Installation
Install the Windows x64 release with Scoop or build vynec from source, then set up optional embedded, CUDA, and graphics tools.
Windows x64 users can install a precompiled Vyne release with Scoop. Source builds use a C++23-capable compiler and xmake; generated desktop programs are C11 and the current compile/run path invokes GCC. Embedded and graphics applications need the additional tools described below.
Windows: install with Scoop
Install Scoop and ensure Git is available, then run in PowerShell:
The Vyne bucket downloads the precompiled Windows x64 release and checks the manifest's SHA256. The package includes the runtime/module/template files used by compilation and project initialization. GCC is installed as a dependency because vynec file.vy compiles and runs. Installing the precompiled toolchain does not require building Vyne with xmake.
Create and run a first project from your preferred projects directory:
Update or uninstall
scoop update refreshes Scoop and its bucket manifests; scoop update vyne installs a newer published package when available. Project files outside the Scoop installation directory remain separate.
Installed versus local builds
Scoop normally installs under %USERPROFILE%/scoop/apps/vyne/current/ and exposes vynec through a shim. It does not replace a source checkout. Use vynec for the PATH-selected installation and ./vynec.exe inside a Windows source checkout for your local executable.
These commands show tool selection if you keep multiple compiler installations. See Hello, World for execution modes and Projects & Tooling for manifests and initialization.
Source-build prerequisites
| Component | Needed for | Notes |
|---|---|---|
| C++23 compiler | building vynec | xmake.lua uses -std=c++23; Windows selects MinGW |
| xmake | build system | scoop install xmake, brew install xmake, or the official install script |
| GCC | generated desktop programs | The current CLI invokes gcc; Scoop installs it as a dependency |
avr-gcc + avr-libc | Arduino firmware | the AVR toolchain from Arduino or your package manager |
arduino-cli + arduino:avr | flashing and serial monitor | arduino-cli core install arduino:avr |
| OpenBLAS | --blas | optional; vendored binaries live under vendor/openblas/ |
arduino-cli + esp32:esp32 | ESP32 firmware | vynec hardware setup installs the selected core/libraries |
| GCC + CMake + make | vraylib compiled desktop programs | vendored raylib builds automatically on first import |
The compiler vendors its other dependencies: stb_image / stb_image_write for vimage, and the FTXUI amalgamated header used by vynec init.
Build from source
The build produces vynec in the repository root. To check it:
For the remaining examples, put vynec on PATH or substitute ./vynec (./vynec.exe on Windows) when working in the source checkout. Scoop already provides the PATH shim.
First program
On Windows, run the produced binary as ./hello.exe.
A bare vynec hello.vy compiles and runs the program. Add --ast when you want the interpreter semantics with no C emission, or --emit-c when you want to inspect the generated translation unit.
REPL
The REPL evaluates expressions and declarations against the live symbol table. Use it for quick numeric checks and API discovery; use --ast on real files when you need reproducible interpreter runs.
Arduino toolchain
Verify the AVR tools before the first firmware build:
Then build, flash, and monitor:
--arduino uno alone produces .elf and .hex files. --flash and --monitor invoke arduino-cli after the build. --fqbn overrides the default board identifier (arduino:avr:uno).
ESP32 setup
Install arduino-cli, select a board/preset, and let the hardware workflow install the core and libraries:
Generic ESP32 boards use --board esp32 (or S2/S3/C3/C6 aliases) instead of the HW-950 preset. See ESP32 & Hardware for memory, display, and touch.
Desktop graphics and audio
vraylib supports compiled Windows (MinGW), Linux, and macOS desktop programs. It requires GCC, CMake, and make; rendering needs an OpenGL 3.3 context. Windows additionally needs mingw32-make on PATH. On Ubuntu install libgl1-mesa-dev libx11-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev; xvfb supplies a virtual display for rendering tests. macOS needs command-line developer tools. Building vynec itself from source additionally requires the C++23 toolchain listed above.
With Scoop, scoop install cmake provides CMake. Also ensure mingw32-make is available on PATH; installing the precompiled Vyne package does not remove the native build requirements for graphics applications.
The first build compiles vendored raylib into %LOCALAPPDATA%/vyne/raylib/ on Windows or the XDG/~/.cache/vyne/raylib/ directory on Unix. VYNE_RAYLIB_CACHE overrides that location. Archives are statically linked: no system raylib package or copied DLL is needed. --emit-c / --emit-asm do not build/link archives. --ast, AVR, and ESP32 are not execution targets for vraylib.
See Managed Graphics & Audio, Native Reference, and Graphics Showcase.
CUDA path
The @device<cuda> backend uses the CUDA driver API at runtime. You do not need CUDA headers or link flags for generated programs. CI and ordinary builds do not need nvcc.
If you change the device kernels and need to regenerate the embedded PTX:
The script uses nvcc -ptx -arch=compute_80 and must run inside a Visual Studio developer environment. The generated header is committed, so this is a maintainer step, not an install requirement.
Editors
- VS Code — extension and language server under
editors/vscode/lsp/, with prebuilt VSIX packages. - Neovim — plugin under
editors/nvim/with file-type detection, syntax, and an LSP client. - Doxygen —
doxygen Doxyfilefor the compiler and runtime internals.
Verify the install
For a source checkout with xmake, run:
xmake test requires Python 3.11+, PowerShell 7+ (pwsh on PATH) for benchmark measurement acceptance, and a host C++ compiler. It runs registered compiler, region-safety, golden/native, interpreter, and optimization acceptance. xmake test-all adds the registered numerical/training cases and existing headless graphics suite. The inventory is scripts/tests/suites.toml; missing/empty discovery and mismatched compile/runtime/output expectations fail rather than reporting success. Standalone runtime C and physical firmware/device checks use their separate suites. xmake test-init runs initialization self-tests and fresh desktop/embedded hardware configuration acceptance. Set PYTHON to choose an interpreter and use python scripts/run_tests.py --suite all --list to inspect registered coverage.