This article introduces the first release of ‘Lash#Cat9’, a different kind of command-line shell.
A big change is that it is communicating with the display server directly, instead of being restricted and filtered by a terminal emulator. The source code repository with instructions for running it yourself can be found here: https://github.com/letoram/cat9. A concatenation of all the clips here can be found in this (youtube-link).
Cat9 serves as the practical complement to the article on ‘The day of a new command-line interface: shell‘. That article also covers the design/architectural considerations on a system level, as well as more generic advancements to displacing the terminal emulator.
The rest of the article will work through the major features and how they came about.
A guiding principle is the role of the textual shell as a frontend instead of a clunky programming environment. The shell presents a user-facing, interactive interface to make other complex tools more approachable or to glue them together into a more advanced weapon. Cat9 is entirely written in Lua, so scripting in it is a given, but also relatively uninteresting as a feature — there are better languages around for systems programming, and better UI paradigms for automating work flows.
Another is that of delegation – textual shells naturally evolved without assuming a graphical one being present. That is rarely the case today, yet the language for sharing between the two is unrefined, crude and fragile. The graphical shell is infinitely more capable of decorating and managing windows, animating transitions, routing inputs and tuning pixels for specific displays. It should naturally be in charge of such actions.
Another is to make experience self documenting – that the emergent patterns on how your use of command line processing gets extracted and remembered in a form where re-use becomes natural. Primitive forms of this are completions from command history and aliases, but there is much more to be done here.
I collected history from a few weeks of regular terminal use along with screen recordings of the desktop window management side. I then proceeded to manually sift through these, looking for signs of poor posture. I found plenty.
This is a humbling experience. The main conclusion drawn is that I am mostly a hapless twit who default to repeating the same things hoping for different outcomes. I consistently confuse ‘src’ and ‘dst’ for ‘ln -s’; ‘ls’ gets spelled ‘sl’ much too often; ifconfig remains the preferred choice to ‘ip’ even though its main output typically is ‘file not found’ these days; nearly every tool that expects regular expressions are first fed plaintext strings. When I actually want to use a regular expression I consistently pick the wrong expression language.
The signal to noise ratio in the history is abysmal. About 90% of scrollback contents were leftovers from cd, ls and tab completion sprinkled with repeated runs of the same command through sudo, with minor tweaks to the arguments or to get a redirection for stderr. Redirections that were then left in the file system, with descriptive names like “boogeraids2000”.
The screen recordings were also revealing. Some notable time sinks:
Copy paste across line-feeds and resizing windows to deal with incorrect wrapping.
Spinning up new terminals to work around man or vim hogging the alt screen.
Digging around in ps/proc/… for PIDs.
Redirecting to temporary files to transfer job outputs between windows or for later comparison.
Switching vim buffers between horizontal/vertical to fight the tiling WM.
All these can be fixed with relatively minor effort.
Get the prompt out of the way.
Starting with the prompt – obvious bits are that its contents should be ephemeral and disappear after running a command. It should reflect information about the current context (directory, etc.) and whatever else of immediate short lived value. The point is to clean this up:
Instead we get this:
Prompt is updated live regardless of input and can change its layout template dynamically.
Prompt format and contents depends on window management state (focus, unfocus).
Silent commands are kept away from the history.
Completions come up without interaction and do not trample/shuffle actual contents.
Commands that only resulted in errors are automatically delay purged.
The previous options for compartmentation was a choice between juggling between a ‘foreground’ job and ‘background’ jobs. For this to work you needed either a fragile weave of signalling (SIGSTP, …) and file redirections — or spin up new terminals, either through a terminal multiplexer (a terminal emulator inside a terminal emulator inside ..) or new windows.
I find those solutions both noisy and distracting. Instead, I now have this:
Every command-line submitted now becomes its own job.
Jobs can reference each other.
Job context (environment variables, working directory, …) is saved and tracked.
The jobs are presented in order of importance (active ones take priority over passive ones).
Spawning new jobs automatically folds old ones into a collapsed form.
Individual controls, status and statistics are added to a stateful bar at the top of the job.
Job contexts can be reused for new commands.
Remember everything, but right to be forgotten.
In the terminal world, all job outputs either get composed unto one shared buffer with a certain amount of memory (scrollback history); fight for a scratchpad (“altscreen mode”) or are redirected to files or other jobs. This happens regardless of stream source or job state (foreground/background).
With real compartmentation and much larger memory and CPU budgets thanks to server side text rendering, we can do much better:
Stdout and Stderr are tracked separately.
All job output is kept, tracked and addressed individually.
Contents can be forgotten, or selectively processed.
Completed jobs can be repeated, appending to the existing output or replacing that of previous runs.
Jobs can be repeated with an edited command-line.
Cooperate with the outer windowing system
Now that the shell can talk directly to the window manager without having the conversation dumbed down by a terminal emulator sitting in between, new integration options are possible:
Snapshot the output of a job to a new window.
Window creation hints to window manager, like vertical split or tabbed.
Open applications and media embedded, with controls for position and size.
Detach and reattach embedded media, preserving input routing.
Directly route contents to clipboard and other data sharing mechanisms.
Trigger GUI file pickers.
Let legacy in
Now with a fairly functional environment, the last part is to account for all the edge cases where we still need access to the old world in various degrees:
Send data from a job to external processing pipes (#0 | grep hi).
Request a new window, attach a terminal emulator to it and run a pty dependent command (!vim).
Setup a PTY and attach a VTxxx view to it: (p! ls –color=yes).
Streamline command structure
The foundation to cat9 is the command-line language itself. All the UI elements that you see, mouse gestures and key bindings map to the same things that you could type in manually:
Hooks and event actions can be added after a command has been setup or is running.
Mouse actions, bindings (clicking shown in clip: view #csel $=crow as in ‘cursor job, cursor row’).
Aliases and pre-commit expansion.
With these basics sorted out, it is time to build something more interesting.
Special Topic: Views on Life
Now that jobs keep their data around in nicely tracked structures rather than a prematurely composed and broken ‘scrollback buffer’, we can do something more. While we have data in its raw form, we can look at it through various lenses to get different representations of the data. These are baked into the ‘view’ builtin.
Simply put, they parse the data and reformat the contents by adding annotations, structures, formatting and so on. The current builtin ones are all shown in this clip:
In this one you see ‘wrap’ and ‘filter’ along with some options like line numbers and column wrapping. Filter even goes so far as to have an interactive mode that live-applies the filter as it is being written.
With the original data retained, re-executing previous pipelines is not needed, and the choice between using the formatted output and the original data is available when copying in/out.
This is one of the features that will be expanded heavily in future versions as we try to improve the presentation of the many ad-hoc text formats.
Special Topic: State Actor
This is a good one. Regular windowing systems provide Clipboard as well as Drag and Drop as forms of interactive data sharing. Some go further and also allow sequenced picking/sharing, like the “share” button popular in mobile operating systems. Arcan adds a state store/restore action to the mix.
This means that at any point, the windowing system can request that a state snapshot is created, or request that the application reverts to a provided one.
Examples of what gets stored in such a state blob here are configuration changes; command history; environment variables; aliases and so on. While this offloads the ‘where are my dot files’ responsibility, more interesting is that states can be transferred between instances at runtime.
Combine this with the job system: by marking a job as persistent, the command creating a job will be added to the state store. In the following clip you can see it being used to an interesting effect:
I first start a new cat9 session, run two jobs and mark one as persistent manual and other as automatic. Shutting down and restarting and you can see how the jobs come back, with the automatic one starting immediately. In the next clip I go one step further and copy the state between two live instances.
When combined with remote shells, this becomes a really potent administration and automation tool. Perform a task once; visually confirm that the results matched expectations; Save the state and replay wherever and whenever. Use that for knowledge sharing, or hook it up to an event source for snapshotting and rollback to give anything history/undo.
Special Topic: Frontending
There is little consistency between many popular tools, no matter if they come as “argv hell”, “CLIs within the CLI” or “lots of small binaries”. This is natural, but also undesired from a user perspective. It feels rather futile to have gone through the strides of building a CLI that behaves like you want it to — just to have the work be undone by the tools you launch from it.
I am no stranger to uphill battles, but the odds of getting the likes of wpa_supplicant, git, gdb/lldb or ffmpeg to change their evil ways and follow the one true path are slim to none. The passive aggressive form of dealing with this is what bash_completion and the likes do – create helper scripts that at least make polite suggestions while building the command line. This works poorly when the tool is interactive. Other options include defining better programmable interfaces, language server style external oracles, then hope for the main drivers to convert.
With the extensive scripting, parsing and rendering options available to us now – there is a more actively aggressive way. In Cat9, you can define multiple sets of builtins and views, and switch between them. This means that you can create a set of builtins for a specific logical function, like networking, programming or debugging, then swap between those as needed.
This, along with views, will be the more active area being developed for future releases. The following short clip shows an early ‘in progress’ such set for networking.
In the clip you can see the set of builtins being swapped to ‘networking’ which new builtins such as ‘wifi’. You can see the live completion of available SSIDs appearing asynchronously as a scan is complete. Commands can still be forwarded ‘raw’ with the output packaged into its own job that can be used by the other builtins. It can also attach polling status about signal levels and connection into the prompt, using all the same infrastructure as the previous demonstrations.
I hope this conveyed some of the benefits of leaving the shackles of terminal emulators and its more abstract form of ‘virtualisation for compatibility through emulation as default’ behind. There are a whole lot more ideas to squeeze into this setup now that all the grunt work has been dealt with.
Better CLIs as part of better TUIs are key for making professional computing more accessible to budding sprout experts and cognitively challenged alike. The building blocks are here for your ‘speech- assisted’ command-lines without having to have a screen reader try and make sense of a poorly segmented word soup, or for your red team approved secret “leave no trace” cleanup sauce.
The last article in this series will dip into the programmable surface – how the APIs replacing curses work and integrate with the display server / window manager.
This release should put us at about half way through the planned work for the networking focus set of releases (0.6.x), a scope roughly defined by the article on A12: Advancing network transparency on the desktop and the one on Arcan as OS design. Alas, it is also the single most difficult and time consuming part left on the entire roadmap.
Before dipping into the major additions and changes, I will break form a little and dwell on what is going on and why.
From the (set of design principles) that we follow; number four “Make State mobile“, five “No State left behind” and six “Privacy fights back” are at the center of attention here.
The idea is to get a protocol which replaces mDNS (local service discovery), SSH (interactive textual shell), X11/VNC/RDP (interactive graphical shell), RTSP (streaming multimedia), HTTP (networked application retrieval and state synchronisation) and a few other lesser knowns, and we are nearly there feature wise.
This is “less” effort than one might think as so much code is needlessly repeated again and again by not leveraging the many bits the designs of these protocols have in common, justified only by legacy and history and not technical and architectural merit.
This is similar to the IPC situation locally that we solve with SHMIF — while others will get to continue to enjoy the cacophony of IPC systems (D-Bus, Wayland, Pipewire, VTxxx, …) where the difficult parts (authentication, discovery, synchronisation, least-privilege separation, zero-copy ownership transfers, queue and resource management, resilience, …) keep on being implemented again and again and again in incompatible ways yet are supposed to work together to solve actual end-user problems — this is the price to pay for being stuck in ‘first order’ forms of reasoning, but simplicity is systemic.
With this protocol as a building block, every single component in Arcan can be de-coupled from one device and re-coupled to running on another – from media parsing and decoding to accelerated rendering and encoding.
To illustrate the point and the self-imposed “grand challenge” — in this photo from one of my labs are the set of user facing devices in the weekly rotation currently capable of running Arcan; each with some quirk or property that makes it interesting to keep in rotation (this is also the least depressing lab, wait until you see the one for displays or the one for input devices).
Together they represent a sort of lighter extreme here:
The devices that are active should be able to share workload and work ‘ as one ‘.
Repurposing any device to a ‘one ephemeral task’ runner should be achievable within minutes, and a queue of prepared runners should make activation near instant.
Installed static state (what it can do) should be known, dynamic state (what you changed) should be extractable.
These tactics should serve to raise the cost for both reliable persistent exploitation and for evading detection considerably. They should also work well for building intuitive and ergonomic compartmentation for harm reduction against both ‘smash and grab’ style attacks, micro-architectural side-channels and against physical theft.
All this across networking infrastructure that is assumed to be unreliable (no global clock), peer-to-peer (no DNS by default) and only accidentally connected to the Internet — what some would call air gapped.
As a refresher, the initial proof of concept – rougly the state in ~2019 – can be seen in this clip:
There will be reason to return to the topic in the near future, but for now, let’s move to the big ticket items for this release. For the both more- and less- detailed list of changes, see the regular Changelog.
Onward to the big ticket items:
The two main tools for using our network protocol are ‘arcan-net’ (standalone) and ‘afsrv_net’ (that transparently maps to script-reachable functions in our Lua scripting layer).
arcan-net now has the first draft implementation of ‘directory’ mode, which will be used for three purposes; as a discovery rendezvous in WANs where other communication might also be needed (proxying or NAT punching), as a trusted third party state store, and as an arcan appl host.
This part of directory mode covers the arcan appl host setup. It lets any arcan installation share the set of appls it has with any other, and act as a state store (configuration persistence).
There are articles in the queue about the implications of this but as an example out of many — it means I can have an offline ‘build box’ that generates device tailored ‘live’ images (e.g. the hacky scripts in arcan-void-mklive for now); injecting authentication keys into the image and whatever device boots from it can load/restore the same persistent desktop from an otherwise ephemeral read-only environment. The image is logged and attested, and can act as source for comparison against the device at a later date.
For the sake of it, arcan.divergent-desktop.org is currently hosting ‘durden’ and ‘pipeworld‘; subject to me breaking things during daily experimentation. It was started like this:
(the –soft-auth makes it about as insecure as a world of self-signed https, it is unwise to run appls from this server on anything sensitive). In this clip you simply see me connecting to it over a fairly slow link:
The state store bits are less promiscuous and still require an authenticated key exchange. Let’s write a simple arcan appl called ‘demo’:
local counter = get_key("counter") or "A"
local img = render_text("Hi " .. tostring(counter))
tag_image_transform(img, MASK_OPACITY, function() shutdown() end)
store_key("counter", counter .. string.char(string.byte("A") + #counter))
end' > demo/demo.lua
The following clip is the result from me running this appl a few times:
You need to squint a bit, but for each run another letter is attached, walking from ‘A’ and onwards. The point is that this state is managed server-side, had the same public key (== identity) been running on another device, it would have continued where the other last synched :- remote, persistent, observable.
arcan-net now caches some binary transfers and ramp ups transfers after confirming something is not already in the cache. This is most useful for synching fonts. It also automatically compresses / decompresses binary file transfers.
afsrv_net has gotten a ‘sweep’ discovery mode, which enumerates the ‘petnames’ in the keystore and periodically tracks which ones that have started- or stopped- responding.
The protocol now covers role negotiation as part of the initial handshake, where each side can be either Source, Sink or Directory. Pair Source-Source or Sink-Sink will disconnect with an error.
Finally, the protocol itself has added role negotiation as part of the initial handshake. Since both sides can act as either source, sink or directory – we now pair that so trying to connect a source to a source or a sink to a sink would generate error notification.
Our ncurses replacement, arcan-tui, has received a number of fixes in its basic widgets, most notably in its ‘readline’ implementation. The corresponding Lua bindings have also received a lot of attention, and are now suitable for writing most kinds of CLI/TUI applications.
The ‘terminal’ frameserver (afsrv_terminal) that has long acted as our terminal emulator of choice, now has an additional mode of operation (ARCAN_ARG=cli=lua) that we call ‘Lash’ (LuA SHell). This pulls in a Lua VM along with the TUI API bindings, coupled to a simple chainloader script that pulls in a custom shell ruleset from (default) $HOME/.arcan/lash.
The following clip demonstrates some capabilities of a shell written on top of this – “Cat9”.
The reasoning behind all this is covered in a separate article, ‘The Day of a new Command Line Interface: Shell‘ while Cat9 will be getting a more thorough introduction a little later when I am satisfied with its feature set and implementation quality.
A large target for the API/bindings is to quickly build / wrap system services (network device control, file system mounting, …) as frontends to existing tools (wpa_supplicant et al.) that integrate and compose similar to how we illustrated with the tray icon handler.
Arcan splits up its resource handling in a number of static namespaces. The rules behind many of these are quite complex, and really only makes sense when considered as a design to avoid some of the many mistakes in Android and dates back to the old days (2.x+) of that OS, and as preparation for sandboxing resource access in the context of loading appls as shown in the networking section.
In this patch, the configuration database managed through the arcan_db tool can be used to define dynamic user namespaces. A simple example would be:
Which would allow Arcan instances read/write access to /home/me with the symbolic name “myhome” and the user presentable label as Home. This fits in with the ‘wrapping tools around tui’ point from the lash/TUI section to grant/revoke access to storage as it becomes available.
The ‘catch all’ dependency sponge afsrv_decode (external client with a stricter ruleset on behaviour for tighter sandboxing and privilege separation) that absorbs parsers (one much beloved exploitation target in offensive security) has received basic support for PDFs and similar vector formats via MuPDF.
In the clip below, you can see how the preview feature in the Durden browser HUD spins up a bunch of decode-pdf processes and then toggles to navigate one.
On systems that support the ‘v4l2-loopback’ device interface, the encode will now support exposing its input as a v4l2 device. This means that all the sharing and streaming features can emulate a webcam device for applications that trust such things. The video below shows exposing a window playing back a movie as being treated as a webcam in Chrome:
This article continues the long-lost series on how to migrate away from terminal protocols as the main building block for command-line and text-dominant user interfaces. The previous ones (Chasing the dream of a terminal-free CLI (frustration/idea, 2016) and Dawn of a new Command-Line Interface (design, 2017)) might be worth an extra read afterwards, but they are not prerequisites to understanding this one.
