hihaheho/desk

🔮 The application platform for your cyberpunk desk

Desk

🔮 The application platform for your cyberpunk desk

🧊 Project Status: FROZEN 🧊

This project is currently on hold. We may consider resuming development in the near future if there's continued interest or demand. For updates on our ongoing projects, please check our Discord or X.

No Release!

Status: All releases are behind the previous schedule for several reasons, so I updated it below.

Reasons:

• I'm building query-flow for deskc and dworkspace.
• I plan to discard the current mirgen and design a new MIR intervening between Desk-lang and low-level IR to generate a GC-less binary, which is fast as Rust in theory (in my mind).
• I've switched to Ubuntu Desktop from M1 Macbook Pro mainly for financial reasons. As a subsequence of it, I can utilize the mold linker.
• I've returned to Neovim after several years of a temporary stay in the VSCode.
• With the lowest priority, I have about three my own projects now to improve IDE experience with the Neovim ecosystem.
• I've built a Keyball39, a split keyboard placing a injection-molded trackball slot in the really ergonomic position.
• I'm going to completely re-design the Desk-lang.

Pre-release (by the end of 2025)

Pre-release includes:

• incremental Desk-lang compiler
• DeskVM with an official scheduler
• a file system for Desk-lang
• Desk-lang visual editor
• A web demo like Rust Playground for Desk-lang

The first release (by the end of 2030)

The first release includes:

• MVP of Desk Craft, a game engine
• a platformer game demo
• a space to publish created games
• real-time collaboration on Web
• paid plans for Desk X (official hosting)

See the draft of the first release

Goals and Philosophy

🎯 Blur the line between living and coding 🎯 Make every software programmable

• 🎮 Intuitive like games
• 🥼 Pragmatic like professional tools
• 🗺️️ Versatile like spreadsheets
• 💗 Accessible to everyone
• 🛹 Minimalist design

Why Desk?

Desk apps are inherently:

• 🎼 code-oriented like data-oriented
• 🔒 statically-typed (data and UI)
• 🤖 programmable (extensible by code)
• 🧲 interoperable with other Desk apps
• 📱 running on everywhere (web, desktop, mobile)

How does Desk work?

Desk is consist of:

• Desk Programming Language and Desk Compiler (deskc)
• Desk Workspace System (dworkspace)
• DeskVM (deskvm)
• Desk-plugins (dplugins)

Desk Programming Language (Desk-lang)

Desk-lang is a programming language that has:

• minimalistic syntax and semantics
• type system with inference
• algebraic effects
• content-addressable by type and UUID
• incremental compilation

Most of the data and programs on Desk apps are finally evaluated as a snippet of Desk-lang.

Desk Compiler (deskc)

Desk compiler is an incremental compiler for Desk-lang.

Crates

• deskc: the incremental compiler
• deskc-lexer scans Desk-lang source code and generates tokens
• deskc-parser parses tokens and generates an AST
• deskc-typeinfer infers types of expressions.
• deskc-mirgen generates MIR

Desk-workspace (dworkspace)

Desk-workspace is a platform-agnostic environment for editing Desk-lang.

Desk-workspace provides these features:

• file system for Desk-lang
• permission management system
• realtime collaboration support

Crates

• dworkspace: the implementation
• dworkspace-codebase defines structs for a codebase

DeskVM (deskvm)

DeskVM is a runtime for Desk-lang influenced by Erlang VM.

Features

• platform-agnostic
• capable of running many programs as a d-process
• type-driven message passing and pub/sub
• interpreter-agnostic: DeskVM can run anything as a d-process
• preemptive scheduling
• custom scheduler support

Crates

• deskvm: the implementation
• deskvm-dprocess defines structs of such as d-process

Desk-plugins (dplugins)

There are many Desk-plugins. Each Desk-plugin implements a single feature as a Bevy Plugin.

• 🚧 Desk Craft for game development
• 🚧 Desk Brain for productivity
• 🚧 Desk Verse for communication
• 🚧 Desk Robot for automation
• 🚧 Desk Board for BI
• 🚧 Desk Calendar for scheduling
• 🚧 Desk Pages for hosting

🚧 not yet implemented

Is it any good?

Yes.

Resources

• Contributing Guide

Join our community

👉

👉

👉

👉

👉

desk's Issues

Tracking Issue for defining Desk-lang

Parse `1.3` to `13/10` in parser

We should parse all floats as rational numbers for now.

`ref` type and `mut` type

'ref type is something like a Arc<T> and 'mut type is something like a Arc<Mutex<T>>.

Syntax

<'ref 'nat> 'ref 1

$ <'mut 'nat> 'mut 1;
// update the value
'mutate & 'mut 'nat & \ 'nat -> 'nat

Consistent NodeID to leverage incremental compilation

Background

• deskc does incremental compilation for Desk-lang code.
• dworkspace emits ASTs with inconsistent NodeId in most cases.

Problem

In most cases, full compilation occurs because NodeIds are changed.

Solution

• Use moka for persistent generated NodeIds.

