to 0.
In GPE, we don't care about the device mode. Let's put the absolute values
into the deviceValuator event and worry about relative valuators on the other
side of the EQ.
Assuming master->last.valuators is in screen coords, SD's are always in device
coordinates.
1. If an event comes in, scale masters->last to the device, drop into device's
last->valuators.
2. Apply motion from the actual event
3. Scale back to screen coords, check if we may need to cross screens
4. Drop screen coords into master->last
5. Rescale to device coords, drop into deviceValuator event and SD->last
6. Drop screen coords into ev->root_x/y
Whoopee...
During GetPointerEvents (and others), we need to access the last coordinates
posted for this device from the driver (not as posted to the client!). Lastx/y
is ok if we only have two axes, but with more complex devices we also need to
transition between all other axes.
ABI break, recompile your input drivers.
Changed all the checks for x&y valuator so the more complex
calculation is only made once.
Added TODOs for valuator/axis 2 and above for future correct
handling of relative reporting of these.
Signed-off-by: Peter Hutterer <peter@cs.unisa.edu.au>
valuators[] is passed from the DDX. Depending on the device mode, update it
with either absolute values or relative values. The deviceValuator event sent
to the client will then contain the respective values.
This isn't quite finished yet, but at least it gives us the ability to use a
tablet as a normal mouse - with all the scaling in place.
Signed-off-by: Peter Hutterer <peter@cs.unisa.edu.au>
Conflicts:
Xext/xprint.c (removed in master)
config/hal.c
dix/main.c
hw/kdrive/ati/ati_cursor.c (removed in master)
hw/kdrive/i810/i810_cursor.c (removed in master)
hw/xprint/ddxInit.c (removed in master)
xkb/ddxLoad.c
Sometimes we didn't have a cursor when coming back from suspend. Reason was
that the suspend caused the server to lose the device that was attached to the
VCP, and a RemoveDevice() would then set the cursor to NULL.
Solution: only set the cursor to NULL if we actually own the sprite.
We pass in the client that wants to create the device anyway, lets use the
parameter instead of hardcoding the serverClient.
Wow. I hope this is merge detritus, otherwise it'd be a sign that I didn't
have enough coffee that day.
XQuartz was crashing because the Appkit thread was trying to GetXXXEvents while the Xserver thread was exiting.
This adds some more sanity checks and avoids that crash
(cherry picked from commit 34ec4bd6ac)
When a new device is added, calculate the event size needed if a DCCE event is
sent and set the EQ's event size to this minimum. This avoids reallocs when a
event is sent (which may happen during a SIGIO).
This variable was used originally to determine which client is allowed to
change the pointer-keyboard pairing. For now, we just let anyone change it and
see how that works out.
(cherry picked from commit 4e2c6dbabdbbaaca213fd08edd422de15d0900cc)
required because of commit 7c0709a736,
which made requestingClient in dix specific to Xprint only.
Add to XPRINT_LIBS in hw/xprint/Makefile.am in front of
$(XSERVER_LIBS) to override definitions in libdix.la for standard xservers.
Follows 571206832d (providing -DXPRINT
to xprint subdirs).
Note it may be possible to restructure the code so that
requestingClient is stored elsewhere than in dix. See discussions
following http://lists.freedesktop.org/archives/xorg/2008-March/033844.html
If this is done it may be possible to revert this commit (if not 571206...).
Rather than freeing/allocing classes each time the device capabilities need to
swap, store them in the devPrivates system.
When a class is unused, it is pushed into the devPrivates, and later recovered
when needed again. This saves us a lot of memory allocations/frees, admittedly
on the cost of some memory.
The DDX (xfree86 anyway) maintains its own device list in addition to the one
in the DIX. CloseDevice will only remove it from the DIX, not the DDX. If the
server then restarts (last client disconnects), the DDX devices are still
there, will be re-initialised, then the hal devices come in and are added too.
This repeats until we run out of device ids.
This also requires us to strdup() the default pointer/keyboard in
checkCoreInputDevices.
X.Org Bug 14418 <http://bugs.freedesktop.org/show_bug.cgi?id=14418>
The device classes aren't deleted anymore on a class change, so there's no
need to store the MD's original classes. We should however restore the MD to
sane defaults when disconnecting the last device, consider this as TODO item.
Instead of a simple counter, use bits to keep track of which device is where
etc. When device enters a window (or sets focus), the bit matching the device
is set, when it leaves again, it is unset. If there are 0 bits set, then
Leave/Enter/Focus events may be sent to the client.
Same theory as before, but this should get around the insanity with
Grab/Ungrab special cases. Those cases are basically untested though.
If an pointer event is being processed during a device grab, don't deliver it
to the focus window, even if the device has a focus class. Reason being that
some pointers may have a focus class, thus killing drag-and-drop.
InitializeSprite won't create a new one if it already exists, with the result
of overwriting the master's sprite. This master sprite is then assigned to the
floating slave, and freed when the slave is reattached later.
Setting the sprite to NULL forces InitializeSprite to alloc a new one, and
this one can be freed without further repercussions.