The value proposition and motivation is still that such a critical part of computing should not be limited to device restrictions set in place some 50-70 years ago. The resulting machinery is inefficient, complex, unreliable, slow and incapable. For what is arguably a strong raison d’être for current day UNIX derivates, that is not a strategic foundation to neither rely nor expand upon.
The focus this time is about the practicalities of the user facing ‘shell’ — the cancerous and confused mass that hides behind the seemingly harmless command-line prompt. The final article will be about the developer facing programming interfaces themselves as application building blocks, how all of this is put together, and the design considerations that go into such a thing.
This article is structured as follows:
What is ‘Shell’ – gives a short primer about the specific role the CLI ‘shell’ plays.
‘Gains’ – goes into the capabilities that new shells can take advantage off.
The following clip is a very quick teaser from using one in-progress replacement shell that has been built using the tools that will be covered here. While the shell itself will be presented in greater detail in another future article, it is available for adventurous souls to poke around with [and can be found in this GH repository].
Starting with other/related work; many have attempted to deal with the embarrassing legacy of terminals and their inherent limitations.
“NOTTY” focused on replacing the “in-band” signalling and command format. This addresses some of the protocol issues, but has an impedance mismatch with what the rest of your desktop or basic rendering expects, and the emulator-shell split remains.
Some, like “Hyper” and “Upterm“, rewrites the terminal emulator and shell in more and more advanced UI frameworks to get better cooperation with an outer graphical shell — inviting in all the complexities of rancid behemoths like Electron and GTK/Qt while still leaving the protocols and TUI libraries in their currently poor state.
Others, like “Notcurses et al.” replaces the key TUI libraries like Curses, Readline and so on. These fix neither the emulator nor the protocols. Worse still, a few make the protocol situation worse by introducing sidebands, hard-coding escape sequences or introducing new ones.
Then there are a number of attempts like jc and relational-pipes that proxy or modifying the exchange format between stdin/stdout in a single pipeline, but that is mostly orthogonal to the problem discussed; solving for the others would provide another pathway for negotiating multiple, concurrent, exchange formats.
What is ‘Shell‘?
First, if you want better and deeper reading into the subject, I would suggest:
(Oil)Why a new Shell – Disambiguation of programming interface versus user interface roles.
There are many more to be had, but piling them on would mainly add to the existing confusion between terms (console, shell, terminal, tui, gui, ..); these terms have contextual and historical interpretations that are slightly incompatible depending on where you are and where you come from, which makes discussing the topic even harder.
Here is a rough breakdown of different components and roles sufficient for the scope of this article:
Here, ‘Shell’ (as part of providing a textual shell as a command-line) is the first in line to consume and work with a terminal emulator, through a preassigned set of file descriptors (0, 1, 2) mapped to a terminal or pseudo-terminal device. Shells and some applications alike test and change their behaviour depending on the state of these descriptors (isatty(3) to tcgetattr(3) to ioctl(2)), sometimes referred to as an ‘interactive’ mode. These continue (unless explicitly told not to) to share and inherit into new jobs over this serial communication line.
The protocol/IPC marked blocks mask quite a bit of nuance as to how data exchange work and they are not created equal; the sockets used to communicate with the display server may be unpleasant, but are still infinitely better than this mix of “tty” devices, signalling groups, sessions and stdio used after the ‘terminal emulator’ stage — you do have to sit down and implement both consumer and producer side of the terminal instruction sets to get a fair grasp on just how bad things are. For the Linux kernel alone, the TTY layer is one that not even the most seasoned of developers wants to touch.
While the ’emulator’ part is often stripped and just referred to as ‘the terminal’ it is very much an emulator of ancient hardware (or rather the amalgamation of tens to hundreds of different ones). That fact should be stressed to emphasise the absurdity of it all — especially given the end goal of reading key presses and writing characters into a grid of cells.
There is valuable simplicity to TUIs (out of which CLIs are but one possibility), but that simplicity is wholly undone by the complexity of terminals and how the ‘instruction set / device model’ they expose makes the shell user experience itself unnecessarily hard to develop and provide.
It should be emphasised that the terminal emulator is also a poor take on a display server. This will become relevant later on. As such, it is at a disadvantage against better display servers for many reasons – one being that each job/client is not given a distinct bidirectional connection for data exchange, but instead share a single triplet of “files” (stdin, stdout, stderr), combined with a protocol that was never designed or intended for this.
This shared triplet as well as ‘multiplexing’ is important. Say that one of the shell-launched jobs is another cli shell itself, like gdb or glorious ed. Since the data is in-band over the shared set of stdio slots mapped to the kernel provided device, the previous shell(s) either needs to be full emulators on their own, or it they cannot safely intervene or layer other things on top of whatever the job is doing.
Even then it has few options for reliably restoring the emulated device state. This is why accidentally cat:ing something like /dev/random will quickly give you a screwed up prompt; it is likely that some of the many sequences that change character map, cursor or flow control was triggered – yet if the shell continues on unawares, the scroll-back history is forever tainted.
There are certainly was to hardcode and reset some state between jobs explicitly – and some shells do – but that also serve to mask the danger and fundamental issue with the design; it is executing random instructions in a complex and varying instruction set.
Back to the model. The ‘graphical shell’ and ‘textual shell’ refer to the abstraction the user actually interacts with. The other ‘shell’ (as in bash, zsh, …) serves at least two roles. First you have a primary role as a ‘window manager’ of sorts. This provides the “prompt”, parsing the command-line into ‘built-in’ command execution; constructing processing pipelines or executing ‘fullscreen’ applications and choosing which pipeline that is currently being “presented”.
The other thing it provides is a scriptable programming environment (as in the scripting part of shell-scripting). This is a secondary feature at best, and not at all necessary. In a command-line environment free from the legacy of terminals, current shells can continue to play this specific role — even if that is a job that should be left to more competent designs.
For the ‘window manager’ role: these range from (visually) simple ones like the foreground/background of bash, zsh and fish to more complex such as tmux and screen (sometimes referred to as multiplexers). The first ones tend to focus on how to articulate jobs and their data exchange, while the second on tiling like window management.
In order for these more complex ones to achieve window management, they also went through the strides of writing additional terminal emulators and embedding other shells (recursive premature composition) in order to output to yet another terminal emulator as there is no proper handover or embedding mechanism in place. It is terminal emulators all the way down and a reason why we have to talk about how shell is complicit in all of this – even the basics of what is expected from ‘reading a line’ requires dipping into terminal drawing, cursor and flow control commands.
This division also reflects the ‘modes’ of how the terminal protocols operate, which, in turn, tie back to what the computer output device actually was in various parts of the timeline. Recall that once upon a time the output was a printer (“line-based”) and only later became monitors (“screen-based”) with incrementally added luxuries like colour and interesting tangents like vector graphics. Moving the cursor around arbitrarily back across previous lines is a privilege — not a right.
You can see some of this bleed through with ‘scroll-back/history’ working poorly (or not at all) in the screen mode, with “tab/context” suggestion popups causing scrolling and weird wrapping visuals when the prompt is at the edge of the last ‘line’ on the ‘paper’ or – heavens forbid – you try to erase previous characters across pages or newlines.
If you want the worst of both worlds, go no further than regular ‘gdb’ (as in the GNU debugger) and go back and forth between ‘tui enable’, for the luxury of seeing the source code you are debugging at the cost of scrolling back through the data you needed, and ‘tui disable’ where every ephemerally relevant output gets committed to the ‘paper’ and the data you needed quickly scroll off into the far away distance.
You can also see it in the ‘tab completion’ output in a line-mode shell having to ‘add lines’ in order to fill in the completion, and those are kept there, polluting the history — as well as in the special treatment certain characters like ‘erase’ received. The man page to ‘stty‘ (or worse still, how a tty driver is written) is a brief yet still frightening look into the special properties of the underlying device itself.
For both modes, the protocols restrict what these two kinds of text-dominant shells can do and how they can cooperate with an outer graphical one. In the model presented so far, there is zero real cooperation between the text and the graphics shells. In reality, there is some, but implemented in a near impenetrable soup of hacks involving a forest of possible sideband protocols — and availability vary wildly with your choice of emulator, the protocol set it is defined to follow, and the contents of a capability database (terminfo/termcap).
As an exercise for the reader, try to work out how and why you can- or why you cannot-
Paste a block of text from your desktop clipboard into the current command-line.
Drag and drop a file into your shell and have it stored into the current working directory.
Click a URL in the command-line buffer history and have it open in your browser.
Redirect the output of a previous command to another window or tab.
Fold / unfold the presentation of output from previous commands.
Have an accurate clock in your prompt that updates by itself.
Neither of these are particularly exotic use-cases, some would even go so far as to say that these are fairly obvious things that should be trivial to support — yet if you think the answers to any of these are simple and easy, you missed something; there is a Lovecraftian horror hiding behind each and every one.
Simplifying and Exemplifying
Using the model from the previous section, we restructure it to this:
The terminal emulator is gone, the rightfully maligned ‘tty’ layer hiding in the kernel is gone. There are now a whole lot of ways for the graphical shell to cooperate with the textual one.
For this to work and provide enough gains, a lot of subtle nuances of the IPC system need to be in place; the one in Arcan (shmif) was specifically designed for this as one of several ‘grand challenges’ that were used to derive the intended feature set many, many, many years ago.
One of the main building blocks is ‘handover allocation’ – where the shell requests new resources in the graphical shell on behalf of an upcoming job, and then forwards the primitives needed to inherit a connection into the job, retaining the chain of trust and custody. Another is the live migration used as part of crash resilience, which eliminates the need for multiplexers as each client can redirect at runtime to other servers by design.
The main Arcan process takes the role of the display server. With that comes a pick of graphical shell, ranging from the modest ‘console‘ to the more advanced (durden, pipeworld, safespaces). Do note that there is a choice in building Arcan as the system display server with authority on GPUs, input devices and so on – or as a regular graphical client that you would run in place of your terminal emulator inside Xorg or some Wayland compositor. You will lose out on several performance gains, and some nuances in how window management integrates, but many features will remain.
For the ‘text shell’ block, there is a little bit more to think about. While it is perfectly valid and intended to use libarcan-tui to write your own here, one also comes included in the box. A regular Arcan build produces ‘afsrv_terminal’ (or arcterm as it is referred to internally).
This is a terminal emulator with a secret; if the argument “cli” is passed, it switches to an extremely barebones built-in text-shell and skips all the terminal emulation machinery. It is intended to provide only the absolutely necessary bits for something like booting a recovery image for an OS. If you are C inclined, this is a fair basis to expand on or borrow from.
In the following clip from the Pipeworld article (a graphical shell), you can see it in use in the form of the small CLI cell where I am typing in commands.
While things go quite fast, you might be able to spot how it transitions from a command-line as part of the graphical shell at 0:02 into a terminal emulator liberated textual CLI shell. You can then see that the jobs which spawn are their own separate processes and do not multiplex over the same pseudo-terminal devices (as that would make more than one ‘tui’ like job impossible or require nesting composition/state through something like screen or tmux, reintroducing the premature composition problem).
The twist is that these jobs are negotiated with the graphical shell being aware of their purpose and origin. This is a feature that runs deep and dates far back, already in use at the time of the One Night in Rio – Vacation Photos from Plan9 article. It is also the reason why jobs spawn as new detachable windows, yet retain hierarchy in the optics of the window management scheme.
Another setup can be found in this clip, also from Pipeworld:
Here we demonstrate how a processing pipeline can be built with separate outputs for each task in a pipeline, while at the same using stdin/stdout to convey the data that is to be processed. Any single one of these can be a strict text client, an arcan-tui one or wrapped around a terminal protocol decoder – yet both interacted with- and tracked- independently.
The ‘cli’ mode takes another argument, =lua. This enables a Lua VM, maps in API bindings, loads a basic script harness that provides some very crude and basic commands, but allows for plugging in a custom shell, like the one mentioned at the beginning of the article.
In this clip we can see a prompt from that shell where we run a job, and popup the ongoing results from that job into a window of its own with a hex view. The graphical shell, here operating in a tiling window manager setup, respects the request for this window to be a tab to the current one and creates it as such.
To add legacy to injury, this clip shows running a new job as a separate vertical-split window, wrapped around a terminal emulator. The standard error output, however, gets tracked and mapped into the shell view of ongoing jobs. Paste actions from the graphical shell has been set to accumulate into the data buffer of a job. This feature disabled and the paste action instead copies into the readline completion set, inserting at the cursor if activated.
In this clip we go even further – the shell opens a media resource, requests it to be embedded into its window. The resource scales, repositions and folds accordingly, yet the user can ‘drag it out’ should she so desire. The video playback in this case is delegated to one-shot dedicated processes, no parsing or exotic dependencies are imposed on the shell process itself.
The process responsible for composition gets to composite and gives the user independent controls for lossless decomposition.
In the following clip we see other forms of metadata interaction – the shell requests that the user picks a file, which it then redirects into a local copy inside the current working directory. The file picking it outsourced to whatever an outer graphical shell provides, and the chosen descriptor is forwarded into the text shell process that then saves it to disk. The process is repeated by picking an image file that is then opened and embedded similarly to the previous clip. Had the textual shell been running remotely or in some distant container, the transfer would have gone through just the same. The underlying mechanism works for explicit load-store, cut and paste as well as drag and drop.
There is much more to be had than the parlour tricks shown in the previous section. What can, immediately, without rose-tinted glasses and speculation, be gained by leveraging this infrastructure?
Data Communication – With an actual IPC system to connect through, it is possible to:
Leave STDIN/STDOUT/STDERR as pure data channels, not mixing in UI events or draw commands.
Accidentally catting a binary file or device cannot break the UI state machine.
Explicit serialisation of state (store / restore runtime config) without filesystem trails.
Every single command in a pipeline is left alone and kept separable between jobs and they do not interfere with shell communication. Thus each tool in a pipeline can provide both in-stream processing and an interactive user interface at the same time.
All interfaces strongly encourage asynchronous processing.
Binary blob transfers pass as file descriptors locally, and scheduled/multiplexed over the network.
Input – Having an event model that is not limited to a range of reserved values in the ASCII table delivered over a pipe allows:
Non-ambiguity – there is a discernible difference between pressing the ‘ESCAPE’ key and the ESC-ASCII character that was used to mark the beginning of an escape sequence.
Modifiers exist, CTRL+C is a symbolic C key with CTRL modifier, that does not equal ^C or it being magically translated to broadcasting SIGINT.
Pasting is separated from entering text, is separated from pressing keys, and can undo as a discrete whole.
Mouse input is predictable and reliable and can be combined with keyboard modifiers.
Keyboard shortcuts are announced and semantically tagged, letting the graphical shell provide automation, rebinding and mapping to assistive devices.
Integration: With the same language for expressing graphical clients as for textual ones:
Clients can be redirected between shells at runtime, even across a network connection.
Graphical shell capabilities can be leveraged for universal file picking.
Decorations like borders and scrollbar are deferred to the outer graphical shell, avoiding mixing data with metadata in the grid by drawing ‘line characters’.
If the graphical shell is sufficiently capable — notifications, alerts, file picking and popups become available and behave according to the rules of the graphical shell.
Visuals and Performance:
The rendering responsibilities have been moved to the display server end of the equation, while the fonts currently in use are being passed as reference objects for features that need it (ligatures, …). There are no pixel buffers being passed around from the ’emulator’ client and the shell is explicit about when it is time to synchronise content onwards.
Tear-free updates and resizing.
Presentation buffer back-pressure control is deferred to the job, no more heuristics in the emulator.
Colours are always 24-bit or from a semantic palette (no more “red” is now “green”).
Embeddable interactive media content (* assuming an outer graphics shell support it).
Synchronised presentation, atomic commit of change sets and only updates between sets are synched.
Glyph caches can be shared between multiple shell instances and other clients.
Glyph indices and availability are queryable so fallbacks can be chosen.
Single buffered ‘chasing the beam’ style rasterisation for lowest possible latency.
Separate on-demand alt-views to propagate compacted accessibility friendly contents.
Semantically tagged input lets screenreader say ‘paste into job #1…’ rather than ctrl-v, proper separation between I/O streams makes it trivial to build ‘audio-only’ shell.
Locale properties on input language and presentation language can change at runtime and is the property of a window, not a process global passed through environment variables.
Everything is unicode.
These are features with direct impact for writing better shells. Then there are parts for writing better TUI applications and other command-line tools in general, but that is for another time.
Time for another fairly beefy Arcan release. For those thinking in semantic versioning still surprised at the change-set versus version number here (‘just a minor update?’) do recall that as per the roadmap, disregarding the optimistic timeline, we work with release-rarely-build-yourself thematic versions until that fated 1.0 where we switch to semantic and release-often.
On that scale, 0.5.x was the display server role, 0.6.x is focused on the networking layer as the main feature and development driver. 0.7.x will be improving audio and some missing compatibility/3D/VR bits. Then it gets substantially easier/faster – 0.8.x will be optimization and performance. 0.9.x will be security — hardening attack surface, verification of protections, continuous fuzzing infrastructure and so on.
After the FreeNode IRC kerfuffle, do note that the oldtimers IRC channel has moved to Libera and whatever remains of the old freenode channel is to be assumed malicious. For some of the younger folks we have added a Discord server pending more manageable and sane alternatives – pick your poison(s).
The detailed changelog can be found here, and the release highlights are as follows:
For the main engine there has been quite some refactoring to reduce input latency; better accommodate variable-refresh rate display; prepare for asymmetric uncooperative multi-GPU and GPU handover; explicit synchronisation and runtime transitions back and forth between low (16-bit) to standard (32-bit) to high-definition rendering (10-bit + fp16/fp32).