Auditor can crash in loop detection with unknown NodeId

I'm waiting for salsa-rs/salsa#87 solved

Dependency Dashboard

This issue lists Renovate updates and detected dependencies. Read the Dependency Dashboard docs to learn more.

Open

These updates have all been created already. Click a checkbox below to force a retry/rebase of any.

• Update Rust crate ariadne to 0.3
• Update Rust crate pretty_assertions to 1.4.0
• Update Rust crate uuid to 1.4
• Update non-major dependencies (bevy, bevy-inspector-egui, bevy_ecs, bevy_egui, bevy_math, bevy_rapier2d, itertools, parol, parol_runtime, pretty_assertions, strum, uuid)
• Update actions/checkout action to v4
• Click on this checkbox to rebase all open PRs at once

Detected dependencies

cargo

Cargo.toml

• parking_lot 0.12
• thiserror 1.0
• uuid 1.3
• bevy 0.10
• egui 0.22
• bevy_math 0.10
• bevy_ecs 0.10
• once_cell 1.18
• maybe-owned 0.3.4
• egui_cable 0.5
• ron 0.8

crates/adapters/dworkspace-in-memory/Cargo.toml

crates/apps/desk-x/Cargo.toml

• bevy-inspector-egui 0.18
• console_error_panic_hook 0.1
• web-sys 0.3

crates/components/desk-command/Cargo.toml

crates/components/desk-physics/Cargo.toml

crates/components/desk-plugin/Cargo.toml

crates/components/desk-theme/Cargo.toml

• serde 1.0

crates/components/desk-window/Cargo.toml

crates/components/deskc-ast/Cargo.toml

• downcast-rs 1.2

crates/components/deskc-errors/Cargo.toml

crates/components/deskc-hir/Cargo.toml

crates/components/deskc-ids/Cargo.toml

• uuid 1.3
• serde 1.0

crates/components/deskc-mir/Cargo.toml

• serde 1.0

crates/components/deskc-type/Cargo.toml

• serde 1.0

crates/components/deskvm-dprocess/Cargo.toml

• anyhow 1.0
• uuid 1.3
• mry 0.2.6
• serde 1.0

crates/components/dson/Cargo.toml

• serde 1.0

crates/components/dworkspace-codebase/Cargo.toml

crates/libs/deskc-macros/Cargo.toml

• quote 1.0
• proc-macro2 1.0

crates/libs/serde-dson/Cargo.toml

• serde 1.0
• serde 1.0

crates/plugins/desk-editor/Cargo.toml

crates/plugins/desk-egui/Cargo.toml

• bevy_egui 0.20

crates/plugins/desk-playground/Cargo.toml

crates/plugins/desk-rapier2d/Cargo.toml

• bevy_rapier2d 0.21

crates/plugins/desk-terminal/Cargo.toml

crates/plugins/desk-touchpanel/Cargo.toml

crates/systems/deskc-hirgen/Cargo.toml

• pretty_assertions 1.3.0
• chumsky 0.9.2

crates/systems/deskc-mirgen/Cargo.toml

crates/systems/deskc-syntax-minimalist/Cargo.toml

• parol_runtime 0.16.0
• anyhow 1.0
• uuid 1.3
• env_logger 0.10.0
• pretty_assertions 1.3.0
• parol 0.21.5

crates/systems/deskc-typeinfer/Cargo.toml

• itertools 0.10
• chumsky 0.9.2
• ariadne 0.2
• pretty_assertions 1.3.0
• env_logger 0.10.0

crates/systems/deskc/Cargo.toml

• salsa 0.16
• anyhow 1.0

crates/systems/deskvm-miri/Cargo.toml

• serde 1.0
• anyhow 1.0
• strum 0.24
• chumsky 0.9.2

crates/systems/deskvm/Cargo.toml

• anyhow 1.0
• uuid 1.3

crates/systems/dworkspace/Cargo.toml

• salsa 0.16
• serde 1.0
• downcast-rs 1.2.0
• anyhow 1.0
• uuid 1.3
• mry 0.2.6

github-actions

.github/workflows/ci.yml

• actions/checkout v3
• actions-rs/toolchain v1
• Swatinem/rust-cache v2
• EmbarkStudios/cargo-deny-action v1

---

• Check this box to trigger a request for Renovate to run again on this repository

Span data

The current parser emits AST with span data of 0..0. parol have a way to get span for syntax elements.

• Define SpanStorage trait
• Use parol with a version that supports span and enable it for generation.
• Make MinimalistSyntaxSpanStorage able to return span data for any NodeId.

Scope escape of type variables

https://counterexamples.org/scope-escape.html

Add test cases to deskvm-dprocess

Helpful parse error

The current parser emits poor errors on parse errors.
We should wait for jsinger67/parol#43.

Add N-bit integer types like `i32` and `u8`.

