uf90: Unicode mathematical notation for portable Fortran

Português (Brasil)

uf90 lets you write mathematical Unicode identifiers in Fortran projects. It translates *.f90u files into ordinary ASCII *.f90 files, which are then compiled by fpm and your existing Fortran compiler.

The project is at an early stage (0.1.x). It is a source-to-source translator, not a compiler or a Fortran language extension.

Quick start with fpm

Install the version published on PyPI with pipx:

pipx install uf90

If your shell cannot find uf90 or fpm-uf90 after installation, run pipx ensurepath and open a new terminal.

The installation provides both the uf90 command and the fpm-uf90 plugin. With fpm and fpm-uf90 available on PATH, run the plugin from a project containing an fpm.toml file:

fpm uf90 build
fpm uf90 run
fpm uf90 test

The plugin finds *.f90u files in the project, updates their corresponding *.f90 files, and forwards the command to fpm. Additional arguments are forwarded as well:

fpm uf90 build --profile release

In practice, this:

fpm uf90 build

is equivalent to:

uf90 sync
fpm build

Minimal example

Write source files in the directories recognized by fpm, using the .f90u extension. This example rotates the point (x₀, y₀) by an angle θ and stores the result in (x₁, y₁):

! app/main.f90u
program rotation
  implicit none
  real :: π, θ
  real :: x₀, y₀, x₁, y₁

  π = acos(-1.0)
  θ = π / 4.0

  x₀ = 1.0
  y₀ = 0.0

  x₁ = cos(θ) * x₀ - sin(θ) * y₀
  y₁ = sin(θ) * x₀ + cos(θ) * y₀

  print *, x₁, y₁
end program rotation

Running fpm uf90 run generates:

! app/main.f90
program rotation
  implicit none
  real :: uc_pi, uc_theta
  real :: x_0, y_0, x_1, y_1

  uc_pi = acos(-1.0)
  uc_theta = uc_pi / 4.0

  x_0 = 1.0
  y_0 = 0.0

  x_1 = cos(uc_theta) * x_0 - sin(uc_theta) * y_0
  y_1 = sin(uc_theta) * x_0 + cos(uc_theta) * y_0

  print *, x_1, y_1
end program rotation

Strings and comments are preserved. The .f90 file is generated output; edit the corresponding .f90u source instead.

Supported notation

• Greek letters: α → uc_alpha, Δt → uc_deltat;
• calculus prefixes: ∂x → partial_x, ∇φ → nabla_phi;
• numeric subscripts: T₁₀₀ → T_100;
• alphanumeric subscripts: Eₙ → E_n, Aᵢⱼ → A_ij;
• superscripts: c² → c_p2.

With uf90-ls, type a LaTeX-style symbol name such as \alpha, \partial, or \nabla and accept the completion to insert α, ∂, or ∇. This works in both VS Code and Neovim; the generated Neovim setup also maps Ctrl+Space to manual completion.

Subscripts use familiar LaTeX-style input: accepting completion for x_n, T_{100}, or A_ij produces xₙ, T₁₀₀, or Aᵢⱼ. The menu opens after _, and closing } triggers braced-sequence completion. Only characters with real Unicode subscript forms are offered.

Unicode does not provide subscript forms for every letter. The current set includes ₐ ₑ ₕ ᵢ ⱼ ₖ ₗ ₘ ₙ ₒ ₚ ᵣ ₛ ₜ ᵤ ᵥ ₓ, ₔ, digits, and the Greek forms ᵦ ᵧ ᵨ ᵩ ᵪ.

Commands without fpm

For direct use or integration with other workflows:

uf90 sync [directory]
uf90 check [directory]
uf90 translate source.f90u [-o source.f90]
uf90 fortls-config [directory]
uf90 editor-config {vscode,neovim} [-o file]

uf90 sync uses .uf90-manifest.json as an incremental cache. uf90 check does not modify files and returns a non-zero status when generated output needs to be updated.

Editors

The experimental 0.2 uf90-ls proxy lets fortls analyze generated .f90 files while editors continue displaying .f90u. Setup guides are available for VS Code and Neovim. The design document tracks the architecture, acceptance criteria, and remaining scope.

From a 0.2 development checkout, install the proxy and its tested fortls in one isolated environment with pipx install --force --editable '.[lsp]'. Then use uf90 editor-config vscode or uf90 editor-config neovim to generate a ready configuration.

Project status and scope

The current version is intended for experimenting with the notation in fpm projects and for finding translation cases that are not covered yet. The interface and mappings may change while the project remains below version 1.0.

uf90 currently:

• translates identifiers without changing text inside strings and comments;
• generates ASCII Fortran that continues through the regular fpm workflow;
• runs automated tests on Linux, macOS, and Windows with Python 3.10 through 3.14.

It does not validate program semantics, replace the Fortran compiler, or depend on LLVM or MLIR.

Examples and development

The examples directory contains six progressive fpm projects. To run them from a repository clone:

python3 -m pip install -e '.[test]'
python3 -m pytest
python3 examples/run_all.py

The maintainer release procedure is documented in docs/RELEASING.md.