The biggest task has been to add support for “direct-to-drain” in the event queue management. This allows for more event producers (e.g. input drivers) to get a short path for minimising input latency. It also allows our scheduler (“conductor”) to reorder event delivery to better accommodate a WM specified focus target.
The major challenge has been that some operations are unsafe to perform when a GPU is busy drawing and scanning out to displays. Misuse can cause subtle visual glitches or stutter in inputs that are fiendishly hard to catch and diagnose. This was a big ticket item as part of the synchronization topic.
Since the Window Manager needs to be able to be first in line to filter- translate- remap- resample- mask- and otherwise respond- to input events — quite some care had to be taken to extend the API to make sure that it remains ergonomic enough to respond to input events in both a safe mode and a low latency one.
With this also comes the ability to allow selected clients to get a higher direct-to-drain queue dispatch so that we can have more efficient hot pluggable external input drivers, something that would, for example, benefit our VR service.
As mentioned before, the main focus of this development branch is still the networking layer, and our related SSH/VNC/RDP/X11 replacement A12 network protocol, which is a priority for some of our sponsors.
While much of this work is not directly visible, some of the bigger changes have been to drop any and all traces of the DEFLATE support and go all in on ZSTD for binary transfers, TPACK and lossless image modes.
Connection-modes: The ‘last’ missing connection mode has mostly been sorted, though there are some nuances still to work out, particularly for serving complex clients like Xarcan.
To elaborate: since the tools (not the protocol or library) were built to mimic the use of X11 style forwarding, the ‘normal’ default push model worked like this:
While simpler, this form has the downside that ‘crash recovery’ won’t work the same, as the listening end don’t know where to redirect ‘some_arcan_client’ here. For the other form, the keyid@ (see keystore below) can have multiple redundancies should one host lost connectivity or fail to respond.
Keystore: A basic keystore has been added for managing cryptographic key material and identities, and while the tooling is lacking, it should be sufficient for testing and non-sensitive use.
The following example shows how to setup one up, using a one-time password for authenticating the public keys, with the ‘push’ form of connectivity:
arcan-net keystore myhost 192.168.0.2 6680
echo 'somepass' | arcan-net -a 1 -s out myhost@ &
echo 'somepass' | arcan-net -a 1 -l 6680
One difference from ssh “type yes to be subject to man in the middle” is that a shared secret is used the first time (-a n == read or prompt secret from stdin and accept the first n public keys that had a valid shared secret). Another one is that the host tag for outbound connections can have multiple possible endpoints by adding more hosts to the same keyid. This ties into crash recovery as that also covers loss of connectivity.
The new default is that if there is no working keystore, arcan-net will refuse to run. You have to opt out of stronger authentication in one or two forms:
With the first form using a password stored in ‘mypw’, while the second will just appear protected (asymmetric primitives generated on the fly and assumed trusted) if you look at the network traffic (i.e. requires active MiM still), anyone is still allowed to connect.
For anything more serious the suggestion would still be to tunnel over Wireguard. I personally have no problems running a12 over untrusted networks for certain tasks — but my threat model is not your threat model and all that.
Backpressure: Changes in both protocol and implementation has been made to better track the current video frame back-pressure as part of congestion control. This is a rather involved topic that will need several iterations to find good parameter sets to switch between depending on the contents — media playback that doesn’t depend on user interaction to produce input can go with deeper buffers, while games would prefer shallow ones, with depth based on latency and bandwidth estimation.
The upside is that improvements here also translate to the local non-networked case — every layer is involved in getting timing better.
What is next for the networking stack is to find an optimal way of leveraging the keystore to answer the question of “which of the devices that know and have a previously established relationship to are currently reachable, at home or overseas” in a privacy preserving way (i.e. service discovery).
Other big ticket items is to make the keystore friendlier for setup and authentication of unknown public keys, expand back-pressure management with dynamic video codec bitrate selection and pre-compressed video passthrough so that a client that exposes a window with a h264 stream first tries to forward that intact, and only recompress if the network conditions are more strict.
XArcan, Waylandand other transports
While not strictly part of this ‘version’ as it is a standalone client and separate repository, our Xorg fork “Xarcan” now handles GPU transitions more gracefully and has received basic clipboard integration as well as accelerated cursor support.
The last updates to this should be pushed and tagged within the coming weeks, but worth mentioning nonetheless, and much more work is planned for this one as we continue to move further away from using Wayland as a client compatibility solution — even if that means writing custom Qt and Chrome-ozone platform plugins.
One thing on the compatibility workbench that, albeit experimental still, is that it is now possible to use X Window Managers to manage and decorate Arcan windows — while still not granting default access to their contents or input routing. Thus for those who are stuck in old ways when it comes to configuration, keyboards and window management, there is a migration path close to being useable.
A longer article on how/why this works, and how it can also be used to move Xorg forward, is being written and should be published shortly.
The Wayland implementation in the arcan-wayland service have also seen some updates, with the main ones being repackaging wl-shm (shared memory) buffers to forward as dma-buf when possible (keeping the GPU transfer cost outside of the composition process), and forwarding pulseaudio sockets into the temporary directory used with -exec and -exec-x11 single client use.
Our “curses/vt100 replacement” client API switches to server-side text rendering only. Cell attributes can now operate in indexed-colour mode where the server-defined colour scheme gets resolved and applied automatically.
API calls for hinting about window contents for server-side scroll controls are now exposed.
The related Lua bindings have been updated and are closer to complete, and now quite usable – with mostly the widgets left that are missing a Lua friendly API.
The NeoVIM frontend have been updated accordingly and now also allows buffer transfers. This is useful in Pipeworld pipelines for building mixed graphical/text/tui processing pipelines and using VIM as the editor for REPL like workflows.
There is also experimental integration with the Kakoune editor that can be found here thanks to Cipharius.
Our “single-purpose” designated special clients for engine core offloading and security compartmentation of sensitive tasks has also seen some new capabilities:
Decode: Added support for proto=img with a similar codebase and sandboxing setup as the aloadimage tool uses. It can also be run in a daemon like mode where the main engine can offload image decoding and vector image re-rasterisation (for low<->high DPI transitions).
The UVC based decode video capture stage has also received controls for toggling tracking dirty regions when running multi-buffered, so if you are using UVC based capture devices such as the ElGato Camlink 4k with stable low frequency changing sources (HDMI or good quality security cameras) fewer wasteful frames should propagate.
Terminal: the terminal emulator received support for dynamically switching colour schemes and server-side defined colour preferences. It can now resize to fit to contents when kept alive after the shell exits. It also received a pty-less ‘piped’ mode for having multiple instances displaying stdin/stdout contents as part of inspectable and individually controllable pipelines.
The low level graphics platform used when arcan acts as the native / primary display server has seen a number of smaller fixes to how modifiers are setup and propagated. More importantly though:
Front buffer text-surfaces: When the WM requests that a source with a TUI (TPACK) is to be mapped as the source to a single display, it will switch to single-front buffer rendering. This means that our terminal emulator and similar simple TUI clients can be run with close to the lowest possible latency and CPU cost.
FBO/EGLScreen scanout transitions: We can now also dynamically switch between the preferred ‘FBO’ direct scanout and the much maligned EGLScreen based recomposition scanout dynamically.
The reblit-to-EGL stage is still the safe default, as there are some really quirky multi-rotated-display resolution switching corner cases to work out still.
Low/SDR/Deep/HDR (+-VRR) transitions: The transitions back and forth between these were a big and fairly complicated part of the HDR/Color Processing work and are now working for the low/sdr/deep cases. There are low level dragons to slay here before the last stretch for working HDR composition, somewhat being blocked by my rather pricey DisplayHDR1000 monitor crashing outright on amdgpu when enabled, while on my intel GPUs the driver crashes instead. Fun times.
On the scripting side, there is now an optional “input_raw” entry point for indicating support for out-of-bound/direct-to-drain input event processing to reflect the changes to core input routing mentioned before.
Non-blocking I/O: The open_nonblock set of functions now has a “data_handler” function for setting a rising-edge event handler to make it cleaner to use than the non-blocking periodic polling approach. Similarly, the write call has added automatic queueing and a completion event. The scheduler will try to favour I/O operations while the platform is busy rendering/scanning out with GPUs in a locked state.
Platform control functions: The video_displaymodes() function has received an overloaded form for switching between desired sink colour depth (Low, Normal, Deep, HDR) as well as variable refresh toggles. Similarly, the map_video_display() function now also supports mapping several layers for hardware platforms with hardware controlled basic composition.
For platforms with low level input translation tables on keyboards (e.g. evdev), the input_remap_translation() function has been added to control/hint/swap desired LMVO (layout/model/variant/option) being applied to one or many keyboards.
Staggered client launch: The state control functions resume_target/suspend_target controls can now be applied to clients that are still locked in the initial preroll/open. This allows for staggering client releases on event storms — or when you want to have more precise control over client environment stage, e.g. remap stdin/stdout/stderr to descriptors delivered over the shmif connection. Pipeworld, for example, uses this to implement its pipe(cmd_1, cmd_2, …) command where the entire pipeline is first built and executed, then activated in one go.
Build / Packaging / OS Specific
NixOS: Thanks to the tireless work of Albin (a12l) and AndersonTorres, we are now packaged in NixOS.
BSDs: We should now be usable on all the major BSDs (Open, Free, Net, Dragonfly) thanks to Jan Beich, Zrj, Antonio, Leonid and others.
Also worthwhile to note is that the OpenBSD performance regression have been solved for 6.9, but it seems 7.0 brought other timing issues. It turned out to be tied to the VRR conductor refactoring making assumptions about kernel tick granularity.
This is now dynamically probed and reverted to a display-clocked scheduling tactic if the OS Kernel has a low scheduler tick configured (the default in OpenBSD being 100Hz still). While not recommended for prolonged use, the “-W processing” synchronisation strategy mitigates sluggishness at the expense of CPU.
A new ‘arcan-dbgcapture’ tool has been added. This acts as a tui- wrapper for core dumps, and is best used as a core_pattern handler on Linux. Here you can see it being used in this way inside Pipeworld:
Time to continue to explain what Arcan actually “is” on a higher level. Previous articles have invited the comparison to Xorg ( part1, part2 ). Another possibility would have been Plan9, but Xorg was also a better fit also for the next (and last) article in this series.
To start with a grand statement:
Arcan is a single-user, user-facing, networked overlay operating system.
With “single-user, user-facing” I mean that you are the core concern; it is about providing you with controls. There is no compromise made to “serve” a large number of concurrent users, to route and filter the most traffic, or to store and access data the fastest anywhere on earth.
With “overlay operating system” I mean that it is built from user-facing components. Arcan takes whatever you have access to and expands from there. It is not hinged on the life and death of neither the Linux kernel, the BSD ones or any other for that matter. Instead it is a vagabond that will move to whatever ecosystem you can develop and run programs on, even if that means being walled inside an app store somewhere.
As such it follows a most pervasive trend in hardware. That trend is treating the traditional OS kernel as a necessary evil to work around while building the “real” operating system elsewhere. For a more thorough perspective on that subject, refer to USENIX ATC’21: It’s time for Operating Systems to Rediscover Hardware (Video).
This is a trick from the old G👀gle book: they did it to GNU/Linux with Android and with ChromeOS. It is not as much the predecessor mantra of “embrace, extend and extinguish” as it is one of “living off the land” — understanding where the best fit is within the ecosystem at large.
From this description of what Arcan is — what is the purpose of that?
The purpose of Arcan is to give you autonomy over all the compute around you.
With “autonomy” I mean your ability to move, wipe, replace, relocate or otherwise alter the state that is created, mutated or otherwise modified on each of these computing devices.
The story behind this is not hard in retrospect; user-facing computers are omnipresent in modern life — they outnumber us. You have phones, watches, tablets, “IoT” devices, e-Readers, “Desktop” Workstations, Laptops, Gaming Consoles and various “smart”- fridges, cars and meters and so on. The reality is that if a computer can be fitted into something, be sure that one or several will be shoved into it, repeatedly.
The fundamentals on how these computers work differ very little; even “embedded” is often grossly overpowered for the task at hand. On the other hand, getting these things you supposedly own to collaborate or even simply share state you directly or indirectly create is often hard- or impossible- to achieve without parasitic intermediaries. The latest take on this subject on the time of writing this is parts of “cloud”; routing- and subjecting- things to someone else’s computing between source (producer) and sink (consumer).
Part of the reason for this is persistent and deliberate balkanisation combined with manipulative monetisation strategies that have been permitted to co-evolve and steer development for a very long time; advances in cryptography has cemented this.
An example: In the nineties it was absurd to think that the entire vertical (all the ‘layers’) from datastore all the way to display would have an unbroken chain of trust. The wettest of dreams that the Hollywood establishment had was that media playback was completely protected from the user tampering with- or even observing- the data stream until presented. That is now far from absurd; it is the assumed default reality and you are rarely allowed to set or change the keys.
The scary next evolution of this is making you into a sensor and it is sold through claims of stronger security features that are supposed to protect you from some aggressor. A convenient side effect is that it actually serves to safeguards the authenticity of the data collected from-, of- and about- you. As a simple indicator: when no authentication primitive (password etc.) is needed, the “ai in the cloud” model of you has locked things down to the point that you best behave if you want to keep accessing those delicious services your comfort depend on.
The overall high-level vision how this development can be countered on a design basis is covered by the 12 principles for a diverging desktop future on our sister blog — but the societal implementation is left as an exercise for the reader.
A short example of the general idea in play can be seen in this old clip:
This demonstrates live migration between different kinds of clients moving to arcan instances with different levels of native integration and no intermediaries. As a native display server on OpenBSD in the centre to a limited application on the laptop to the left (OSX) and to a native display server on the pinephone on the right.
The meat and remainder of this article will provide an overview of the following:
The following diagram illustrates the different building blocks and how they fit together.
SHMIF – Shared Memory Interface
SHMIF is a privilege barrier between tasks built only using shared memory and synchronisation primitives (commonly semaphores) as necessary and sufficient components. This means that if an application has few other needs over what shmif provides, all native OS calls can be dropped after connection setup. That makes for a strong ‘least-privilege’ building block.
It fills the role of both asynchronous system calls and asynchronous inter-process communication (IPC), rolled into one interface. The main inspiration for this is the ‘cartridges‘ of yore — how entire computers were plugged in and removed at the user’s behest.
There is a lot of nuance to the layout and specifics of SHMIF which are currently out of scope, notes can be found in the Wiki. The main piece is a shared 128b event structure that is serialised over two fixed size ring buffers, one in-bound and one out-bound. The rest is contested metadata that is synchronously negotiated and transferred to current metadata that both sides maintain a copy of.
This is optionally extended with handle/resource-token blob references when conditions so permit (e.g. unix domain socket availability or the less terrible Win32 style DuplicateHandle calls), as that is useful for by-reference passing large buffers around which can be preferred for accelerated graphics among many other things.
The data model for the events passed around is static and lock stepped. This model is derived from the needs of a single user “desktop”. It has been verified against both X11, Android as well as special purpose “computer-in-computer” clients like whole-system emulation à QEmu, specialised hybrid input/out devices (e.g. streamdeck) and virtual- augmented- and mixed- reality (e.g. safespaces).
A summary of major bits of shmif:
Connection: named, inheritance, delegation, redirection and recovery.
Color profiles and Display controls (transfer functions, metadata).
Non-blocking State transfers (clipboard, drag and drop, universal open/save, fonts).
Blocking state controls (snapshot, restore, reset, suspend, resume).
Synchronisation (to display, to event handler, to custom clock, to fence, free-running).
Color scheme preferences.
Key/value config persistence and “per window” UUID.
Coarse grained timers.
Most of this is additive – there is a very small base and the rest are opt-in events to respond to or discard. All event routing goes through user-scriptable paths that are blocked by default, forcing explicit forwarding. A client does not get to know something unless the active set of scripts (typically your ‘window manager’) routes it.
Upon establishing a connection, requesting a new ‘window’ (pull) or receiving one initiated by the server side (push), the window is bound to an immutable type. This type hints at policy and content to influence window management, routing and to the engine scheduler.
A game has different scheduling demands from movie playback; an icon can attach to other UI components such as a tray or dock and so on. This solves for accessibility and similar tough edge cases, and translates to better network performance.
A12 – Network Protocol
SHMIF is locally optimal. Whenever the primitives needed for inter-process SHMIF cannot be fulfilled, there is A12. The obvious case is networked communication where there is no low latency shared memory, only comparably high-latency copy-in-copy-out transfers.
There are other less obvious cases, with the rule of thumb being that two different system components cannot synchronise over shared memory with predictable latency and bandwidth. For instance, ‘walled garden’ ecosystems tend to disallow most forms of interprocess communication, while still allowing networked communication to go through.
A12 has lossless translation to/from SHMIF, but comes with an additional set of constraints to consider and builds on the type model of SHMIF to influence buffer behaviour, congestion control and compression parameters.
The constraints placed on the design are many. A12 needs to usable for bootstrapping; operate in hostile environments; isolated/islanded networks and between machines with unreliable clocks; incompatible namespaces and possibly ephemeral-transitive trust. For these reasons, A12 deviates from the TLS model of cryptography. It relies on a static selection of asymmetric- and symmetric- primitives with pre-shared secret ‘Trust On First Use’ management rather than Certificate Authorities.
Around the same era that browsers started investing heavily into sandboxing (late 2000s, early 2010s) Arcan, then a closed source research project, also focused on ephemeral least privilege process separation of security and stability sensitive tasks. The processes that carry out such tasks are referred to as ‘frameservers’.
