This partly reverts commit 41e76d11dff6dd9bc08bf829751f9634f32cdd38. I
removed the "superfluous" errno-handling, which in fact is needed. Turns
out pathconf() might leave errno unchanged.
Reported-by: Lubomir Rintel <lkundrak@v3.sk>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
It can return -1 (feature not supported, denied by a security module, etc.),
resulting in wrong allocation later on.
Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
(remove superfluous errno-checks)
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Lets be safe here and activate the context before performing the
buffer-swap. Drivers might be inconsistent and require this in case the
background process changed the DRM state.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We need to call drmModeSetCrtc during wakeup to prepare for page-flips. An
immediate modeset is needed, otherwise we cannot be sure the FB/Crtc
configuration is still the same.
Force an immediate DRM modeset on wakeup to reset the screen.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Be more verbose during wake-up and set VIDEO_HOTPLUG so we force the
hotplug-check. I don't know why it worked until now, but we definitely
need to force it as we have no idea what display-status changed.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Instead of letting xkbcommon write to stderr, we now forward these
messages via the llog handler.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If the format string contains a trailing newline, we now skip writing our
own. We also skip the fn+file information as it would be written on the
next line.
Trailing newlines occur only when forwarding messages from other
libraries. In this case we don't need the fn+file information, anyway.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We should use library-logging in uterm to avoid cluttering stderr for
applications. Hence, use llog instead of log directly.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
This adds a small US-keymap to the uterm library which is loaded if we
cannot find a suitable system default keymap. This allows using the keymap
if no XKB keymaps are installed.
Thanks to Ran Benita for the minimal US keymap.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Instead of generating C source files, we now use our binary-linker make
target which can link in any arbitrary binary file.
This way, we can delete genshader and instead link the files directly.
This speeds up compilation and makes the code more consistent.
We also strip the shaders from useless comments and whitespaces to reduce
memory consumption.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Shader sources may not be 0 terminated if we mmap() them. Hence, we need
to pass the length to the shader compiler. As glShaderSource() allows this
<0 as zero-terminated strings, we can simply add a parameter for the
shader length.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
No need to keep these static helpers outside of SHL. Instead, include them
in SHL to avoid any external linking.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
The freetype2 font backend lacks support for proper combining marks and I
do not intend to ever implement that. Use pango!
If you don't want heavy dependencies, you can use the unifont or 8x16
backends.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
TSM was extracted from kmscon sources so it can more easily be used by
other emulators. It is available at:
http://cgit.freedesktop.org/~dvdhrm/libtsm
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
cdev sessions are outdated. Use libuvt instead. Remove all references to
cdev-sessions and clean up the build chain.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Remove uvtd. The idea is outdated and not really needed. With recent
systemd-logind changes, all we need is a shim between legacy-programs
(like XServer) and logind. We can easily do that via libuvt without
requiring a huge daemon like uvtd.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Remove all wlterm sources. I never intended to maintain it longer than any
major terminal-emulator needs to get ported over to wayland. Hence, remove
it. It was buggy, anyway.
If anyone is interested, a GTK+ based wlterm is available on:
http://cgit.freedesktop.org/~dvdhrm/wlterm
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If we get a display-refresh event, redraw the screen. Otherwise, we might
stay blank during VT switches if we get a delayed set-master.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If we get a display-refresh event from the session layer, we must redraw
the screen to avoid staying blank.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
On UTERM_REFRESH events we now forward the event to all sessions on the
bound seat. Sessions can then react to it and repaint the screen.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If drm-wakeup fails, we now retry after 20ms continously to gain
DRM-Master again. If we succeed we send UTERM_REFRESH events for all
available displays so existing code can pick it up.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We only use uint8_t as length-counter so MAX_DATA_SIZE must be <256 to
make our length-checks work. This fixes a bug where we read invalid data
if someone uses modified genunifont data files. As this just produces
garbled glyphs if garbled data is given, this is actually not critical at
all.
Reported-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We need to use signed types to actually see errors from ftell(). Fix it by
using "long" for status bits now.
Reported-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Instead of passing display-ops to every helper store it in vdrm-objects.
During vdrm-setup we set the pointer and reuse it in all helpers. This
simplifies the function headers and allows calling them from
helper-callbacks without requiring a pointer from the respective user.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
wlterm doesn't work with current Wayland/Weston; the events are never
sent and the window doesn't come up. We need to flush the display, as
per wl_display_dispatch_pending(3):
To proper integrate the wayland display fd into a main loop, the
client should always call wl_display_dispatch_pending() and then
wl_display_flush() prior to going back to sleep.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
With mesa-9.2 fragment shaders are required to define default precisions
as defined in the standard. We didn't do this.. whoops. Add the mediump
declarations and everything should be working again.
fixes fdo bug: #68934
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Khronos updated their GLES2 headers and broke backwards compatibility,
hurray. Fix it for real now.
Reported-by: Yichao Yu <yyc1992@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Khronos "fixed" their new OpenGLES2 headers and added a bunch of _EXT
suffixes. Now we need to test for both, the old and new macros, yay!
Thanks to Yichao Yu for spotting that.