Add Prox events to the if-clauses with the other events
that are usually sent from the input devices.
Ensure that the event deliverers won't try to deliver
events of type '0' (some extended events doesn't have
an equivalent core-type)
Small modification by Peter Hutterer.
Signed-off-by: Peter Hutterer <peter@cs.unisa.edu.au>
XkbFinishDeviceInit does the following:
xkbi->kbdProc= pXDev->kbdfeed->CtrlProc;
pXDev->kbdfeed->CtrlProc= XkbDDXKeybdCtrlProc;
If we directly copy the device classes for the VCK, pXDev->kbdfeed->CtrlProc
at the time of copying is still XbkDDXKeybdCtrlProc. So at some point
XkbDDXKeybdCtrlProc is called, and calls itself, and calls itself, and...
Setting the device's classes to NULL seems to fix things. The memory isn't
lost, it gets stored into the devPrivates and freed at device closing time.
To recap: the original XC-SECURITY extension disallowed background "None" if
the window was untrusted. XACE 1.0 preserved this check as a hook function.
XACE pre-2.0 removed the hook and first abolished background "None entirely,
then restored it as a global on/off switch in response to Bug #13683.
Now it's back to being per-window, via a flag instead of a hook function.
Before we enable the device through the driver, we size it up and make sure
that the events in the event list contain enough bytes for a possible
ClassesChangedEvent lateron.
The latter is used to increase the amount of allocated memory for the event
list. This will be needed for ClassesChangedEvents that can be of more or less
arbitrary size (larger than 32 anyway).
Rather than letting the DDX allocate the events, allocate them once in the DIX
and just pass it around when needed.
DDX should call GetEventList() to obtain this list and then pass it into
Get{Pointer|Keyboard}Events.
AllowAll is the last check before the parent window is checked. This allows
to override a DenyAll in the parent window in a simpler way than before (the
previous method required all devices to be in the permit list).
In the case that the device cursor was the first in the list of cursors
the window knew about, unsetting the cursor could lead to a segfault
due to pPrev being NULL. Instead catch the condition and correctly remove
the node from the list. Since there is no cursor now set on the device,
we simply return success as the parent windows cursor will propogate
down later.
Signed-off-by: Peter Hutterer <peter@cs.unisa.edu.au>
Relative events that generates both core and extention
events will have its axis cliped and screen changed by
miPointerSetPosition when the events are processed. For
absolute and non core-generating relative events the
axis must be clipped if we shouldn't end up completely
outside the defined ranges (if any).
Don't use a possitive value as a marker for if a max-value
is defined on the valuators. Use the existence of a valid
value range instead. This will also make it possible to
define arbitrary start and end-values for min and max as
long as min < max.
The correct thing would be to return the ClientPointer. However, if the client
for some reason has a core grab on another device (e.g. as result of a passive
grab), return the coordinates of the grabbed device instead.
This makes the use of nautilus a bit saner.
If the window being changed is set as the motion hint window for any device,
the device's motion hint window is set to NULL. Which is kinda what the old
code did, except that it did it with only the VCP.
Composite uses an unmap/map cycle to trigger backing pixmap allocation
and cliprect recomputation when a window is redirected or unredirected.
To avoid protocol visible side effects, map and unmap events are
disabled temporarily. However, when a window is unmapped it is also
removed from grabs and loses focus, but these state changes are not
disabled.
This change supresses the unmap side effects during the composite
unmap/map cycle and fixes this bug:
http://bugzilla.gnome.org/show_bug.cgi?id=488264
where compiz would cause gnome-screensaver to lose its grab when
compiz unredirects the fullscreen lock window.
This is a significant shift in how input events are perceived. The common
approach was to treat a core event as a different entity than the XI event.
This could result in the XI event being delivered to a different client than
the core event. This doesn't work nicely if they come from the same device.
Instead, we treat an input event as a single event, that is delivered through
two separate APIs. So when delivering an event, we first try the XI event,
then the core event. If the window want's neither, we go to the parent and
repeat. Once either core or XI has been delivered, the processing stops.
Important: Different to the previous method, if a client registers for core
button events, the parent window will not get XI events. This should only
cause problems when you're mixing core and XI events, so don't do that!
Generic events don't fit into this yet, they cause me headaches.
This should restore the correct passive grab processing. When checking for
passive grabs, the core event is emulated and we check first for XI grabs on
the window, then for core grabs. Regardless of which event activates the grab,
the XI event is stored in the device's EQ.
When replaying the event, we take the XI event and replay it on the next
window, again including the emulation of the core event.
If two devices are attached to the same master device, pressing button 1 on
each of them leads to two button presses from the same device. Some apps
really don't like that.
So we just put a counter in place and only send the first press and the last
release.
(first_valuator + num_valuators) must never be larger than the number of axes,
otherwise DIX freaks out. And from looking at libXI, anything larger than 6 is
wrong too.
(cherry picked from commit 9f6ae61ad1)
General code cleanup, whitespace, dead code removal, added missing prototypes.