In principle ‘frameservers’ are simply normal applications using SHMIF with an added naming scheme to them and a chainloader (arcan_frameserver) that is responsible for sanitising and setting up respective execution environments.
In practice ‘frameservers’ have designated roles (archetypes). These control how the rest of the system delegates certain tasks, and gives predictable consequences to what happens if one would crash or be forcibly terminated. It is also used to put a stronger contract on accepted arguments and behavioural response to the various SHMIF defined events.
The main roles worth covering here are ‘encode‘, ‘decode‘ and to some extent, ‘net‘.
Decode samples some untrusted input source e.g. a video file, a camera device or a vector image description, and converts it into something that you can see and/or hear, tuned to some output display. This consolidates ‘parsing’ to single task processes across the entire system. These processes have discrete, synchronous stages where incrementally more privileges, e.g. allocating memory or accessing file storage, can be dropped. The security story section goes a bit deeper into the value of this.
Encode transforms something you can see and/or hear into some alternate representation intended for an untrusted output. Some examples of this are video recording, image-to-text (OCR) and similar forms of lossy irreversible transforms.
Net sets up transition and handover between shmif and a12. It also acts as networked service discovery of pre-established trust relationships (“which devices that I trust are available”, “have any new devices that I trust become available”) and as a name resource intermediary e.g. “give me a descriptor for an outbound connection to <name in namespace>”.
Splitting clients into this “ephemeral one-task” and regular ones lead to dedicated higher level APIs for traditionally complex tasks, as well as acting as delegates for other programs.
It is possible for another shmif client to say “allocate a new window, delegate this to a decode frameserver provided with this file and embed into my own at this anchor point” with very few lines of code. This lets the decode frameserver act as a parser/decode/dependency sponge. Clients can be made simpler and not invite more of the troubles of toolkits.
The ‘engine’ here is covered by the main arcan binary and invites parallels to app frameworks in the node.js/electron sense, as well as the ‘Zygote’ of Android.
It fills two roles — one is to act as the outer ‘display server’ which performs the last composition and binds with the host OS. The scripts that run within this role is roughly comparable to a ‘window manager’, but with much stronger controls as it acts as ‘firewall/filter/deep inspection’ for all your interactions and data.
The other role is marked as ‘lwa’ (lightweight arcan) in the diagram and is a separate build of the engine. This build act as a SHMIF client and is able to connect to another ‘lwa’ instance or the outermost instance running as the display server. This lets the same code and API act as both display server, ‘stream processor’ (see: AWK for Multimedia) and the ‘primitives’ half of a traditional UI toolkit.
Both of these roles are programmable with the API marked as ‘ALT’ in the diagram and will be revisited in the sections on ‘Programmable Interfaces’ and ‘Appl’.
The architecture and internal design of the engine itself is too niche to cover in sufficient detail. Instead we will merely lay out the main requirements that distinguish it against the many strong players in the core 2D- supportive 3D- game engine genre.
Capability – enough advanced graphics and animation support for writing applications and user interfaces on the visual and interactive span of something ranging from your run-of-the-mill web or mobile ‘app’ to the Sci-Fi movie ‘flair over function’ UIs. It should not rely on techniques that would exclude networked rendering or exclude devices which cannot provide hardware acceleration.
Abstraction – The programmable APIs should be focused on primitives (geometry, texturing, filtering), not aggregates/patterns (look and feel). Transforms and animations should be declarative (“I want this to move to here over 20 units of time”), (“I want position of a to be relative to position of b”) and let the engine solve for scheduling, interpolation and other quality of experience parameters.
Robust – The engine should be able to operate reliably in a constrained environment with little to no other support (service managers, display/device managers). It should avoid external dependencies. It should be able to run extended periods of time — months, not hours or days.
Resilient – The engine should be able to recover from reasonable amounts of failures in its own processing, and that of volatile hardware components (primarily GPU). It should be able to hand over/migrate clients to other instances of itself.
Recursive – The engine should be able to treat additional instances of itself as it would any other node in the scene graph, either as an external source node or a subgraph output sink node.
SHMIF has been covered already from the perspective of its use as an IPC system. As an effect of this, it is also a programmable low-level interface. A thorough article on using it can be found in (writing a low level arcan client), and a more complex/nuanced follow up in (writing a tray icon handler). The QEmu UI driver, the arcan-wayland bridge and Xarcan are also interesting points of reference to hack on.
TUI is an abstraction built on top of SHMIF. It masks out some features and provides default handlers for certain events — as well as translating to/from a grid of cells- or row-list- of formatted text. It comes with the very basics of widgets (readline, listview, bufferview). Its main role in the stack is to replace (n)curses style libraries and improve on text dominant tools as a migration strategy for finally leaving terminal emulation to rest.
ALT is the high level API (and protocol*) for controlling the Engine. The primary way of using it is as Lua scripts, but the intention is a bit more subtle than that. For half of this story see ‘Appl’ below. Lua was chosen for the engine script interface in part for its small size (with that low memory overhead and short start times), the easy binding API and minimal set of included functions. It is treated and thought of as a “safe” function decoration for C more than it is treated as a normal language.
The *protocol part is that the documentation for the API double as a high level Interface Description Language to generate bindings that would use the API out of process — allowing both Lua “monkey patching” by the user and process separation with an intermediate protocol. This makes the render process and ALT into a dynamic sort of postscript for applications with animations and composition effects rather than static printer-friendly pages.
This is not a discrete component, but rather a set of restrictions and naming conventions added on top of the core engine. To understand this, a rough comparison to Android is again in order.
The Android App is, grossly simplified, a Zip archive with some hacks, a manifest XML file, some Java VM byte code, optional resources and optional native code. The byte code traditionally came from compiling Java code, but several languages can compile to it. The manifest covers some metadata, importantly which system resources that the app should have access to.
The Arcan Appl (the ‘l’ is pronounced with a sigh or ‘blowing raspberries’) has a folder structure:
A subdirectory to some appl root store with a unique name
A .lua file in that directory with the same name.
A function with the same name as the directory and the .lua file.
Resources the appl can access and data stores it can create and delete files within are broken down into several namespaces. The main such namespaces are roughly: application-local-dynamic, application-local-static, fonts, library code and shared.
Similarly to how the Android app can load native code via JNI, the Arcan appl can dynamically load shared libraries. In contrast to Android where native code is pulled in for supporting the high level Java/Kotlin/…, the high level scripting in Arcan is to support the native code so that the tedious and error prone tasks are written in a memory safe, user hackable/patchable code by default.
The mapping of the namespaces themselves, restrictions or additional permissions, configuration database and even different sets of frameservers are all controlled by the arguments that was used to start each engine process.
The database act as a key / value store for the appl itself, but also as a policy model for which other shmif capable clients should be allowed to launch (* enforcement for native code is subject to controls provided by the host OS), as well as a key / value store for tracking information for each client launched in such a way.
Other resources permissions are not directly requested or statically defined by the appl itself, it the window manager that ultimately maps and routes such things.
There are a number of reference appls that have been written and presented on this page throughout the years. These are mainly focused on filling the ‘window manager’ role, but can indeed also be used as building blocks for other applications.
These have been used to drive the design of ALT API itself; to demonstrate the rough scope of what can be accomplished, and they are usable tools in their own right.
The ones suitable as building blocks for custom setups or ‘daily drivers’ are as follows:
Console — (writing a console replacement using Arcan) which act as fullscreen workspaces dedicated to individual clients, with a terminal spawned by default in an empty one. This is analogous to the console in Linux/BSD setups and comes bundled with the default build (but with unicode fonts, OSD keyboard, touchscreen support, …).
Durden — Implements the feature set of traditional desktops and established window management schemes (tiling, stacking/floating and several hybrids). It is arranged as a virtual filesystem tree that UI elements and device inputs reference.
Safespaces — Structurally similar to Durden as far as a ‘virtual filesystem’ goes, but intended for Augmented-, Mixed- and Virtual Reality.
Pipeworld — Covers ‘Dataflow’; a hybrid between a programming environment, a spreadsheet and a desktop.
There are several intimidating uphill battles with established conventions and the network/lock-in effect of large platforms — no interesting applications equals no invested users; no developers equals no interesting applications.
The most problematic part for compatibility comes with ‘toolkit’ (e.q. Qt and GTK) built ones. Although often touted as ‘portable’, what has happened time and again is a convergence to some crude and uninteresting capability set tied to whatever platform abstraction can be found deep in the toolkit code base — it is never pretty.
There is fair reason as for why many impactful projects went with ‘the browser as the toolkit’ (i.e. Electron). The portability aspects for the big toolkits will keep on loosing relevance; the long term survival rate for well-integrated ‘native’ feel portable software looks slim to none. The end-game for these rather looks like banking on one fixed idea/style or niche.
The compatibility strategy for Arcan is “emphasis at the extremes” — first to focus on the extreme that is treating other applications as opaque virtual machines (which include Browsers). Virtualization for compatibility is the strongest tactic we have for taking legacy with us. This calls for multiple tactics to handle integration edge cases and incrementally breaking the opaqueness – such as forced decomposition through side-band communicated annotations, “guest-additions” and virtual device drivers.
The second part to the strategy is to focus on the other extreme that is the ‘text dominant’ applications, hence all the work done on the TUI API. As mentioned before, it is needed as a way out of getting ‘Terminals’ with command lines that are not stuck with hopelessly dated and broken assumptions on anything computing. Terminal emulators will be necessary for a long time, and Arcan comes with one by default — but as a normal TUI client.
TUI is also used as a way of building front-ends to notoriously problematic system controls such as WiFi authentication and dynamic storage management. It is also useful for ‘wrapping’ data around interactive transfer controls; leave the UI wrapping and composition up to the appl stage.
The distant third part of the compatibility strategy are protocol bridges — the main one currently being ‘arcan-wayland’. For a while, this was the intended first strategy, but after so many years of the spec being woefully incomplete, then seriously confused, it is now completely deranged and ready for the asylum. That might sound grim, yet this is nothing compared to the ‘quality’ of the implementations out there.
One area that warrants special treatment is security (and the overlap with some of privacy). This is an area where Arcan is especially opinionated. A much longer treatment of the topic is needed, and an article for that is in the queue.
The much condensed overarching problem of major platforms are that they keep piling on ‘security features’ (for your own good they say) and (often pointless) restrictions or interruptions that are incrementally and silently added through updates — with you in the dark of what they are actually supposed to protect you from, and the caveats with that.
The following two screenshots illustrate the entry level of this problem:
Note: The very idea that the second one even became a dialog is surprising. Most UIs that predates this idiocy had trained users to route data using their own initiative and interaction alone through “drag and drop”, “copy and paste” and so on. It is a dangerous pattern in its own right, and a mildly competent threat actor knows how to leverage this side channel.
There is a lot to unpack with these two alone, but that is for another time.
The core matter is that these fit in some threat model, but unlikely to be part of your threat model. The tools to actually let you express what your threat model currently is, and tools to select mitigations to fit your circumstances — are both practically nonexistent.
Compare to accessibility: supporting vision impairment, or blindness, has vastly different needs from deafness which has different needs from immobility. Running a screen reader will provide little comfort to someone that is hard of hearing and so on. Turning such features on without the user being informed; or even rudely interrupting by repeatedly asking at every possible junction, is rightly to be met with some contention.
In the other end, someone working on malware analysis has different needs from someone approving financial transactions for a company has different needs from someone forced to use a tracking app from an abusive partner or employer. Yet here protections with different edge cases and failure modes are silently enabled without considering the user.
The security story is dynamic and context dependent by its very nature. A single person could be switching between having any of the needs expressed above at different times over the course of a single day. More technically, it might be fine for your laptop to “on-lid-open: automatically enable baseband radio, scan for known access points, connect to them, request/apply DHCP configuration” coupled with some other service waking up to “on network: request and apply pending updates” and so on from the comforts of your home. It might also get you royally owned while at an airport in seemingly infinitely many ways.
To tie things back to the Arcan design. The larger story comes from the 12 principles linked earlier, and a few of those are further expanded into the following maxims:
The Window Manager defines the set of mitigations for your current threat model.
This is hopefully the least complicated one to understand. To break it down further:
The window manager is first in line to operationalise your intents and actions.
The window manager reflects your preferences as to how your computer should behave.
You, or someone acting on your behalf, should always have the agency to tune or work around undesired behaviours or appearances.
Any interaction should be transformable into automation through your initiative and bindable to triggers that you pick.
Automation and convenience should be easy to define and discover, but not a default.
The point is to make the set of scripts that is the Appl controlling the outermost instance of Arcan (as a display server) the primary control plane for your interests. If you think of each client/application as possibly sandboxed-local or sandboxed-networked, the scripts define the routing/filtering/translation rules between any source to any sink — a firewall for your senses.
There is no IPC but SHMIF.
Memory safety vulnerabilities (typically data/protocol parsers) were for a long time a cheap and easy way to gain access into a system (RCE – Remote Code Execution).
The cost and difficulty increased drastically with certain mitigations, e.g. ASLR, NX, W^X and stack canaries – but also through least-privilege separation of sensitive tasks (sandboxing). Neither of these are panaceas but they have raised the price and effort so substantially that there is serious economy and engineering effort behind just remote code execution alone (public examples) — which is far from what goes into a full implant.
Bad programming patterns break mitigations. If you don’t design the entire solution around least-privilege, very little can safely be sandboxed. In UNIX, everything is a file-descriptor. Subsequently, blocking write() to file-descriptors breaks everything.
What happens when trying to sandbox around non least-privilege friendly components is that you get IPC systems. Without a systemic perspective you end up with a lot of them, and they are really hard to get right. Android developers had serious rigour and a lot of effort in Binder (their primary IPC system) yet it was both directly and indirectly used to break phones for many years — and probably still is.
Few IPC systems actually gets tested or treated as security boundaries, and eventually you get what in offensive security is called ‘lateral movement’.
This is the story of (*deep breath*) how the sandboxed but vulnerable GTK based image parser triggered by a D-Bus (IPC) activated indexer service on your mail attachment exposed RCE via a heap corruption exploited with a payload that proceeded to leverage one of the many Use-after-Frees in the compositor Wayland (IPC) implementation, where it gained persistence by dropping a second stage loader into dconf (IPC) that used PulseAudio (IPC) over a PipeWire (IPC) initiated SIP session to exfiltrate stolen data and receive further commands without your expensive NBAD/IDS or lazy blue team noticing — probably just another voice call.
In reality the scenario above will just be used to win some exotic CTF challenge or impress women. What will actually happen is that some uninteresting pip-installed python script dependency (or glorious crash collection script) just netcats $HOME/.ssh/id_rsa (that you just used everywhere didn’t you?) to a pastebin like service– but that’ll get fixed when everything is rewritten in Rust so stay calm and continue doing nothing.
The point of SHMIF is to have that one IPC system (omitting link to the tired xkcd strip, you know the one); not to end them all, or be gloriously flexible with versioned code-gen ready for another edition of ‘programming pearls’ — but to solve for the data transport, hardening, monitoring and sandboxing for only the flows necessary and sufficient for the desktop.
Least privilege parsing and scrubbing
Far from all memory safety vulnerabilities are created equal. The interesting subset is quite small, and somehow need to be reachable by aggressor controlled data. That tend to be ‘parsers’ for various protocols and document/image formats. If you don’t believe me, believe Sergey and Meredith (Science of Insecurity).
This can (and should be) leveraged. Even parsing the most insane of data formats (PDF) have fairly predictable system demands. With a little care they can do without any system calls at all after they have been buffered, and it is really hard to do anything from that point even with robust RCE.
This is where we return to the ‘decode’ frameserver — a known binary that any application can, and should, delegate parsing of non-native formats to. One which aggressively sandboxes the worst of offenders. With support from the IPC system that tunes the parsing and consumes the results, it also becomes analysis, collection and fuzzing harness in one. Leveraging the display server to improve debugging.
Someone slightly more mischievous can then, run these delegates on single-purpose devices that network boot and reset to a steady state on power-cycle. Let them consume the hand-crafted and targeted phishing expedition, remote-forward a debugger as a crash collector to a team that extract and reverse exploit chain and payload, replicate-inject into a few honey pots with some prices for the threat actor to enjoy. This clip from pipeworld looks surprisingly much like part of that scenario, no?
Many data formats are becoming apt at embedded compromising metadata. Most know about EXIF today, fewer are aware just how much can be shoved into XMP – where you can find such delicacies as metadata on your motion (gait), or tracking images hidden inside the image as a base64 encoded jpeg inside the xmp block of a jpeg. A good rule of thumb is to never let anything touched by Adobe near your loved ones. If you would manage to systematically strip the one, something new is bound to pop up elsewhere.
By splitting importing (afsrv_decode) and (afsrv_encode) to very distinct tasks with a human observable representation and scriptable intermediary model the transition from the one to the other — you also naturally get designs that lets you define other metadata to encode. If that then is what gets forwarded and uploaded to whatever “information parasite” (social media as some tend to call it) that pretends to strip it away, but really collects it for future modelling, starts to trust it, well shucks, that degrades the value of the signal/selector. The point is not to “win”, the point is to make this kind of creepiness cost more than what you are worth so some are incentivised to try and make a more honest living.
Compartment per device.
With A12 comes the capability to transition back and forth between device-bound desktop computing and several networked forms. This opens up for a mentality of ‘one task well’ per discrete (‘galvanically isolated’) device, practically the strongest form of compartmenting risk that we can reasonably afford.
The best client to compartment on ‘throwaway devices’ is the web browser. The browser has dug its claws so deep into the flesh of the OS and hardware itself, and expose so much of it to web applications that the distance between that and downloading and running a standalone binary is tiny and getting shorter by the day — we just collectively managed to design a binary format that is somehow worse than ELF, a true feat if there ever was one.