Reported-by: Yichao Yu <yyc1992@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We need to correctly set the VTNR for each new pty, otherwise the pty
cannot set the XDG_VTNR correctly. Note that we do this only for real VTs,
that is, vtnr > 0.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
A caller can now pass a VT-num to the PTY which will get set as XDG_VTNR
in the environment of new childs.
This can be used to tell systemd-logind to associate the session with
the correct VT. But note that this still cannot overwrite VT-associations
if the pty is created from within an existing session.
Reported-by: Thomas Hebb
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
This helper returns the VT-number for the given VT if, and only if, it
is a real VT. In all other cases 0 is returned.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Wayland compositors are not required to send buffer release notifications
on an implicit buffer-attach. That is, an attach->commit->commit series
does not cause a buffer-release notification after the second commit.
Hence, send a buffer-attach on every commit that requires a new
buffer-release notification.
For the related wayland-protocol changes, see:
http://cgit.freedesktop.org/wayland/wayland/commit/?id=7165bf21917ed1c4f5ec823a1e1642a9e83aa984
This fixes a bug where wlterm only redraws the window-content when
resizing (due to explicit buffer-attach calls during redraw).
Reported by: MoD
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
This adds an --xkb-keymap <FILE> option to kmscon. When given, kmscon
will try to compile the keymap from the file before trying the other
options (like the XkbKeymap option in xorg.conf).
This is useful for users who have a customized XKB keymap, which is
usually kept in a single file. Example, in X:
xkbcomp $DISPLAY my_keymap.xkb
Customize my_keymap.xkb to your liking, and then in .xinitrc, or
xorg.conf:
xkbcomp my_keymap.xkb $DISPLAY
Now you can also do this in kmscon.conf.
Additionally, lacking such an option, kmscon is quite difficult to use
without an installed xkeyboard-config package, which provides the
infrastructure for the "rules" configuration mechanism. We might even
want to distribute some plain xkb file as a last ditch, for
robustness, if even the default RMLVO fails... without a keyboard a
terminal is not very useful.
(changed Ran's patch to use *_from_string() instead of *_from_file())
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
This reads in a complete file as a string and returns it to the caller.
The string is 0 terminated (which isn't guaranteed by mmap()) so this
helper is needed if we have to work with APIs that don't accept buffer
lengths.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
This implements two new keyboard shortcuts zoom-in and zoom-out that
increase/decrease font size of the current terminal.
This is similar to how wlterm does it and allows runtime modification of
fonts.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
If the terminal screen is smaller than the real screen, we never paint to
the margins. This doesn't hurt as long as we never resize the terminal.
The uterm layer clears all framebuffers during allocation.
However, uterm behavior may change and our terminal may get resized (eg.,
during hotplugging) so we really should clear all the margins.
We now clear them on every frame as it is a trivial task. However, if we
speed up rendering, we should probably set a "needs_clear" flag that
simply clears the framebuffer.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We currently retrieve one font object per screen (which is in fact shared
in the font-layer), but we can also move it one layer up to the terminal
layer. This shares the same font per terminal between all screens.
This simplifies on-the-fly font modifications and speeds up monitor
hotplugging.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
pango_ft2_render_layout_line() requires the baseline offset as argument,
not the vertical extent of the glyphs. This is important as we currently
align glyphs to the bottom edge instead of the baseline.
Fix this by passing the cached baseline offset so all glyphs are correctly
aligned to the baseline. We then clip according to the cell-extents as
usual.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
It is currently pretty annoying to use multiple wlterm windows stacked on
top on each other when the terminal-background is black. The border
doesn't use multiple colors so it is hard to distinguish from the
main-frame.
This patch changes the border color to white (as most terminal-backgrounds
are black by default) and additionally draws a black 1px frame around it.
This guarantees that the frame is even visible with white terminal
backgrounds.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
We now keep a link to our parent seat and set it to NULL when our session
is unregistered. In this case, any further request that depends on the
session being registered and probably a valid seat pointer, we will stop
with ENODEV.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
These ioctls are used to retrieve and set the VT-mode. That is, the
signals and PID information that correspond to the VT controlling process.
It is notified whenever the VT gets active/inactive so it can react to it
or prevent/acknowledge it.
This doesn't implement the signal-sending logic, yet. It only implements
state-tracking.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Session IDs provide a unique ID for each registered session. They are used
to sort the sessions so next/prev return the correct neighbour-sessions.
Furthermore, they can be used by outside users to refer to a session
directly without knowing the session implementation.
Sessions with ID=0 are always linked at the end and considered to be
"unnamed".
The algorithm to switch to a session with a given ID is:
Try to find the first session with the exact same ID. If there is none,
return the session with the next higher ID. If there is none, return the
ID'th session in the list. If there is none, return the last session.
This provides a fairly predictable way of switching between session. It is
modeled after the classic VT F1-F12 keys that switch between sessions. If
a session is not given, these keys will switch to the next higher session
instead.
All "unnamed" sessions are put at the end. So if you have only F1-F4, then
F5-F12 will map to unnamed sessions.
Please note that new sessions are always linked at the end of their group.
So new unnamed sessions are at the far end, new named sessions are linked
in the sorted list but behind all sessions with the same ID.
Hence, the caller should avoid multiple sessions with the same ID,
otherwise, some sessions might not be reachable even though they're named
(unless you use next/prev of course).
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