Made Xquartz come to foreground later in startup, so it doesn't appear for Xquartz -version
(cherry picked from commit 36922e8ff4)
From the X11 protocol spec:
"If background None is specified, the window has no defined background."
This means that toolkits and apps cannot rely on the "transparent" nature
of the current implementation! At some point before the next release,
XACE will switch back to a solid background as the default.
(first_valuator + num_valuators) must never be larger than the number of axes,
otherwise DIX freaks out. And from looking at libXI, anything larger than 6 is
wrong too.
Initialise num_events to 1, so we always send a proximity event, and then
optionally valuator events. Also make sure mieq can deal with valuator
events sent after proximity events.
If we have one global filter, one pointer may change the filter value and
affect another pointer.
Reproduceable effect:
blackbox and xterm, start dragging xterm then click anywhere with the other
pointer (attached to different masterd device!). The button release resets
the filter[Motion_Filter(button)] value, thus stopping dragging and no event
is sent to the client anymore.
Having the filters set per device gets around this.
In some cases (e.g. using x2x) the previous model broke, with a window ending
not counting down to 0 anymore. Special treatment for NotifyUngrab seems to
help here.
Current solution: If a NotifyGrab is to be sent ignore it. If a NotifyUngrab
enter is sent, only adjust the semaphore if it is on 0. Likewise, do the same
for a NotifyUngrab leave if the semaphore is on 1. This seems to work alright
so far.
We free the ValuatorClassRec quite regularly. If a SIGIO is handled while
we're swapping device classes, we can bring the server down when we try to
access lastx/lasty of the master device.
XkbInitKeyboardDefviceStruct may call FatalError if it fails. FatalError then
cleans up all the devices, resulting in a segfault if the pointer is
uninitialised.
Old code was fundamentally broken, fixes now are:
- free the MDs current device classes
- copy the device classes instead of flipping the pointers
- check for the old MD, not the new one.
Anything in dev->key, dev->valuator etc. of a MD must always be a copy of the
original class. The intial classes of an MD (the ones set up before an SD is
attached) as well, as we may have to restore them if no SD is attached
anymore.
Turns out it's really really hard synchronising device state across multiple
duplicated events if they all share the same struct. So instead of doing so,
when the SD changes deep-copy all it's classes into the MD. The MD then has
the same capabilities, but the state can be set separately. This should fix
xkb, key state, repeat etc. problems.
Updating the device state allows us to remove the SwitchCoreKeyboard from the
event gathering, it's all done during event processing now.
If a slave device is attached to a master device, then we need to send a
mapping notify to the client.
Mapping notify needs to be sent if
- different slave device but on same master
- different master
This gives you funny behaviour with the ClientPointer. When a
MappingNotify is sent to the client, the client usually responds with a
GetKeyboardMapping. This will retrieve the ClientPointer's keyboard mapping,
regardless of which keyboard sent the last mapping notify request. So
depending on the CP setting, your keyboard may change layout in each app...
This can happen if we check for a passive core grab and our device is a
floating slave device. Doesn't really change anything as SDs can't send core
events but it stops the server from segfaulting.
Each time a different slave device sends through a master, an
DeviceClassesChangedEvent is enqueued. When this event is processed, all
classes of the matching master device are changed, and the event is sent to
the clients.
Next time the master is queried, it thus shows the evclasses of the last slave
device. The original classes are stored in the devPrivates.
TODO: if all slave devices are removed, the master's original classes need to
be restored.
For pointers: don't try to set master->valuator fields if there is no master.
For keyboards: check if device is valid before trying to access the fields in
miPointerGetScreen (btw. this disables DGA events for floating keyboards).
Also stop the hideous number of ErrorFs if we request the paired device for a
floating dev.
XkbRemoveResourceClient wants to access xkbInfo if it exists, so make
sure we NULL it after freeing it. It doesn't make much sense to move
the RemoveResourceClient call first, as there's not much point in
notifying clients while we're shutting the server down anyway.
Subject: [PATCH] fix some performance gaps in Xace
The XaceHook function is used in several hotpaths.
The problem with it (performance wise) is twofold:
* The XaceHook function has a big switch() statement for the hook number in it
* The XaceHook function uses varargs to reassemble the final dispatch arguments again
Both are expensive operations... for something that is known at compile time
This patch turns the hotpath XaceHook call into a direct call to avoid
the switch and varargs; this gives me over 10% performance gain
on the x11perf benchmark.
These hints allow an acceleration architecture to optimize allocation of certain
types of pixmaps, such as pixmaps that will serve as backing pixmaps for
redirected windows.
If we inherited a signal mask from the parent process that ignores SIGUSR1,
then we will send SIGUSR1 to the parent to indicate when we're ready to
accept connections. Unfortunately, we send this notification way too
early, right after creating the sockets rather than just before entering
the main loop.
Move it to just before Dispatch() so we're not lying quite so much.
FixKeyState needs to be able to handle XI events, otherwise we get "impossible
keyboard events" on server zaps and other special key combos.
(cherry picked from commit 5ee409794e)