The browser offer ample opportunity for persistence and lateral movement, yet itself aggregates so much sensitive and useful information that you rarely need to seek elsewhere on the system.
In these cases lateral movement as covered before is less interesting. Enough ‘gold’ exists within the browser processes that it is a comfortable target for your disk-less ephemeral process parasite to sit and scrape credentials and proxy through; ‘smash and grab’ as the kids say.
There is generally an overemphasis on memory safety to the point that it becomes the proverbial ‘finger pointing towards the moon’ and you miss out on all the other heavenly glory. There are enough great and fun vulnerabilities that require little of the contortionist practices of exploiting memory corruptions, and a few have been referenced already.
A category that has not been mentioned yet is micro-architectural attacks — one reason why the same piece of hardware is getting incrementally slower these days. You might have heard about these attacks through names indicative of movie villains and vaguely sexual sounding positions, e.g. SPECTRE and ROWHAMMER. Judging by various errata sections between CPU microcode revisions alone, there is a lingering scent in the air that we are far away from the end of this interesting journey.
Instead of handicapping ourselves further, assume that ‘process separation’ and similar forms of whole-system virtualization is useful for resiliency and for compatibility still, but not a fair security mechanism; sorry docker. Instead, split up the worst offenders over multiple devices that, again, are wiped and replaced on a regular basis. You now cost enough to exploit that a thug with a wrench is the cheaper solution.
At this juncture, we might as well also make it easier to extract state (snapshot/serialize) and re-inject into another instance of the same software on another device (restore/deserialize). In the end, it is a prerequisite for making the workflow transparent and quick enough that spinning up a new ephemeral browser tab should be near instant.
Another gain is that you reduce the amount of accidental state that accumulates, and you get the ability to inspect what that state entails. This is the story about how your filthy habits built up inside and between the tabs that you, for some reason, refuse to close. Think of the poor forensics examiner that has to sift through it — toss more of your data.
Anything that provides data, should be able to transition to producing noise.
Consider the microphone screenshot from earlier, or ‘screen’ sharing for that matter. What value does it have to your actions over abstracting away from the device?. Having a design language that is ‘provide what you want to share’ might look similar enough to a browser asking for permission to use your microphone or record your desktop — but there is quite some benefit to basic the user interaction to work at this other level of abstraction.
Some are strictly practically, like getting the user to think of ‘what’ to ‘present’ rather than to try and clean the desktop from any accidentally compromising material. Being explicit about source makes it much less likely that the iMessage notification popups from your loved one showing something embarrassing will appear in the middle of a zoom call with upper management.
By decoupling the media stream from the source (again, fsrv_decode and fsrv_encode), there is little stopping you from switching out or modifying the stream as it progress. While it can be used for cutesy effects such as adding googly eyes to everything without the application being directly the wiser — it also permits a semantically valid stream to slowly be morphed to and from a synthetic one.
With style transferring GANs improving, as well as adversarial machine learning for that matter, AI will no doubt push beyond that creepy point where AI synthetic versions of you will plausibly pass the Turing test to any colleagues and loved ones. This also implies that your cyber relations or verbal faux pas will be more plausibly deniable. You can end a conversation, and let the AI keep it going for a while. Let a few months pass and not even you will remember what you actually said. Taint the message history by inserting GPT3 stories in between. Ideally the cost for separating truth from nature’s lies will cost as much as dear old Science.
This is a building block for ‘Offensive Privacy’ — Just stay away from VR.
Now for something completely different. In the spiritual vein of One Night in Rio: Vacation photos from Plan9 and AWK for multimedia, here is a tool that is the link that ties almost all the projects within the Arcan umbrella together into one – and one we have been building towards for a depressing number of years and tens of thousands of hours.
The core is based around ‘cells’ of various types. These naturally tile in two dimensions as rows and columns. The first cell on each row determines the default behaviour of the row, and moving selection around will ‘pan’ the view to fit the current cell.
Each row has a scale factor, and the same goes for the cell. This means that different sets of cells can have different sizes (‘heterogenous zooming’) with different post-processing based on magnification and so on. Some cells can switch contents based on magnification, while others stay blissfully ignorant.
In the following short clip you see some of this in play, using a combination of keybindings as well as mouse gestures. This might make some feel a bit nauseous as this is not something that generalises well to groups of observers (we have solutions for that too) — but it is a different effect when it is your interaction that initiates and control the “zoom”.
The heterogeneous zooming allows for a large number of cells to be visible at the same time. Even when scaled down, client state such as notifications and alerts can still be communicated through decoration colours and animations.
Befitting of tiling workflows, everything can be keyboard controlled. Just like in Durden, mouse gestures, popup menus, keyboards and exotic input devices simply map to writes into file-system like paths.
When- or why- would tens to hundreds of simultaneous windows be useful? Some examples from my day to day would be monitoring, controlling and automating virtual machines, botnets remote shell connections, video surveillance, ticketing systems and so on.
Dataflow Computing and the Expression Cell
Each cell has a type – with the default one being ‘expression’, where you type in expressions in a minimalistic programming language. The result of that expression mutates the type- or the contents- of the cell.
Expressions are processed differently based on which scope they are operating in – with the current scopes being:
Expression – Arithmetic, string processing and other ‘basic types’ operations.
Cell – Used for event handlers, changing, annotating or otherwise modifying visual or window management behaviour of the current cell.
System – Global configuration, shutdown, idle timers and so on.
Factory – Producing new cells or mutating existing ones.
In the following clip you can see some basic arithmetic; changing numeric representation; string processing functions and the ‘cell’ scope as a popup to tag a cell with a reference identifier which is then used in another expression.
In a sense this is 3 different kinds of command-line interfaces wrapped into one. These can modify, import from- and export to- cells of other types. This is where things gets more interesting and powerful.
The following video is an example where I first copy the contents of a file to the clipboard of a terminal (the somewhat unknown OSC;52 sequence). I then create an expression cell where I set its contents to a text message. I then start vim in the first terminal and then run the cell-scope expression:
type_keyboard(concat(a1.clipboard, b1), “fast”)
This reads as ‘send the clipboard content of the A1 cell and the content A2 as simulated keypresses into the currently selected cell using a moderately fast typing speed model. From the perspective of vim itself, this looks just like someone typing on a keyboard.
There are many more cell types than just terminals, command lines and expressions. Video playback; capture devices; images; native Arcan clients, such as our Qemu backend or Xorg fork and many more. In this clip you can see some of those – including Pipeworld running pipeworld. You can also see the current state of autocompletion and expression helper.
In this video I first open a capture device (web camera). I then spawn a terminal and copy the contents of a shader (GPU processing program) to its clipboard. This gets compiled and assigned the name ‘red‘. Lastly I sample a frame from the capture device using this shader.
What all of this means practically is that we can gather measurements, trigger inputs and stitch audio, video and generic data streams from clients together in a fashion that more resembles breadboards as used in electronics, and sequencers as used in sound production, than traditional desktop or terminal work.
Terminal, CLI and Pipelines
Lets go back to the command-line for a bit, as we have yet to poke at pipelines. In this clip I run the system scope function: pipe(“find /usr”, “grep –line-buffered share”)
A lot of magic is happening inside of this one. Each client is spawned separately and suspended. Then their runtime stdin/stdout pipes gets remapped based on the current pipeline before the chain is activated. Note that when I reset the cell representing the “find /usr” cell, the grep one remains intact and unaware that find was actually killed off and re-executed.
The API is lacking somewhat still, but technically there is nothing blocking the ‘pipes’ from being a mix between terminal emulators (isatty() == true), stdin/stdout redirected normal processes (isatty() == false) and fully graphical ones.
Typing # into an expression cell gives us a dusty old terminal emulator. We can also add some command to run afterwards, like #find /tmp. ‘Resetting’ a cell means re-evaluating the expression or command it represents.
In the following clip I first list the contents of /tmp and then setup a timer that will reset the cell ~every second. This is indicated by the coloured flash. Then I spawn a normal terminal and create a file in tmp, and you can see it appearing. To show that the ‘terminal’ behind the first command is still alive, I also swap out its colour scheme and have it resize to fit its contents.
Typing ! into an expression cell switches it into a CLI one. It uses a special mode in afsrv_terminal (the terminal that comes with Arcan) where no terminal protocols are emulated, no ‘pty devices’ are needed, and the CLI is native. As such we are no longer restricted to running a shell like bash that hides (“premature composition”) the identity, state and inputs/outputs of its children and commands.
Note that each discrete command becomes its own window, and the cell itself dictates the layout and scale of new and old command outputs.
Individual commands can be re-executed, clients run simultaneously and states such as clipboards are kept separate. Clients can switch to being graphical or embed media elements; they can announce their default keybindings; handle dynamic open and save from cut/paste and drag/drop; runtime redirect stdio; they can spawn sub-windows that attach on the same logical row and much more.
Advanced Networking, Sharing and Debugging
In the following clip you can see how quickly I go from a cell with external content and a debugger attached to the process associated with it, with lots of other options for process manipulation and inspection. Protections like YAMA are kept enabled, yet there is no sudo gdb -p mess going on. To learn more about this, see the article on Leveraging the Display Server to improve Debugging.
Lets go deeper. In the following clip I have set ‘arcan-dbgcapture’ as the kernel ‘core_pattern’ handler. Whenever a process crashes, the collector grabs crash information and the underlying executable. These gets wrapped around a Textual-UI with some options of what to do with the information. I then add a ‘listen’ cell that exposes a connection point for this TUI (‘crash’ in the video). Anything that connects to this point gets added to the row that this cell controls. To show it off, I run a simple binary that just crashes outright a few times, and you can see how the collections appear.
Since it is built using arcan-shmif, these connection points can be routed over a network – so swapping that ‘crash’ point out with something like a12://my.machine, embed into your base VM image, distribute to your fuzzing cluster or CI and enjoy network transparent debugging.
Moving on to the networking path and streaming/sharing. The following clip shows me migrating a terminal and qemu cell from a workstation to my laptop via the A12 network protocol. Note that the cell colours and font size change automatically as it is the target a client is presenting on that controls the current look and feel.
(The slight delay for the qemu window is a bug in the qemu Arcan display backend that does not properly re-submit a frame on migration, so it does not appear until the next time the guest updated its output).
Almost all cells can have an ‘encoder’ attached to them. This is simply a one-way composition transform that will convert the contents to some other format. A very practical example is recording to a video file or RTMP host, or something more obscure like OCR for pixels to text conversion.
In the following clip I set a cell encoder that act as a VNC server and then connect to that using the built-in VNC client that is part of Mac OS X. I then destroy the encoder and assigns a new one to a terminal. The OS X VNC viewer reconnects automatically, and you can see that input also works over the return path.
A cell type that blends well with this is ‘composition’ which puts the contents of multiple cells together using some layouting algorithm. The following clip shows that being used.
Attach an encoder to the composition cell and you have compartmented partial desktop sharing, another potent building block for interesting collaboration tools.
Trajectory and Future
Pipeworld will join Safespaces in acting as the main requirement ‘driver’ in improving Arcan and evolving its set of features, while Durden takes the backseat and moves more towards stabilisation.
These projects are not entirely disjunct. Pipewold has been written in such a way that the dataflow and window management can be integrated as tools in these two other environments so that you can mix and match – have Pipeworld be a pulldown HUD in Durden or 360 degrees programmable layers in Safespaces with 3D data actually looking that way.
The analysis and statistics tools that are part of Senseye will join in here, along with other security/reverse engineering projects I have around here.
Accessibility will be one major target for this project. The zoomable nature helps a bit, but much more interesting is the data-oriented workflow; with it comes the ability to logically address / route and treat clients as multi-representation interactive ‘data sources’ with typed inputs and outputs rather than mere opaque box-trees with prematurely composed (mixed contents) pixels and rigid ‘drag and drop’ as main data exchange.
Another major target is collaboration. Since we can dynamically redirect, splice, compose and transform clients in a network friendly way, new collaboration tools emerge organically from the pieces that are already present.
Where we need much more work is at the edges of client and device compatibility, i.e. modify the bridge tools to provide translations to non-native clients. A direct and simple example is taking our Xorg fork, Xarcan, and intercept ‘screen reading’ requests and substitute for whatever we route to it at the moment – as well as exposing composed cell output as capture devices over v4l2-loopback and so on.
I can editorialise about this for hours, and although the clips here show some of what is already in here, there is so much more already existing and — much more to be done.
Hot on the heels of the recent Arcan release, it is also time for a release to our reference desktop environment ‘Durden‘.
To refresh memory somewhat, the closest valid comparison is probably the venerable AwesomeWM – but there is quite a lot more to it in Durden. It has been my daily driver for around 5 years now and implements all popular window management styles, and some unique ones.
During this time, it has been used to drive development of Arcan itself — but that will start to wind down now as there is little need for more major features. Instead updates will mainly be improvements to the existing ones before we can safely go 1.0. The two other projects, Safespaces and [Undisclosed], will take its place in helping the engine reach new heights.
Durden has an unusual way of putting a desktop together where everything is structured as a file-system and reconfigurable at run-time with results immediately visible:
This filesystem can be accessed through build-in HUD and popup menus, as well as externally controlled through a unix socket. Through the arcan-cfgfs tool it can even be mounted as a FUSE-file system.
All higher level UI primitives — no matter if it is decoration buttons, keybindings, statusbar and so on are simply references paths in this file system. There is over 600 such paths at the moment.
You can easily extend or ‘slim it down’ by adding or removing ‘tools’ scripts and ‘widgets’ scripts (for the HUD).
We are at the point where many traditional X window managers styles can be transferred through scripts that translate your old dot files into paths that can be run as schemes (atomic set of paths).
If you are interested in helping out with developing such translation scripts, get in contact.
Here is a link to the full Changelog. The rest of the post will describe some of the major changes but since most of them are not particularly visual, videos will be used more sparingly this time around.
Core / Input
Shutdown has received a new important option, ‘silent’. This will still cause a shutdown of Durden but the clients will treat it as if the connection has crashed and enter a sleep-reconnect cycle. Whenever you start Durden again, the clients should come back as if nothing happened.
Touch – This layer has been refactored to make it much easier to add or modify classifiers. New classifiers have been added for (relative mouse, touch-fit and touch-scaled).
More controls have been added to the touchscreen and trackpad input handler through /global/input/touch for runtime tuning, but many more controls still exist as part of the device profiles (see devmaps/touch/…). Touch device profiles have received more options and bindable gestures, as well as different handling for ‘enter n-finger drag’ versus ‘step n-finger drag’.
Rotary – Through /global/input/rotary and devmaps/rotary/… you can now control binding and mapping for devices like the Surface Dial and GriffinPower Mate. These are great alternatives to the mouse wheel for actions like scrolling. This tool also has basic experimental support for 3D mice, though those are much more complex.
This is not strictly limited to this narrow class of devices. The tool can be used to provide any display+input device pair to run as a ‘minimap’ screen, or integrate with wm- defined properties like titlebar buttons (in the video), client announced custom keys, workspace and window miniatures, custom bindings and so on.
Popup – This tool allows you to spawn either a custom menu (defined in the devmaps/menus folder) or any subtree of the normal file-system as a popup. Either tied to another UI component or relative to the current known cursor position. The video below shows how they are mapped to parts of the statusbar and titlebar.
Todo – This is a simple task tracker / helper that can be found under /global/tools/todo. You set a task group, add some tasks with a shortname and description and activate. A task is picked at random and appears as a statusbar button. Click on it to postpone or mark for completion and you get a new one, and so on. The best use of this tool is integration with your other ticketing systems if you, like me, work on many projects and sometimes have a tough time choosing what to focus on.
Tracing – This is simply a debug assistant that will merge the engine tracing facility in Arcan with all the logging going on in Durden into one coherent timeline. Saves to the about:// trace format in Chrome or used through a converter with the much better Tracy.
Extbtn – The ‘external button’ tool was also covered in part by another article (Another low-level Arcan client: A tray icon handler). It allows you to map external clients as icon providers into the statusbar tray. This is also a way to mix the status bar contents with multiple external ‘bar-tools’. The tool inside the Arcan source repository has some helper scripts for doing so using the lemon-bar protocol.
This video is from that article, showing off attaching a terminal emulator as well as Xorg with wmaker as a tray icon popup.
Autostart –This tool allows you to define, view and modify paths that will run on startup or reset.
HUD – The built-in HUD file browser now triggers on clients (on user input) requests universal save/load, and sorting / searching received a ‘fuzzy matching’ mode (thanks to Cipharus). By typing ‘%’ into the HUD command line you can switch sorting mode, where fuzzy_relevance is a new entry.
Colour-Picker Widget – HUD settings that request an input colour, such as setting the background to a fix colour, will now popup several different palettes or reference image to pick from. In the video below you can see it being used to set a single colour wallpaper.
Statusbar – The popup tool now gets mapped to custom menus for workspace- switching buttons when right-clicking. This also applies to window titlebar. Display buttons are now dynamically added on display hotplug for quicker switching or migrating windows by drag-and-drop.
If window titlebars are marked as hidden due to client-side decorations, or default- setting them to off, they can now be set to ‘merge’ into the center-fill area of the statusbar instead – saving precious vertical space.
Mouse controls have been added to the tiling window management modes. This means that you can re-organise / swap / migrate and so on by drag-and-dropping, with a live preview as to where in the hierarchy they will attach. The video below shows that in action:
A ‘column-‘ tab mode has been added that dedicates a side column for each ‘window as tab’.
The previous ‘CLI group’ mode of handling new connections have been re-written in favour of a new swallowing mode (where a window can share the hierarchy spot of another, swapable). It is more robust than the previous tactic, but currently more limited when dealing with multiple clients.
