Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
..
Failed to load latest commit information.
wlimpl
CMakeLists.txt
README.md
arcan-wayland.1
boilerplate.c
bondage.c
shmifevmap.c
structs.h
syscalls.c
waybridge.c

README.md

Introduction

This tool bridges wayland connections with an arcan connection point. It is currently in a beta- state and should be useful, but expecting quirks and side-effects to exist.

Principal Articles

Crash Resilient Wayland Compositing

Building/Use

The build needs access to the arcan-shmif and arcan-shmifext libraries, either detected through the normal pkgconfig or by explicitly pointing cmake to the -DARCAN_SOURCE_DIR=/absolute/path/to/arcan/src

The appl- that arcan is running also needs to expose an approriate connection point (some, like durden and prio expose this as an ARCAN_CONNPATH env in instances of the terminal emulator) and you need to set the appropriate XDG_RUNTIME_DIR for both Waybridge and for any Wayland clients you wish to connect to the point.

     mkdir build
     cd build
     cmake ../
     make
     XDG_RUNTIME_DIR=/tmp arcan-wayland

The build-scripts will download additional protocol specifications from the wayland-protocols repository the first time the build is setup. If there's no internet connection at that time, build will fail.

Hacking

The code has a somewhat odd structure:

waybridge.c - setup and allocation/routing
boilerplate.c - structures and tables
shmifevmap.c - translate from arcan -> bridge/client/surfaces
wlimpl/* - subprotocol function implementations

Some subprotocol implementations (those that call request_surface) also attach separate event handlers that override the mapping done in shmifevmap.c. See also the part in the CMakeLists.txt that takes unstable protocols and generate implementation files.

The main paths to follow to get some kind of comprehension of what is going on is how surfaces are allocated and how the surface allocation request are propagated.

To debug, add the -trace level argument and a call trace to stderr will be added that cover (based on level argument bitmask, so add the values):

1   - allocations only
2   - digital input events
4   - analog input events
8   - shell (wl\_shell, xdg\_shell, ...) events
16  - region events
32  - data-device events
64  - seat events
128 - surface events
	256 - drm (accelerated buffers)
	512 - alerts

Allocation outputs are encoded as follows:

'C' - client-bridge connection
'S' - shell surface
's' - shell subsurface
'm' - pointer surface
't' - xdg- toplevel surface
'p' - xdg- popup surface
'x' - dead
'o' - unused

Then the correlated arcan events can be traced with ARCAN_SHMIF_DEBUG=1

All individual protocols can be disabled by the -no-(protocol alias) switch in order to control client behavior as this may vary wildly from client to client depending on the set of requested vs. available protocols.

Notes and Issues

  1. Stride - The shm- buffer blit doesn't take stride differences into account. This fails on MPV in SHM depending on source video size, though the other buffer transfer modes and the shm-to-gl approach works.

  2. Initial Control connection - this one should be deprecated and removed in favor of getting it from the client connection when it is created, the current form is a remnant from before the -exec refactor.

  3. Dynamic resizes are a bit edgy due to how differently this stage is handled in arcan and the 'idea' in Wayland. This will be fixed when the 'conductor' refactoring in arcan finishes.

Depending on what toolkit is being used, chances are that some magic dance is needed in order to get a client to connect using wayland, and similarly for the specific shell protocol to use when multiple are available, for the choice of buffer passing (shm or egl) and for the GPU device actually being used. With other Wayland compositors, chances are that this effect is masked by the client in question falling back to XWayland and some clients will fail outright if there's no XServer available even though the clients themselves actually only use their respective Wayland paths. In short, it's a mess.

a. For SDL, check the environment variable for SDL_VIDEODRIVER=wayland b. For QT, try the -platform wayland argument. c. For EFL, it's EVAS_ENGINE=wayland_shm or EVAS_ENGINE=wayland_egl along with ELM_DISPLAY=wl and optionally ELM_ACCEL=none if there's issue with gl acceleration.

The -exec mode sets these automatically.

Limitations

There are a number of arcan features that do not have a corresponding Wayland translation, and will therefore be a no-op or hidden in whatever scripts arcan is running. There are also some features where the translation is not entirely compatible. Such incompatibilities/limitations are tracked separately in the arcan wiki.

XWayland

XWayland support is currently omitted for two reasons. One is that the Xarcan approach is more efficient and blends better with the current arcan WMs (durden/prio) due to the whole 'self contained' thing, along with the need to have sharing-interception and not be explicitly dependent on wayland for X support.

The other is that for XWayland support, one must write an entire window manager (or import the one from weston or wlc, neither work flawlessly). Based on the 'Why isn't Xwayland just a Wayland Client' post (wayland-devel, 2017-sept.) it seems like there's a chance we get patches that makes the special xwayland path unecessary and thus it is better to wait and see.

BUGS

  1. gnome-apps, mouse motion registers but not button presses on some popups, likely related to input regions and grabbing and the WM script-side being guilty.
  2. SDL2, buffer- size and mouse cursor alignment is off in ex. 0ad
  3. gnome-apps, visible with calculator - some interaction between durden and subsurfaces etc. still seem to misbehave.
  4. The defunct wl- shell protocol is still in use in places like SDL2, except the protocol lacks options for communicating when a shutdown is initiated server-side, or it goes into a spinlock EXCEPT if some Qt specific protocols are present

TODO

Rough estimate of planned changes and order:

  1. fork- mode
  2. input fixes (regions, D to A mouse wheel, ...)
  3. data-device to clipboard
  4. enforce stronger error handling (not allow surfaces to switch roles etc.)
  5. move more 'decisions' that is part of the durden atype handling
  6. make a simple automatable wayland-only WM appl

(x - done, p - partial/possible-tuning, s - showstopper, i - ignored for now)

  • (for-each finished milestone, verify allocation/deallocation/validation)

The many 'p's are simply that we lack a decent conformance suite to actually determine if we are compliant or not, because Wayland.

  • Milestone 1, basics

    • Boilerplate-a-plenty
    • 1:1 client to bridge mapping
      • *:1 client to bridge mapping
    • Seat
      • [p] Keyboard
      • [p] Mouse
      • Touch
    • Mouse Cursor
    • [p] Popup
    • [p] EGL/drm
  • Milestone 2

    • [p] Cut and Paste (full 'data device manager')
    • [p] Subsurfaces
    • [p] XDG-shell
      • Focus, buffers, cursors, sizing ...
      • [p] Forward shell events that can't be handled with shmif
      • [p] Positioners
      • [p] zxdg-v6 to xdg-shell mapping
    • [i] Application-test suite and automated tests (SDL, QT, GTK, ...), seems that canonical attempts to tackle this
    • [i] XWayland (WM parts)
  • Milestone 3, funky things

    • SHM to GL texture mapping
      • Single-exec launch mode (./arcan-wayland -exec gtk3-demo)
    • [p] Transforms (Rotations/Scaling - most is done server side already)
    • [p] Multithread/multiprocess client processing
      • Live icon discovery / mapping
    • Pulseaudio stripping
      • Dbus stripping
    • Dynamic Keyboard Translation table generation
    • Benchmarking/Inspection tools
    • [p] Sandboxing
    • Migration/Reset/Crash-Recover
    • Drag and Drop (cursor states)
  • Misc. protocols:

    • Idle Inhibit Unstable
    • Pointer Constraints
    • Pointer Gestures
    • Input Method
    • Keyboard Shortcuts
    • Tablet
    • Presentation Time
    • Viewporter
    • Xdg-output
    • Xdg-foreign
      • Dma- buf
      • Qt- specific protocols for SDL shutdown issue