A new version of Arcan has been tagged. There is no demo video or fancy screenshots this time around; those things will have to wait until updates come to the related projects (mainly Durden) in a few weeks. Most of the work that remains on the 0.5 series isn’t much to look at by itself – but is nevertheless conceptually and technically interesting.
Some generic highlights from the last ~8 months of work:
The ‘Prio’ side-project – described in more depth in the One night in Rio – Vacation photos from Plan9 post. Outside of its value as a security- research target, an hommage to ‘the elders’ and a reminder that there are many more “ways of the desktop” out there than the ‘popular by legacy and traction’ Win/OSX/… one. Prio also serves as a decent base for rapidly putting together highly customised environments.
A number of new supporting tools (src/tools) – There’s aloadimage for image loading that will serve as a testing and development tool for security work like sandbox hardening, quality improvement work in terms of performance and – more importantly HDR rendering and HDR- formats. There’s aclip for building CLI- and scriptable external clipboard managers. There’s shmmon for connection- inspection, monitoring and debugging, and there’s waybridge for wayland client support. Speaking on the Wayland support, it’s getting to a stage where things like gtk3-demo, weston-terminal etc. start showing up – but so far, it has unfortunately been a very unpleasant beast to work with, and with the current pacing, it will take, at least, another good month or two still until it’s really usable.
VR grunt work – One of the near-future prospects that interest me the most on both a personal and a professional level – is the one of getting rid of much of the unintuitive and overcomplicated cruft that browsers, UI toolkits and the “traditional” desktop puts between the individual and computing. “Desktop VR” as it has been presented so far is little more than low-definition “planes-in-space”. With the layering and divison of responsibility that Arcan brings to the table, much more interesting opportunities should arise. Putting flowery visions aside, the support that has been integrated right now is far from stellar, but the “heavy on experimentation, light on results” phase of figuring out how everything from device interfacing- to the scripting API- is supposed to work – is nearing its end. The first part in this is the tool vrbridge that provides device- control and aggregation (the current hardware reality of coding for VR is mostly a closed-source vendor-lock in mess where you won’t get access to the primitives without swearing allegiance to bulky full-engine APIs) which will be fleshed out during the coming few releases.
TUI – Covered in the (regrettably crap) blog post on Chasing the dream of a terminal-free CLI is the (soon to be) developer facing API that takes advantage of SHMIF features to try and provide a way for building text oriented / command-line interfaces that gets rid of the legacy baggage and limitations that stem from having the CLI shell to always work through terminal emulator protocols like VT100, instead making it talk with the display server directly. This will be accompanied with Lua bindings and a (bash/zsh/…)- like shell environment. The SensEYE sensors and translators will also be reworked to use this API.
Xarcan – Is a patched Xorg that interfaces with SHMIF and provides ‘X in a box’ like integration. For all the ‘beating up the old man’ that Xorg seem to get, the backend coding was neither more or nor less painful than Qemu or SDL proved to be. Normal use should be just fine, but dipping into glamor+accelerated graphics seem to be good at provoking graphics driver crashes, possibly since we bind the GL context to a render node rather than the card- node. See the README in the git repository for more details.
Platform refactoring – The main ‘low-level’ platform backend, egl-dri – has been extended with some basic switchable ‘synchronization strategies’ for dynamically changing scheduling priorities between energy efficiency, lower input latency, smoother animations and so on. The egl-nvidia code has been integrated with the egl-dri platform now that unified buffer project seems to have stalled. There are some caveats on activating and using it with the NVIDIA closed-source blobs, covered further in the wiki. Most GL use has been refactored to be dynamically loadable and reloadable, getting us much closer to multi-vendor-multi-GPU use and live driver upgrades.
LED subsystem rework – This was very recently covered in the Playing with LEDs post, so I will only mention this briefly, but the way LEDs like keyboard NumLock, CapsLock, ScrollLock, Display Backlights and more advanced “gamer keyboard” like setups – were all hooked up has been reworked and mostly moved to a tiny protocol over a pipe.
SHMIF improvements – The internal segmentation/IPC API has been extended to support negotiation for privileged features, such as access to display lookup tables and VR metadata (also, placeholder: HDR, Vector). Extended accelerated graphics (OpenGL etc.) has been split out into a shmifext- library so that all the other project backends use the same method for getting accelerated GPU access. This will primarily play a role when we need to respond to hotplug events or load balance between multiple GPUs. A new ‘preroll’ stage has been added to the connection process in order to provide an explicit synch-point for script- dependent initial metadata, which should cut down on connect-and-draw latency.
An interesting development that is being experimented with, but won’t be pushed for a while, is reusing the Lua API (most of the API was actually designed with this in mind) as an external protocol for moving the ‘appl’ layer out into its own process. This will be relevant for two reasons. The first one is to make it possible to control the engine using languages other than Lua. It also makes it possible to run things in an X11- like “separate window manager” division of responsibility in an ‘opt-in’ way. The second, more distant, reason is that it works to provide a render- API for network like transparency – though there are likely to be more hard-to-foresee problems lurking here when there’s added latency and round-trips.
A slightly more detailed changelog is available on the wiki.