The option for controlling which workspace new clients spawn on has also been added, along with the (tiling) option of having subwindows spawn as children to their parent window.
Visual / Accessibility
There is now a shared core for caching, loading and generating non-textual icons. Before all icons were picked from a font file, but can now be picked from pre-rastered sources or shaders, and then resampled for the target display output density it is being used on.
The global/displays/<display or current> path has received a ‘zoom’ group that allows you to set or step per-screen magnification around a specific region or relative to the cursor.
Multiple UI components now have optional shadow controls. /global/settings/visual/shadow allows for enabling/disabling the effect, adjusting size, colour and intensity and so on.
The flair – tool has received an effect category for window selection, with a ‘shake’ or ‘flash’ option included.
An ‘invert light’ (color preserving) shader has been added to the collection that can be applied per window, see target/video/shaders.
After that article, there was only one (and a half) real feature left to safely claim parity and that can be covered rather quickly. Thereafter we can nibble on the bites that are in Arcan, but not in Xorg — the reason for the difference in scope is best saved for a different time, although it is a good one.
First, let us not forget that there are more vectors for qualities that are significant to users than just features. Client compatibility is something that has been much lower on the list of priorities, yet is an important quality.
The reason is that prematurely adding support for something like a new display server backend to a toolkit, game engine or windowing library without both necessary and sufficient features in place will lead to a scattered actual feature set. There will be theoretical features, and then the features some clients actually might use some version or interpretation of. These two sets will slip further and further apart unless each affected project has exceptionally alert developers, and the reference implementation having basic hygiene in place regarding conformance verification and validation tools.
We have experimented with such implementations, of course, but that was mainly to test the design and current implementation; to find out where our APIs gets annoying to use, so that nobody else suffers.
The other reason for downplaying compatibility is that client applications has strong biases on how things are “supposed” to work, with the worst offenders being just ‘GUI toolkits’. Small changes can have large impact on application design and you are forced to adapt a certain worldview that lessens your perspective. As a trivial thought exercise, just consider something like a smartphone in a world where touch input was not invented, but eye tracking was flawless.
To re-emphasise the point of the Arcan project as such. It is not, and never was, to reinvent X. The principle program (12 principles for a diverging desktop future) say as much. The project is about showing that there were other paths not taken by the big players, and hint at what those paths offered when followed to their logical conclusion.
Most of the lessons in here are not amazing discoveries from diligent research. Rather, they are uncovering wisdom hidden inside the circuits of elderly arcade machines. It is the reason for header-bar image here still being a cropped PCB from that era.
Enough editorialising, towards the Xorgcism! This will again be painfully long and dry. Appy-polly-loggies in advance for that. Quicknav links are as follows:
The main missing feature was about network support. Luckily I do not have to cover that again — there are already two articles on the topic: one that digs into the situation in Xorg; one that covers our solution. If there is anything left to re-emphasise, it would be that the networking part is a completely separate process that reinterprets the networked client data.
There are three reasons for this – this first is that optimal local rendering and optimal remote rendering are incompatible ideas that were historically similar enough that the discrepancy could be tolerated. Nowadays we want both and that takes a different architecture where everyone involved adapts to which one of the two that is currently desired.
The second is for security; the client IPC is a privilege boundary. Network access is yet another privilege boundary. Mixing multiple privilege boundaries within the same process practically merges the two into one in this case, greatly reducing security. Even if you have a separate network process, should all it is doing be proxy-patching packets, you gain very little actual protection. This is because your ‘network process’ is just yet another router or proxy on the way to the real parser in the privileged process, and the only primitive often needed (writing to a file-descriptor) is hard or impossible to ‘sandbox’.
The third is for resilience. Recall the lengths we have previously gone to in order to have crash resilience. There has been much more work done in that direction since the linked article, and one of the final stages is to make sure client state can gracefully survive GPU loss and, should the circumstances permit, machine loss. These things do not work cleanly as an after-thought – the consequences permeate everything.
The other feature left was the fabled drawing API – to which I would also consider font management — as that lies somewhere in the middle; fonts were already covered in the previous article as part of the section on ‘mixed-DPI’ drawing.
The thing is, Arcan started with a drawing API – initially for specifying interactive animated presentations, documents and some ehrm, darker things of which only beer and cocktails in shady bars might, briefly, bring back to the surface. The display server angle turned into a detour and necessary evil after certain actors could not ‘leave well-enough alone’. All the projects shown here over the years have been using this API and been driving its development in lieu of a requirements specification / waterfall like design phase.
Anyhow, the ‘animated presentations’ bit brought in all kinds of input devices, as well as audio and video. Video parsing and decoding is a finicky process at the best of times, and the typical libraries for this were obvious exploitation vectors. This meant that process separation with least-privilege was clearly needed. With that you get a realtime/media IPC system, so clean that up, add buttons and there is your basic display server.
As a side-note: Having a ‘drawing API’ on the consumer side is not hard per se and, in fact, is the strategy that won, with local raster being a distant fallback. If you want accelerated rendering through OpenGL or Vulkan, it ends up being command buffers (hey, draw commands!) sent to the GPU that will actually turn it into a signal and/or pixels – the drawing won’t happen in-process through the grace of putting pixels in a buffer. You could add non-accelerated drawing to Wayland with a one line addition to wl_shm defining a buffer format as SVG, wouldn’t that be interesting? ( ͡° ͜ʖ ͡°)
J’accuse! – I hear you zoladats triumphantly proclaim. That is not for the clients touse now is it‽. So lwa-ts me explain the two sides to this story. A longer version with charts and code and everything can be found in the article on AWK for Multimedia.
The big bit about a drawing API is the level of abstraction it provides. If it is at the low level of “putPixel()”, you lose — the framebuffer is dead, long live the framebuffer. If it is about “bind set of uniforms, buffers, gpu program and supply commands” you get what the high end graphics libraries already provide. The real challenge is finding the right set of mid-level functions that provide something more without getting morbidly obese to the scale of whatever flavour ‘absorb all’ UI toolkit that is in fashion these days; to understand how, where and when the JS/HTML5/CSS/… soup fail.
Thus, a network friendly, compact, mid-level drawing++ related API is the target. One that can be used out-of-process and as an “intermediate language” (think LLVM-IR) for higher level UI components or document interpreters to render to, while at the same time having the feature set expected of ‘web-app’ like targets.
The “++” represents the process control, audio and meta-IPC routing that is also necessary to cover the span of what is expected of ‘a desktop’. The processing of these commands translate to another intermediate layer, ‘AGP’ which is a simplified subset of OpenGL that was needed as a decoupling mechanism to eventually switch to-/provide- Vulkan and Software rasterisation. It aims at at a network friendly level at about the quality of DOOM3/WebGL, not great, but thereafter video compression starts to win out.
The following figure illustrates the different ways of running Arcan to cover all combinations of ‘drawing API’:
So Arcan can be a client to itself, although it is a separate binary (arcan_lwa – (Lightweight-application) marked as ArcanLW in the figure.
ArcanLW will connect to Arcan ‘as a display server’ and act like a normal client. The side benefit here is that if you know how to write an app with the one, you can write or patch the window manager in the other. You get one API to control, manipulate or inspect all the routing for the entirety of your desktop. It privileges separates by design, it has been stable for years.
The last piece to this puzzle is to be able to provide RPC like access to this API. Here is why Lua matters more.
For those not in the know, exposing a C API to the Lua VM is about as simple as it gets and it fits other low level calling conventions; register the symbol name and function pointer of the function in the VM namespace; pop arguments from the stack based on the desired types; process; push the function return to the stack.
It can get more involved than this, but not by much. That actually doubles as a protocol decoder, thus it can be used as a decoupling mechanism. Define a bytecode and there is your \out of process\ drawing/WM API to bind to other languages.
Migration and Recovery
While we did brush upon crash recovery in the previous post, as well as a previous article on the subject – there is still a lot to it. Enough work has been done since to warrant yet another article.
People aware of some GTK history knows that it actually has the underused ability to jump between X Displays. The ‘underused’ part can probably be blamed on the, ehrm, ergonomics of spinning up and maintaining multiple X servers to jump between. Politely put, it was not for the faint of heart.
It is a good idea in principle, although there is not much benefit to do it on the toolkit level versus at the IPC level and, as it turns out, much easier — the feature will always apply regardless if the toolkit had enough foresight and resources to implement it- or not.
In X, if a window manager crashes — it is not that big of a deal since a new window-manager-as-client can reconnect and start rebuilding its view of the world. In reality it is rather janky and there are possible left-overs when the window manager additions to the scene graph in Xorg was of a more complex nature.
To learn from this, we come back to resource allocation in Arcan: Each clients gets a primary segment (window++) and then dynamically renegotiates for more (with rejection being the default). The window manager can kill any and all secondary allocations and ‘reset’ the primary — forcing it to try and renegotiate new resources.
This means you have one guaranteed ‘window’ and the rest that can come and go. This is partly to allow for much crazier window management schemes, while at the same time not losing the more simple ones. A client developer gets a known and guaranteed set of available features to serve as a baseline, with sets of extras to add- or react- to.
The WM also gets access to a key/value store for tracking WM specific information about its clients. Both clients and WM gets GUIDs for repairing the clients to this information so things re-appear where expected. It also gets the ability to dynamicallyredirect clients to other display servers – as well as inform about alternate ones should the present one disappear.
The WM also gets a specific event entrypoint, ‘adopt’ which references clients of unknown origin. This allows controlled handover between WMs for live-switching only relevant clients, as well as a recovery mechanism for the WM to recovery from errors in itself.
This brings us to a handful of possible error scenarios and recoveries:
Live-lock in WM →Watchdog process sends signal →(same as 1.)
Crash in Engine → Client-side IPC library detects, connect loop (same or new address) →injects ‘RESET’ event.
Live-lock in Engine → Watchdog process sends kill →3.
The extra twist is that the WM can ‘fake’ a crash for one or many clients, so that the procedure from 3 can be triggered on command. It can also send a new address for the client to connect to. This opens up for dynamically migrating clients between local and remote servers. The short clip below shows this going from local to remote:
While the internals for how all this works is quite complicated — both the WM and the client facing code is quite trivial.
A lesson to learn from XTEST, XRANDR and the tools that take advantage of these extensions- is that no matter how well you do your requirements engineering, people will disagree with your world view, and sometimes loudly so. Some just need their ‘hacks’ for the sake of it — the act of hacking and the feeling of agency it brings has value that should not be dismissed outright.
Is it possible to provide an outlet for such expressions without painting yourself into a design corner? The answer here are ‘hook scripts’. To understand how and why, we need something of a schematic.
This provides two possible paths for intercepting and manipulating WM behaviour. In the pre- stage the API that the WM sees can be intercepted and patched before the WM scripts themselves have a chance to run.
The other happens after the initialisation entry point but before the actual event loop starts. This allows you to intercept the event handlers themselves and both pre- and post- hook those.
When combined these let you ‘monkey patch’ any part of the WMs view of things, and work around the “stuff-you-don’t-like”. It can also be used to inject features that were not allowed as part of Arcan itself, for whatever reason.
A practical example is how to add external input drivers:
This opens up a single-use connection point (extio_1) and additional ones (_2, _3, …) for each time the hook script is added. Then your external input driver is set to connect to that name, whereby ‘myapp’ sees the inputs as coming from the engine itself.
This is yet another point where the Lua tactic shows its strength — if hundreds of thousands of beginner- to intermediate- level developers managed to do it for World of Warcraft and lots of games thereafter; anyone who can juggle things as clumsy as bourne shell scripts around should be able to ‘scratch their itch’ with ease.
Alternate Representations and Accessibility
Now we are at the more unique stuff. In most other display systems, a client performs some version of the following:
Allocate a window.
Tie it to a pixel buffer.
Signal when it is ready to be used.
Add the ability to control parts of its position and composition order and you are surprisingly close to done — the ‘odd man out’ is practically Wayland where even this is convoluted to the extreme.
Arcan has a different view on clients. While the clients can perform the sequence above, the WM can also tell the client to have a new window — and know if the client accepted it or not.
This was brushed upon in the article on ‘Leveraging the Display Server to improve debugging‘ — go there for the longer explanation. The short form is that if we add the notion of more abstract types to our windows, like ‘Debug’ or ‘Accessibility’ and so on – it implies asking the client ‘can you provide such a view of yourself’.
Thus there is no need for a separate ‘input-accessibility-bus’ or complexities of that nature — it is all about re-use of existing mechanisms to form new features.
The obvious normal case for a feature like this is ‘screen sharing’ – there is strictly no need for the client to be able to enumerate resources and displays in order to implement the sharing and collaborations features; it should not need to know how the sausage gets made.
The rest of the desktop is perfectly capable of providing the UI language to annoy a user for such details, or to automatically map and provide the contextually relevant subset. The role of the sharing client should be to receive and serve, not to define and control.
Application Launching and Handover allocation
Arcan can act as a launcher for applications. As such, it knows if an application dies, and will respond to that. This is done in order to improve security, as well as to be able to provide control flows that are common to gaming consoles, smart phones, set-top boxes and so on.
Most of this feature did not come from the strict idea of acting as a ‘launching frontend’ but as a side effect of wanting to be able to privilege-separate any parsing of untrusted contents. This practically means spawning new limited processes that can send the results back, and controlling their life-cycle.
With the launcher path, it means that a client does not ‘connect’ as an external source, but rather inherits its IPC connection as part of being run. In fact, locking down the ‘external connect’ path is a single ‘Hook Script’ that blocks out the ‘target_alloc(str)’ function.
What it affords us is client identity and authenticity. This is a hard problem to implement as an afterthought with the primitives that a POSIX-style userspace provides you, and ultimately it ends up failing spectacularly or juggling cryptographic primitives and failing subtly; the chain of trust is broken.
It also affords us a client-unique key-value store for the metadata config that is needed to ‘remember’ what some window related to a client did – across crashes, reboots and restarts. It also lets us inform our scheduler, resource allocation and the likes to treat these clients special, in order to make ‘fork-bomb’ like Denial of Service attacks much harder.
The twist to this is ‘handover allocation’. Clients themselves may want to repeat the cycle by spawning new processes, as part of the intended feature set, as well as privilege separation / sandboxing of its own. The alternative would be ‘premature composition’ where the client becomes a display server of its own, which is a bad idea.
The following figure illustrates both the ‘launcher’ path as well as the ‘handover’ form:
The user defines the set of allowed targets that ‘MyWM’ can execute (binary + arguments + metadata) through the ‘arcan_db’ tool and the ‘add_target’ command shown at the bottom of the figure.
MyWM refers to this by calling the launch_target builtin function, which sets resources up and spawns the new process with ‘mybin’ running based on what is in the database as ‘thething’. The code in mybin decides that it wants a new window to delegate something, say embedded privilege separated video playback. It requests a new window of the special handover type.
MyWM is friendly enough and ‘accepts’ by calling accept_target as part of the eventloop tied from the launch_target call. New primitives are allocated and pushed to mybin. Mybin launches the new program ‘somebin’ which inherits and opens the connection to Arcan the same way ‘mybin’ did. At the same time, mybin gets an abstract cookie that can be used to define and reposition ‘somebin’ within its own coordinate system and scene graph subgraph. X style reparenting, but without most of the chaos.
A (not) so big gamble on client compatibility is that for the future, we really need to improve how virtual machines (a class of applications that include ‘terminals’, emulators and ‘modern’ browsers) integrate with the grander system.
All the big actors are moving in this ominous direction as a preface to ‘clouding’ these virtual machines and then keeping things there.
One of the special needs of virtual machine class clients, is that of state controls – suspend, resume, snapshot and restore.
Arcan is ahead of the curve there since we used emulators as the reference model for the ‘needs’ of such a client from the start. They have been leveraged for testing performance, throughput and latency ever since – emulation has hard real-time requirements after all, and a whole bunch of free test cases in the shape of ‘Tool-Assisted Speedruns’. Another such thing is that ‘state controls’ are a natural thing in well-written emulators.
In the Arcan API, this boils down to trivial commands in the WM layer – something like:
Is literally enough to tell one client to pack up its state and forward it to client_b, and to tell client b to restore from the incoming state – even if client_b is networked.
This also means that the ‘sharing’ window feature can be combined with the ‘alternate representation’ feature to fake new devices inside of the virtual machine. Something like the following:
Would create a composition (of src1, src2) and inject into (vid_client_a) that can then present it as a ‘camera’ or ‘screen sharing’ and so on.
This might come as a surprise to some, but audio and video processing are really quite similar. If you only do graphics and have not studied audio, or the other way around, you are missing out. It might take a course or two on ‘Digital Signal Processing’ or something to that effect to not hit any grand barriers, but that should be part of basic computing hygiene by now.
Not only do they have a large overlap in the theoretical foundation, but their internal structure from an operating system position does as well. Separating the two is something that has increased systemic complexity by a fair amount. Instead of using the same IPC primitives; transfer modes; device navigation; security; hotplug notification; output routing and redirection; we traditionally maintain two or more ecosystems for doing 90% of the same thing – even though the HDMI cables themselves tell us to stop.
What then follows is that a lot of applications need to combine the two in a coordinated and synchronised way. So instead of these two domains work over the same IPC system that has done the painfully heavy lifting, a third is added (Binder, D-Bus and so on), increasing systemic complexity by magnitudes.
While the feature itself is in a more primitive stage, much of it piggybacks from the work done on video and input latency – so when those stages are closer to complete the last bits of audio will get filled in as well. Again similarly to graphics where we can’t realistically aspire to be the next ‘Unreal Engine’, the role of audio support here is not to try replacing ‘pro’ audio but to fill out the ‘simple to middle’ level with coherent controls that meshes well with desktop style coordination – and wrap it in controls that can delegate to ‘pro’ audio consumers, much like we handle features like colour management.
