FatalError. These direct calls to exit(1) made it impossible to do
anything ddx-specific in these cases; note that most of these calls
occur during argument processing.
been decided that the priority is to preserve the server's internal
API/ABI so that third-party drivers that depend on symbols like
noPanoramiXExtension, etc., would not need to be recompiled. Toobad gcc
on Linux doesn't support ELF's weak symbols as that would have been a
reasonable solution for preserving the ABI. N.B.: While symbols, i.e.
functions and variables revert to the old name, I did not revert build
names, i.e. -DXINERAMA, to the old -DPANORAMIX. There was no need, and
it's just a build issue that has no impact on the binary output of the
build.
generating Alt_L, etc, fix the implementation so that you get Meta_L
for the left Meta key and Meta_R for the right Meta key. Ditto for Alt,
Control, and Shift.
programs/Xserver/Xext/shm.c
3.37 and programs/Xserver/Xext/xvdisp.c 1.26 got zapped when Xinerama2 was
merged into the tree. (Xinerama has since been reverted to 1.1, but
that's another story.)
decided that the priority is to preserve the server's internal API/ABI
so that third-party drivers that depend on symbols like
noPanoramiXExtension, etc., would not need to be recompiled. Toobad gcc
on Linux doesn't support ELF's weak symbols as that would have been a
reasonable solution for preserving the ABI. N.B.: While symbols, i.e.
functions and variables revert to the old name, I did not revert build
names, i.e. -DXINERAMA, to the old -DPANORAMIX. There was no need, and
it's just a build issue that has no impact on the binary output of the
build.
been decided that the priority is to preserve the server's internal
API/ABI so that third-party drivers that depend on symbols like
noPanoramiXExtension, etc., would not need to be recompiled. Too bad
gcc on Linux doesn't support ELF's weak symbols as that would have been
a reasonable solution for preserving the ABI. N.B.: While symbols, i.e.
functions and variables revert to the old name, I did not revert build
names, i.e. -DXINERAMA, to the old -DPANORAMIX. There was no need, and
it's just a build issue that has no impact on the binary output of the
build.
Lanigan's xinerama tree on Sourceforge.Net. No attempt has been made to
handle previous, non-standard versions of the protocol. Nor has any
attempt been made to preserve the ABI of previous versions -- that part
will be added at a later time, and then probably only on systems that
have nice object/linker semantics, e.g. ELF systems with weak symbols.
have redirected subwindows in manual mode. Those clients are marked
Critical and given a significant scheduling boost whenever they receive
a damage notify event. This dramatically improves update frequency.
If the kernel reported a large number of keys, readKernelMapping would walk
off the end of the kdKeysym array.
Fix usage of _IOWR; the 'size' argument is actually a datatype.
drawable pixmap rather than directly into the frame buffer. Rewrite
logic in kdoffscreen to make space for new allocations, now deals
correctly with locked areas.
the server seems to act as before. With RedirectSubwindows (root,
automatic), the server looks just like a regular X server. Now to go
rewrite the (currently lame) compositing manager to get some real
action on the screen.
Some of the fixes here are to make valgrind quiet with various ioctls used
by kdrive/linux.
Also fixed a bug where fbdev initialization was out of order in fbdev.c and
smi.c
Alan Coopersmith
Harold L Hunt II
Kaleb Keithley
Egbert Eich
Keith Packard
Eric Anholt
Phil Blundell
Matthew Allum
Anders Carlsson