Data

enum Type {
    /// N-bit signed integer
    Signed { bit: u8 },
    //// N-bit unsigned integer
    Unsigned { bit: u8 },
    ...
}

Subtyping

• uA <: uB if A <= B
• iA <: iB if A <= B
• uA <: iB if A < B (0x1111_u can be casted to 0x01111_i)
• uN <: 'nat for any N
• iN <: 'int for any N

Coherence

Introducing either or both of the following rules breaks the coherence of the type system

• 🚫 uN <: iN for any N
• 🚫 iN <: uN for any N

For example: Is <'int> 0b11 -1 or 3?

Syntax

'do <'u3> 0b000;
'do <'u8> 0xff;
$ 1_u8;
$ -2_i32;
*<
  & 'u8
  & 'i32
>

Naming of u8

• Following the naming of 'int and 'nat, it seems that i8 or n8 would be nice. Like i8 is a subset of 'int, n8 is a subset of 'nat. This is not strange.
• Following that u8 means "unsigned integer", it seems that s8 (means "signed integer") and u8 would be nice.
• How about p8 (means "positive integer") and i8

But I choose u8 and i8 to minimize surprises for the programmers. Non-programmers may be surprised at u8 but u8 is rare enough for most Desk codes.

How about bits type like b8

It sounds nice, but the following subtyping rules break the coherence of the type system.

1. bN <: uN for any N
2. bN <: iN for any N

So we must throw Rule 2 or both of them away.

Rule 1 is safe and uN <: bN is also safe, so we can say uN = uN. uN is used to represent bits in many languages.

Literal Types

Something like dependent types.

let a: "a" = "a"

Π, Σ, and [t; N]

Match exhaustive check

Polymorphic effect typing

Fix subtyping of map

Formal specification of Desk-lang

Add (automated) sponsors section in README

I don't want to use a GitHub action like this for security reasons. I need a cli tool to generate sponsors svg.

Use NewType pattern for meta ID

'loop and 'break

Find most matching type in subtyping of product and sum types

Support sum type in a case of match expression.

Support more unicode quotes for raw string.

Currently, the parser only supports "«»" and "‹›".

Support comments right after `,` or `;`

Syntax

?, // hahaha

?; // yeah

Why?

Sometimes it’s slightly or super annoying to write a comment for an expression.

For example;

// Here is a comment for whole let expression
$ /* Here is a comment for the hole */ ?;

to be:

// Here is a comment for the whole let expression
$ ?; // Here is a comment for the hole

And

*<
    1,
    2,
    // Here is a comment for 3
    3,
>

to be:

*<
    1,
    2,
    3, // Here is a comment for 3
>

How to avoid ambiguity?

Answer: We introduce new tokens, CommaComment and StmtEndComment.

CommaComment: ",[\s--\r\n]*//";
StmtEndComment: ";[\s--\r\n]*//";

Allow reorder of arguments

Design a generic performant-first interpreter

Current implementation

The interpreter we are using now evaluates MIR (Mid-level IR) directly. It's easy to understand but may have bad performance.
MIR is a collection of CFG (Control Flow Graph), and each instruction of basic blocks is in A-normal form.

Purpose

The purpose is to create a new interpreter for cases that requires a generic performant interpreter.
Because DeskVM can handle multiple types of interpreters, we don't need to replace the current one.

Design requirements

• It must implement the Interpreter trait.
• It must be generic because we can have many interpreters specialized for a specific domain.
• Performant execution is the first, and performant compile is the second.
• It must have a resource management system such as GC.
• No JIT compilation (the reason is written in the Related notes).
• It should be sound.
• We should keep design and code as simple as possible while performant is first.
• The interpreter must be suspendable and resumable for scheduling.

Related notes

• We'll create an AOT compiler in the future.
• On desktop platform, DeskVM has an ability to run any executable file as a d-process in the VM. It means we can use any executable binary including binaries AOT-Compiled Desk code by just performing a delegation effect in a code. So we don’t need JIT anymore.
• Since Desk-lang has no side effects, the only side-effective opcode is perform-an-effect.
• [Needs proof] Because Desk-lang may have a soundish type system, we don't need to worry about crashes by type problem.
• Desk-lang has no operator other than calling a function.
• Desk-lang doesn't define built-in functions, so we can define ones for this interpreter.
• MIR distinguishes data structures such as arrays and linked lists.
• MIR can contain region-analyzed references.

Related crates

• deskc-mir defines the structure of MIR
• deskvm-miri implements the current de facto interpreter
• deskvm-process defines the process struct like ErlangVM's one and the interpreter trait
• deskvm implements the virtual machine like ErlangVM

References

• BEAM bytecode of Erlang VM
• Ruby bytecode

Allow to omit the output type of perform expression

Add tests for DeskVM

Because mry lacks some functionalities, we need manually implemented mocks.

Use `SmolSet` in `deskc_types` for `Product` and `Sum`

Remove many clones in typeinfer

We should change the signature of methods in Ctx to use self instead of &self.

Recursive Type

• syntax
• typeinfer

Use bigint instead of i64

Desk must support bigint

Currying by partial application of product type

Add sponsors section in

`gcref` and `gcmut`

Garbage collected version of ref and mut.