This becomes especially important for the networked desktop use case, for accessibility (the ‘screen reader’) as well as for VR and AR experiences where positional audio suddenly becomes one of the most powerful tools available.
Exotic Input and VR
This is another chapter both for the future, for the present and for improving the past. The previous article mostly skimmed over the input model, and as with the other features — there is a lot of nuance to it.
The input model in Arcan covers both the expected basics (mouse, touch and ‘translated’ – i.e. keyboards) as well as game (‘assistive’) devices, eye trackers and similar sensors. This goes all the way to weird hybrid-I/O (see Interfacing with a stream deck). It is possible to safely and securely attach external input devices as shown in the section on hook scripts.
Two other “input” parts are client-announced input labels and VR.
Starting with the client-announced input labels – any client can announce a set of active input labels. These are short high level tags e.g. “Copy_Clipboard” with some kind of description in the current language settings, “Copy the word at the cursor”, some nature of its type (analog or digital) as well as a suggested default keybinding (Ctrl+C), if digital.
The WM can leverage this feature to provide a coherent binding interface, work around conflicts with existing bindings from other clients or the WM itself. Subsequently it can be used to implement your ‘hotkey’ manager (best done as a hook script) without overreaching and giving it access to anything other than what is strictly necessary.
The WM simply sets the appropriate label ‘tag’ to the next input event that gets routed to a client. This also meshes with other accessibility features, such as routing through a voice synthesizer to say that you pressed ‘copy clipboard’ rather than ctrl-down-C and then guessing whatever that meant in the context of the active application.
VR is an interesting example of a complex input model with a short window of opportunity; the ultimate avatar matches your every muscle at high sample rates.
This is a complex operation that aggregates sensors from multiple timing sources, including arrays of video cameras and laser projects. It does not mesh well with the traditional event loop structure as queues gets saturated instantly.
The span between ‘present’ to ‘possible’ inputs is also much wider – not everyone should need a full body suit, or have all fingers and toes present, and the sample streams themselves are sensitive as a number of strong biometrics can be extracted.
The following diagram tries to take a stab at explaining how the layers mesh together to provide VR composition and input management.
Akin to launch_target mentioned previously, there is a function for launch_avfeed available to the WM for launching the ‘decode’ frameserver. This provides a privilege separated producer process that provides a single input feed. Internally it works like any old client would, but since we have a well defined role and interface for it, the consequences for forcibly killing/restarting it are predictable.
It can be used to retrieve video feeds from webcams, and the file browser in Durden uses it for live previews of images and video. For VR, it allows us to access and map the many possible tracking cameras (inside out, outside in, hands, …) then compose/embed (AR), filter and perform computer vision to improve position tracking and so on.
For device control itself, we have yet another process/binary that is mapped to the function vr_setup. This one provides display metadata about the display panel and lens configuration, as well as hotplug events of ‘limbs’ (slots matched to the average human anatomy). The WM decides which ‘limbs’ map to which objects in 3d space, and how or if they should be forwarded to any clients.
The sample streams will not be processed as a normal input event, that would be too costly. Instead they continuously update designated locations in shared memory, along with timestamps and checksums – based on the currently set limb-map.
A client connects and identifies its primary segment as the L eye, the R eye or LR in a side-by-side configuration. For the L and R cases, a secondary segment is then allocated through the primary, to cover for the missing eye. It requests an extended mapping for VR data transfer (similarly to how colour management transfer its metadata back and forth).
A Path from Curses – TUI
On top of the mid-high level drawing API and the low-level client API we have in the client facing side of the ‘SHMIF’ IPC system, there is also a separate client API for text dominant clients. This API is built on top of SHMIF, and exposes a stable subset of its features.
The included terminal emulator, afsrv_terminal, is built using this API – and it has a basic alternate shell that completely skip terminal protocols and pseudo-terminals.
The API provides basic window segmentation and positioning that integrates with the window manager of the desktop, and similar integration features specifically for command line so that the WM can decide how to present and interpret completions and suggestions. It can trigger alerts and notifications, as well as announce inputs in a similar fashion as described in the ‘exotic input and VR’ category.
The output buffer format, tpack, has been mentioned before. It enables server-side rendering, both for efficient network remoting (since SHMIF is involved, the network transparency parts mentioned earlier applies). It is also for optimal sharing of font caches and resolved glyphs (and GPU texture atlases) transparently between clients, allowing for zero overdraw and minimal input to output latency.
Colors and palette can be defined and redefined by your window manager; TUI clients themselves then have the option of 24-bit colours that they pick themselves – but also the preferred option of semantic labels that map to the active palette. This means a way out of the terminal-legacy “Red is now Green” problem of specifying colours.
Content synchronisation is always explicit, so that no window should have tearing or draw output that will be replaced miliseconds later, or waste a vsync on an incomplete update.
All text is unicode and the clipboard has a side channel so large paste operations do not block or otherwise interfere with stdin/stdout. Locale (input/output) language can be provided, and a getopt- like replacement for input arguments that are packed over an environment variable in order to not complicate legacy argc/argv further.
Attention alerts can be sent, as well as notifications. There are controls for describing content length and byte-precise seeking, as well as job progress status.
Speaking of stdin/stdout/stderr – those can be redirected dynamically by the server. Additional input/output pipes can be dynamically announced and added. As part of that feature a TUI client can also announce support for input and output types, both as immediate requests or capability hints for universal open/save and file browser integration.
There are a few basic primitives to get started. Examples of such primitives are readline, buffer view (everyone deserves a hex editor) and list view, but the core will not go deeper than that. The focus, instead is on providing discoverable command line interfaces and one-off commands that focus on processing and providing data for data flow graphs instead of pipelines.
This time around, the changes are big enough across the board that the sub-projects will get individual posts instead of being clumped together, and that will become a recurring theme as the progress cadence becomes less and less interlocked.
We also have a sister blog at www.divergent-desktop.org that will slowly cover higher level design philosophy, rants and reasoning behind some of what is being done here. A few observant ones have pieced together the puzzle — but most have not.
This release is a thematic shift from low level graphics plumbing to the network transparency related code. We will still make and accept patches, changes and features to the lower video layers, of course — ‘Moby Blit’ is still out there — but focus will be elsewhere. Hopefully this will be one of the last time these massive releases make sense, and we can tick on a (bi-)monthly basis for a while.
As such, it is dedicated to the memory of a fellow explorer and dear friend, Morten, whose insights have been instrumental to the work here and elsewhere. A taste of the depths of such contributions are well illustrated by the ‘Farewell’ post over at Mamedev. Thanks for everything buddy :'(.
If all of this isn’t enough, there is much more in the Changelog.
This release brings us the first working version of ‘arcan-net’. This is an arcan-shmif server that translates clients running locally to another instance across the network. It has been in use for a while, but don’t trust it (yet) in tougher security compartments than you would VNC/RDB.
Not only that, but it is a lot more powerful than the corresponding form in X. It uses a protocol we are lovingly calling ‘A12’ — It is not the X12 some people wanted, but at least it is a twelve. It was also the last missing piece before being able to safely claim to be far beyond feature parity with Xorg, so expect the comparison article series to get its next instalment soon enough. With that it is also time to start chasing compatibility (client support) and performance over features.
Single Client Forwarding (‘X11’-like), Desktop Forwarding (‘RFB/RDP/SPICE’-like)
Remote Input and Meta-data (‘Synergy’- style)
Audio, Video and Binary Transfers
Basic blob- caching
Lossless and Lossy video compression, defaults based on window types
Interactive adjustment of compression parameters
X25519 + Chacha8 + Blake3 for authenticated encryption
Crude Privilege separation
This meshes really well with other arcan-shmif features, such as the one on external input drivers mentioned below. The video below shows a single nugget from how ‘transparent’ things can be, live migration, drag and drop:
This is yet another reason as to why ‘crash resilience‘ matters – if you solve the one, you get a lot more interesting things in return.
On-Demand Client Debugging
When the WM initiates window creation, it can now push windows of a debug type. Should the client cooperate, the first time that happens the client will still get a new window to provide its debug data through.
Should the client fail to comply, or the WM explicitly asks it, a debug interface built into the arcan-shmif library will now silently activate, allowing the horror movie scenario of ‘the call is coming from inside the house’ to unfold.
The default distribution now includes the ‘KMScon- like’ ‘console’ window manager, though it also accepts graphical clients, so if all you need is a “kiosk like” one-application runner per virtual ‘terminal’ (or the template for one), this fits that bill.
it has since been extended some, getting support for an OSD keyboard and basic wayland client support as well via the script below. The video below shows it and the keyboard working on a Pinephone:
If you have an unrequited love for minimalism confused as simplicity (minimalism doesn’t have room for love, being simply so minimal), this is the tradeoff path that will get you there, eventually.
Builtins, Hook Script Chaining
There are multiple facilities for encouraging opt-in code reuse between applications and window managers, both for developers and advanced end users. These are provided in the ‘system-scripts’ namespace and consume the folder names ‘builtin’ and ‘hooks’. The idea is to let the heavier subprojects like durden act as incubators, and the implementations gets cleaned up and re-usable as such shared builtin scripts.
This time around, the builtins have been extended with:
builtin/decorator.lua – for adding borders and mouse controls for border manipulation to a surface
builtin/wayland.lua – for adding some of the special considerations that wayland clients require
builtin/osdkbd.lua – for building custom on-screen display keyboards
This means that the WM side of thing for adding Wayland and XWayland support to a project boils down to about 2/3rds of this script that is used for testing. API is stable and supported, has been for a while, will remain as such.
The engine facility for ‘hook scripts’ that are run transparently to the application itself have been improved to allow chaining multiple ones together – greatly improving their utility (actually make them useful).
Previously, these were limited to a single one, which was less than ideal. Now that multiple ones can be chained together, many features that can be implemented in a window manager agnostic way can be shared between all of them.
This is the safer, more secure and much more powerful way to get all the ‘xdotool/xrandr/…’ kind of hacks going without needlessly exposing them over some other IPC.
Some are primarily used for testing, like this auto-shutdown one:
arcan -H hooks/shutdown.lua durden shutdown=200
Regardless of the other scripts running, this will issue a shutdown sequence after 200 logical ticks on a 25Hz timer. Since these scripts can simulate and inject other tricky event paths, like display and input hotplug and playback, they are ideal for a window manager testing corpus as the rest of the system can run headless.
Others are more interesting, such as ‘hooks/external_input.lua’ which opens up dedicated single-client connection points that are allowed to inject input, and only that. This is also used to extend the engine with input drivers that would not be allowed into the main engine due to their dependencies or licenses.
This can even be combined with the network proxy mentioned above to allow, for instance, sharing of input devices from other machines (which would enable Synergy- like virtual KMS workflows) and proprietary drivers running inside of containers.
One external driver example is the Tobii 4C eye tracker driver. Even though their hardware is cool, it is still a company that are about as 90ies terrible and backwards with their software ecosystem as can be; shrinkwrap licenses restricting allowed use cases of the sensor data (*facepalm*), closed source drivers, firmware clamped hardware capability and little in terms of serious competition. This is just depressing giving the potential eye tracking actually has outside of vile adtech. There used to be a driver available on their website; it was pulled in a move that can only be assumed to be about further fscking their customers over. A few searches should land you with repositories that mirrored the .debs.
A rough Arcan input driver as part of the repository here, and some docker container to reduce the taint here. Inject the blobs from the driver and experiment with eye tracking inside of Arcan. Other exotic input and LED output clients will also be exposed through that repository rather than the main one.
As covered in the article on ‘Interfacing with a stream deck device, there is also a support tool for ElGato “streamdeck” like devices added that lets you map window contents, ui controls, custom macros etc. unto the device, though the real meat of this features comes with the next release of Durden.
The following clip is from that feature, recorded in an AirBnB somewhere in Shinjuku:
You can see how server-side decorations are being leveraged to map the titlebar contents and controls to the display, mixed with window manager controls like workspace switching with live previews, as well as client-announced custom inputs.
XWayland Client Isolation
We give up. The Wayland protocol service has received support for XWayland clients.
For the last five years or so, the plan was to only support native Wayland clients. After all this time, it turns out that the X11 backends for various toolkits are still infinitely more reliable and robust through X than through native Wayland.
As with anything else here, we treat it, and the obligatory meta-window manager every Wayland Compositor has to have in order to work around the hackish way XWayland uses Wayland, as a separate process for the privsep space to be as tight as possible.
The best way of using this is for the individual X clients that you might need to run; browsers and games being the main suspects.
arcan-wayland -exec-x11 chromium
This will spawn an instance of the wayland protocol server side, which spawns the virtual window manager, which inherits a clipboard window and spawns an Xwayland instance along with chromium.
This means that each client gets their own infrastructure, working directory and so on. It consumes more memory and sometimes has a noticeable startup time (most of that being MESA), but enables much stronger separation between clients.
If that is undesired, it can, of course, be run as a normal service:
arcan-wayland -xwl DISPLAY=:0 xterm
This would give you a global Wayland display that also provides XWayland on demand, and clients that expect to spawn multiple processes that connect at different times, to work.
This does not mean that Xarcan is defunct or will become be abandoned as the goals are quite different. In fact, Xarcan has recently been synched to changes in upstream Xorg. The Wayland path is currently to get the transparent rootless one, while Xarcan behaves more like Xephyr; ‘Virtual Machine’ like controls where you provide your own window manager and so on. Both modes have justified use cases.
That said, it is likely that rootless window support will be added to Xarcan as well to free legacy client support from the hands of Wayland entirely, as I am, quite frankly, so thoroughly disappointed in- and tired of- coding anything Wayland to the point that I’d much rather mess with Xorg internals than read more chapters from this scattered in the wind, half-java RMI love letter, half hand MS-COM+ anti-patterns handbook. Had the hours spent debugging this been billable, I’d buy a Tesla then return it for the money.
Speaking more of this second coming of TERMCAP – this round of “protocol” pokémon added: xdg_decoration, kwin_decoration, xdg_output, zwp_dma_buf, and probably some others that I forgot. This seems to now be enough to run the winner of ‘what the actual f are you doing?’ clients, Firefox – until a nightly update breaks it again, that is.
Although it is not a very visible feature, it is still very important in its own right. In preparation for proper VFR/VRR (variable frame-rate/refresh rate), better biased client processing (two sides of the same coin when you factor in network transparency), a lot of the old synchronisation code has been consolidated and reworked into a ‘conductor’. The conductor exposes a uniform set of synchronisation strategies for the policy layer (WM) to switch between.
The strategies themselves will need a lot of tuning before they provide the best contextual balance and trade-off for gaming (minimise latency), processing (maximise throughput), visual fidelity (filters, colour processing, animation smoothness) and minimising energy consumption (mobility, cloud hosting). There is no ‘one size solution fits all’ here, and these are problems that in their own rights would consume a master level thesis, or three, on the subject.
It might help to think of it as analogue to the role of the heart in the human body. A single beat has far reaching consequences in a complex system. The ideal rate depends on what the body is supposed to do; running a marathon is different from a sprint and different from bedrest – yet all need to be supported. If you are academically inclined and actually want to dip into the fundamentals, finish a course or two on queue theory then contact me.
To this effect, there are API additions on both the high and the low end. For the WM side, we have added the ability to set a synchronisation focus target that gets preferential treatment so that when mapped to a variable- output and deadlines are closing in, we can try and bias the way it receives input and how the frames gets synchronised compared to other clients that are fighting for the same resources using differently staged releases for ‘the current focus’ and ‘the herd’ to avoid a twisted take on the sleeping barber problem.
On the low end, we can now continuously forward information about the next upcoming deadline on a per client basis, as well as provide different deadlines based on active strategy, segment type, window management state and client state (networked or local for instance).
With this feature we have also added much needed tracing features for most of our layers so that it is possible to on-demand collect a minimal overhead description of the sequence of events and their timings. There are so may combinations of asynchronous, threaded and non-blocking operations here across so many layers that measuring and troubleshooting anything performance related pretty much demands it. If in doubt when it comes to tracing and viewers, disregard any suggestions about webcrap like Google im-‘Perfetto’, and give https://github.com/wolfpld/tracy the love it deserves, it is a beautiful piece of work.
New Tool: Arcan-trayicon
The list of tools has also been expanded with a convenience wrapper / launcher utility, ‘arcan-trayicon’ which lets you register two SVG icons (inactive and active) into some UI element in a supporting window manager. When clicked, the wrapped application is launched and attached as a popup to this icon. This combined with some bits from the next section should be all that is needed to wrap system services and network interface tools with a UI yet not bring any of that D-Bus stench with it.
TUI, Server-side Rendering, Widgets and non-VTx shell
The rich-text oriented client developer facing API, arcan-tui, has undergone serious backend refactoring. It can now render into a specialised text-encoding scheme (TPACK), and a server-side renderer for the same format has been added to Arcan.
This is the first step to drastically cut down on rendering costs and memory consumption as the glyph cache can now be generated and shared between otherwise independent windows of the same type, as well as using accelerated rendering without unnecessarily forcing the TUI clients to have GPU access. For tracking the feature, check the text rendering section in the wiki on TUI.
It also provides the server with a textual representation to work with for other features, such as lossless magnification and screen readers (see Text-To-Speech below).
It has also received more abstract UI widgets so that certain features gets a shared base. Some of these are visible in the debugging video that was shown earlier. This round we added three widgets (out of a planned total of four):
Bufferwnd – take a buffer and provide a read/write hex-/ascii-/unicode- viewer or editor interface to it.
Listwnd – take a list of items, tag separators, submenus, inactive and hidden items, and prevent it as a selectable list. This is a suitable base for more advanced components (tree view etc).
Readline – act as libreadline/linenoise to query the user for input. It still lacks some of the important bells and whistles to fully cover that use-case, and hopefully those holes will be plugged shortly.
The last planned component (on top of more features and quality work to the above) is Linewnd that will provide scrollback buffer like behaviour. This will eventually remove the last remnants of the libtsm ‘screen’ that we used to build things in the past, but are now slowing us down.
The pre-existing ‘screenshot’ copy widget hidden inside every arcan-tui window also got an added ‘annotation mode’ where it is possible to write and highlight parts of the copied surface.
Speaking of arcan-tui, there is also a prototype neovim UI built using it – it is roughly on par with current curses/terminal emulator based neovim, but will soon leverage many more of the features in arcan-tui, and will be used as a requirements ‘specification’ to make sure that we have not missed anything. It can be found in the repository here.
On a related note, the included terminal emulator, afsrv_terminal, has received a special mode (ARCAN_ARG=cli afsrv_terminal) that serves as a basis for terminal emulator liberated command-line shells. The video below shows some of the current state of that shell, running inside a window manager that is well, still part of my big pile of secrets.
Worthwhile to note is that the ‘new’ windows are negotiated by the shell, and based on what kind of a command that is to be executed – sets up the environment accordingly. Wayland routes through arcan-wayland and so on.
Text to Speech and other Decode improvements
To recap, the ‘decode’ frameserver is the special dedicated client that is intended to absorb all dangerous and likely insecure parsing from the main process to a ‘kill on sight, single role’ one that gets aggressively sandboxed.
It is used both for basic decoding and for asynch- live content previews in parts like the durden filesystem browser. It is also used in a ‘plan9 plumber’-like fashion for clients to hand over media presentation (both embedded and stand-alone).
When finished, no parser working on untrusted data should remain in the arcan process, no matter if Arcan is used as the outer display server or as a ‘browser/electron-lite’ like app-engine nested inside itself or as an X/Wayland client.
In this round, we have added support for text-to-speech and chiselled out basic 3D model transfers. The text to speech, combined with the existing OCR feature of the ‘encode’ frameserver, provides WMs like Durden the last needed missing primitives pieces for greatly improved accessibility.
These improvements also have their role in VR, with positional audio and HRTFs (still disabled in upstream builds) – notification sounds, longer text-to-speech passages and so on can, if carefully configured, be much more natural and comfortable than visual ‘widget’ like alerts when you can position the source ‘in the room’.
The way headless mode (no attached displays, ever) operates has been reworked as a separate binary with a slightly different setup. It will now output to an instance of the ‘afsrv_encode’ process, while also using it as input layer.
This allows us to run the whole thing on a server in a container and either record or stream its output, but also to serve a remote desktop or single app host:
Of course, you can also switch that ugly ‘vnc’ for a12.
Combined with the dynamic part of the network transparency bits, you can have a headless server where your clients normally ‘live’, then connect to it and temporarily redirect individual clients to a machine of your liking.
As part of our ongoing effort in improving systemic resilience (the ability to recover from unforeseen events), the egl-dri (native graphics on Linux/BSDs) privileged separation part that does driver negotiation has received support for two levels of watchdog triggers.
The first will reset the scripting layer, protecting against bad scripts getting stuck in an infinite loop, while the second will trigger crash recovery for the engine process itself – trying to at least recover clients.
To complement this, and continue down the path towards proper non-cooperative, asymmetric multi-GPU support (runtime swapping graphics stack and driver versions and persisting across multiple graphics stacks at the same time, say GBM+KMS/EGLStreams), it is now possible to set a configuration where one GPU acts as a stand-in for another, and have the WM cooperatively trigger a swap.
This requires both client and server to be able to completely tear down and rebuild accelerated graphics use. This also provides an easy path for testing that we do not leak state between rebuilds (define a GPU as a backup for itself and set a timer that randomly resets), which makes the last missing step — transferring individual clients and render jobs between being native to one or many different GPUs at once — approachable.
This article is is the main course to the appetiser that was The X Network Transparency Myth (2018). In it, we will go through how the pieces in the Arcan ecosystem tie together to advance the idea of network transparency for the desktop and how it sets the stage for a fully networked desktop.
Some of the points worth recalling from the X article are:
‘transparency’ is evaluated from the perspective of the user; it is not even desirable for the underlying layers to be written so that they operate the same for local rendering as they would across a network. The local-optimal case is necessarily different from the remote one, the mechanisms are not the same and the differences will keep on growing organically with the advancement of hardware and display/rendering techniques.
side-band protocols splitting up the desktop into multiple IPC systems for audio, meta, fonts, … increases the difficulty to succeed with anything close to a transparent experience, as the network layer needs to take all of these into consideration as well as trying to synchronise them.
To add a little to the first argument: it should also not be transparent to the window manager as some actions have drastically different impact on the user interface side to security and expectations. For example, Clipboard/DND locally is not (supposed to be) a complicated thing. When applied across a network, however, such things can degrade the experience for anything else. Other examples is that you want to block some sensitive inputs from being accidentally forwarded to a networked window and so on, it has happened in the past that the wrong sudo password has, indeed, been sent to the wrong ssh session.
This target has been worked on for a long time, as suggested by this part from the old demo from 2012/2013. Already back then the drag/slice to compose-transform-and-share case exposed out of compositor sharing and streaming; something that only now is appearing elsewhere in a comparably limited form.
We are on the third or fourth re-implementation of the idea, and the first one that is considered having a good enough of a design to commit to using and building upon. There are many fascinating nuances to this problem that only appear when you ‘try to go to 11’.
As per usual, parts of this post will be quite verbose and technical. Here are some shortcuts to jump around so that you don’t lose interest from details that seem irrelevant to you.
Starting with some short clips of the development progress – and then work through the tools and design needed to make this happen. It might be short, but there is a whole world of nuance and detail to it.
Composited Xarcan (desktop to pinephone), compression based on window type:
Here is a native arcan client with crypto, local GPU “hot-unplug” to software rendering handover and compression negotiation (h264):
Here is ‘server-side’ text rendering of text-only windows, font, style and size controlled by presenting device — client migrates back when window is closed:
In the videos, you can see (if you squint) instances of live migration between display servers over a network, with a few twists. For example, the decorations, input mapping, font preferences and other visuals change to match the machine that the client is currently presenting on and that audio also comes along, because Arcan does multimedia, not only video.
What is less visible is that the change in border colour, a security feature in Durden, is used to signify that the window comes from a networked source, a property that can also be used to filter sensitive actions. The neo-vim window in the video even goes so far as to have its text surfaces rendered server side, as its UI driver is written using our terminal-protocol liberated TUI API. This is also why the font changes; it is the device you present on that defines visuals and input response, not the device you run the program on.
Also note how the clients “jumps” back when the window is closed on the remote side; this is one of the many payoffs from having a systemic mindset when it comes to ‘crash resilience‘ – the IPC system itself is designed in such a way that necessary state can be reconstructed and dynamic state is tracked and renegotiated when needed. The effect is that a client is forcefully detached from the current display server with the instruction of switching to another. The keystore (while a work in progress) allows you to define the conditions for when and how it jumps to which machines and picks keys accordingly.
That dynamic state is tracked and can be renegotiated as a ‘reset’ matters on the client level as well, the basic set of guaranteed features when a client opens a local connection roughly generalises between all imaginable window management styles. Those that are dynamically (re-) negotiated cannot be relied upon. So when a client is migrated to a user that has say, accessibility needs, or is in a VR environment, the appropriate extras gets added when the client connects there, and then removed when it moves somewhere else. This is an essential primitive for network transparency as a collaboration feature.
Basic Primitives: Arcan-net, SHMIF and A12
There are three building blocks in play here, a tool called arcan-net which combines the two others: A12 and SHMIF.
A12 is a ‘work in progress’ protocol – it’s not theX12 that some people called for, but it’s “a” twelve. It strives to be remote optimal – compression tactics based on connectivity, content type and context of use, deferred (presentation side) rendering with data-native representation when possible (pixel buffers as a last resort, not the default); support caching of common states such as fonts; handle cancellation of normally ‘atomic’ operations such as clipboard cut and paste and so on.
SHMIF is the IPC system and API used to work with most other parts of Arcan. It is designed to be locally optimal: shared memory and system ABI in lock free ring-buffers preferred over socket/pipe pack/unpack transfers; minimal sustained set of system calls needed (for least-privilege sandboxing); resource allocations on a strict regimen (DoS prevention and exploit mitigation); fixed based set of necessary capabilities and user-controlled opt-in for higher level ones.
SHMIF has a large number of features that were specifically picked for correcting the wrongs done to X- like network transparency by the gradual introduction of side-bands and good old fashioned negligence. Part of this is that all necessary and sufficient data exchange used to compose a desktop goes over the same IPC system — one that is free of unnecessary Linuxisms to boot. While it would hurt a bit and take some effort, there are few stops for packing our bags and going someplace else, heck it used to run on Windows and still works on OSX. Rumour has it there are iOS and Android versions hidden away somewhere.
Contrast this with other setups where you need a large weave of IPC systems to get the same job done; Wayland for video and some input and some metadata; PulseAudio for audio; PipeWire for some video and some audio; D-Bus for some metadata and controls; D-Conf for some other metadata; Spice/RFB(VNC)/RDP for composited desktop sharing; Waypipe for partial Wayland sharing, X11 for partial X / XWayland sharing: SSH+VT***+Terminal emulator for CLI/TUI and less unsafe Waypipe / X11 transport; Synergy for mouse and keyboard and clipboard and so on. Each of these with their own take (or lack thereof) on authentication and synchronization, implementing many of the most difficult tasks again and again in incompatible ways yet still end up with features missing and exponentially more lines of code when compared to the solution here.
Back to Arcan-net. It exposes an a12 server and an a12 client, as well as acting as a shmif server, a shmif client and taking care of managing authentication keys. In that sense it behaves like any old network proxy. While not going too far into the practical details, showing off some of the setup might help.
On the active display server side:
email@example.com# arcan-net -l 31337
This will listen for incoming connections on the marked port, and map them to the currently active local connection point. To dive further into the connection point concept, either read the comparison between Arcan vs Xorg or simply think ‘Desktop UI address’; The WM exports named connection points and assigns different policies based on that.
On the client side we can have the complex-persistent option that forwards new clients as they come:
This triggers the SHMIF implementation tied to the window of a client to disconnect from the current display server connection, connect to a remote one through arcan-net, then tell the application part of the client to rebuild essential state as the previous connection has, in a sense, ‘crashed’. The same mechanism is then used to define a fallback (‘should the connection be lost, go here instead’). This is the self-healing aspect of proper resilience.
There are WM APIs for all the possible network sharing scenarios so it can be handled as user interfaces without any command line work.
I mentioned ‘authentication’ before, where is that happening? So this is another part of the implementation that is still settling. We are reasonably comfortable with the cryptography engineering, which is currently undergoing external/independent review.
For those versed in that peculiar part of the world, the much condensed form is currently AE/EtM, ChaCha8 for stream cipher, BLAKE3 for KDF and HMAC, x25519 for asymmetric key exchange (optionally with initial MinimaLT like ephemeral exchange) and PAKE-like (possibly changing to IZKP) n-unknown-initial PK authentication (rather than PKI), with periodic rekeying.
For great security (and justice): As demoed, moving applications to present on other devices is a neat thing for getting more use out of computers that are otherwise cast aside when you have a big beefy workstation; put those SBCs to use, security compartment per device and happily ignore most of the endless stream of speculative attacks instead of allowing mitigations to bring processing power back a decade.
Take that clusterboard of SOPINE modules, boot them into a ramdisk with a single-use browser and let that be your ‘tab’, pull the reset pin when done and that very expensive chrome- or less expensive firefox- exploit chain will neither be able to grab many interesting tokens nor gain meaningful persistence or lateral movement. Hook it up to the same collection/triaging system used for the fuzzers you have running and pluck those sweet exploit chains out of the air ready for re-use responsible disclosure.
Take those privacy eroding apps and adtech drivel that managed to weasel their way into a position of societal necessity and run them ‘per device’ but far away from your person; feed their sensory inputs with sweet, GPT-3-esque plausibly modelled nonsense, tainting whatever machine-learning trend that feeds on this particular perversion until they are forced to pay cost in order to discern truth from tales.
For great mobility: Ideally, the use and experience should be seamless enough that you would not need to care where the client itself is currently running – that is to continue down the path of Sun Ray thin clients towards thin applications as a counterpoint to the mental- and technical dead end that is the modern web.
Applications that are capable of serialising their seed- and delta- states and responds to the corresponding SHMIF events for RESTORE/STORE and RESET can first remote render as the safe default, while in the background synchronise state with the viewing device (should the same or compatible software exist there) and switch over to device-local execution when done.
Then have an Arcan instance provide ‘your real composited desktop’ running ‘in the cloud’. Let the applications and their state live there while travelling across unsafe spaces, and redirect the relevant ones to your device when it is safe and necessary to do so.
For great collaboration:
Think of an ‘window’ as a view into an application and its state as a living document that can both be shared and transferred between machines and between people. What one would call ‘clipboard cut and paste’ or ‘drag and drop’ is not far from ‘file sharing’ if you adjust the strength of your reading glasses a little bit. There are pitfalls, but that is our job to smooth over.
With decorations, fonts, semantic colour palette, input bindings and so on being defined by the device a client is presenting on rather than the settings of the local client through some god-forsaken config service or font specification, differences in aesthetics, accessibility and workflow can be accounted for rather than accentuated.
Take that ‘window’ and throw it to your colleagues, they can pull out what state they need or add to it and return it back to you when finished. Splice your webcam and direct it to your friend and there is your video conferencing without any greedy middle-men.
For great composability:
The arcan-net setup, as shown before, covers remote presentation/interaction with SHMIF clients. This includes exotic external input drivers and sensors, enabling advanced takes on “virtual KMS” workflows that applications like Synergy provides.
Take one of those spare tablets gathering dust and use it as a remote viewer/display – or forward otherwise useless binary blobbed special input devices (Tobii Eyetrackers come into mind) from the confines of a VM.
Take those statusbar- applications, run them on your compute cluster or home automation setup and forward to your viewing device for SCADA- like monitoring and control.
Personally, my fuzzing clusters ‘phone home’ by forwarding wrapper and control clients whenever a node finds something to my desktop. By leveraging the display server to improve debugging over the same communication channels I get my crash inspections over without juggling symbols, network file systems and what not around. Redirect the interesting ones to the specialist on call for such purposes.
The range of applications that various combinations of these tools enable is daunting. Your entire desktop, with live application state, can literally be made out of distributed components that migrate between your devices – drag and drop:able.
The reference implementation is in heavy/daily use, but there are a number of things to sort out before we would dream to push- or put- it to use for sensitive tasks. If you want to run your terror-cell or other shady business with the tools, go ahead. Anything or anyone more honest and decent – stay away until told that it is reasonably safe.
All the infrastructure around this will be actively developed for a long time, and there is a fair list of coming interesting and advanced things as the focus for this work. Track the README.md for changes.
Some of the highlights from my perspective:
Safety features (side channel analysis resistance, transitive trust) – machine learning is listening in and interactive web is no different; the field is ripe with tools that can reconstruct much of plaintext from rather minor measurements without having to attack the cryptography engineering or primitives themselves.
Spliced interactive/non-interactive subviews – for group collaboration.
Compressed video passthrough negotiation – to ensure that no pack/unpack stage for already compressed sources like video persist.
ALT/AGP packing (Arcan rendering command buffers) – to stream both mid-level graphics and its virtual-GPU like backend as draw primitives when no better representation is available.
Latency/Performance work – Better domain specific carrier protocols (UDT and the likes), progressive compression for certain client types, buffer backpressure mitigations strategies. Network state deadline estimation for better client animations.
But more on those at a later time.
Having, hopefully, wet your appetite – lets close this one by explaining what went on in the demo.
As you might have seen in the example command lines at the top, or in the previous articles, “connection points” is a key primitive here. They allow the window manager (WM) and the user to define ‘UI addresses’ so clients know where to go, and the WM knows which policy to use to regulate the IPC mechanisms provided.
Any ‘segment’ (group of audio/video/event-io roughly corresponding to a window) has a primary connection point, and a fallback one. Should the primary connection point fail, the client will try to reconnect to a mutable fallback, and then rebuild itself there. This allows clients to move between display server instances, something that covers elaborate sharing, more safe/robust multi-GPU support etc.
This fallback is provided through the WM via a call to target_devicehint. It comes in two main flavours, “soft” (use only on a severed connection), and “hard” (leave now or die).
In the videos, the Durden WM is used. To make a long story short, it has an overwhelming amount of features (600+ unique paths last I counted), structured as a virtual filesystem (even mountable). In this filesystem, the specific path “/target/share/migrate=connpoint” tells the currently selected window to “hard” migrate to a connection point. In this case it is the special flavour of a12://my_other_machine that indirectly fires up arcan-net to do most of the dirty work.
With the stacking workspace layout mode, there is a feature called “cursor regions” – parts of the workspace that trigger various file system paths depending on if the mouse is moving, dragging or clicking.
The way things are setup here then, is that:
When a window is dragged into the left edge area of the screen, draw a portal graphic.
Default the contents of the portal to be noise, spawn an instance of the ‘remoting’ frameserver connecting to the host IP of the left laptop (why you can see parts of the remote wallpaper in one of the videos).
If connected, show the contents of this remoting connection.
If the window is dropped over the portal, send the /target/share/migrate command, pointing to the remote server in question.
If the drag action leaves the region, kill the remoting connection and hide the portal.
Onwards and upwards towards more interesting things. /B