unit-syntax 0.1.0

No project description provided

unit-syntax extends the Python language in Jupyter/IPython to support expressions with physical units:

>>> import unit_syntax.ipython
>>> speed = 5 meters/second
>>> 2 seconds * speed
10 meter

Behind the scenes this is translated into standard Python that uses the excellent Pint units library.

Getting Started

Install the package with pip install unit_syntax. Tip: In Google Colab, you can run this directly in a cell by prefixing the command with !.

To enable unit-syntax in a Jupyter/IPython session run:

import unit_syntax
unit_syntax.enable_ipython()

Note that in Jupyter this must be run in its own cell before any units expressions are evaluated.

Syntax

Units apply to the preceding value (literal, variable, function call or indexing), and have higher precedence than other operators:

x * 1.21 gigawatts

This is equivalent to x * (1.21 gigawatts), and desugars to something like x * Quantity(1.21, "gigawatts"). The high precedence means units apply to the literal not the whole expression.

Values can be converted to another measurement system:

(88 miles / hour) furlongs / fortnight

Pint transparently supports numpy when available:

velocity = [5, 7] meters/second**2
location = velocity * 2 seconds
distance_traveled = numpy.linalg.norm(location)

Units must start with an identifier

10 cm * sin(45 degrees)

This is a syntax error because sin is parsed as part of the units; to resolve add parentheses: (10 cm) * sin(45 degrees)

The units term follows this grammar:

units:
    | '(' units ')'
    | NAME '/' units
    | NAME '*' units
    | NAME units
    | NAME '**' NUMBER
    | NAME

The syntax takes advantage of the fact that that in python its illegal for a NAME to follow a "primary" (literal, function call etc), so there's no ambiguity.

Why? How?

I like using Python+Jupyter Notebook as a calculator for physical problems and often wish it had the clarity and type checking of explicit units. Pint is great, but (IMO) its necessary verbosity makes it hard to see the underlying calculation that's going.

unit-syntax is an IPython/Jupyter custom input transformer that rewrites expressions with units into calls to pint.Quantity.

This is possible without ambiguity in the python grammar because it's otherwise invalid for a "primary" (literal, function call etc) to be immediately followed by

unit-syntax cannot (currently) be used for standalone python scripts outside of IPython/Jupyter, but that's in principle possible through meta_path import hooks.

Prior Art

The immediate inspriration of unit-syntax is a language called Fortress from Sun Microsystems. Fortress was intended as a modern Fortran, and had first-class support for units in both the syntax and type system.

F# (an OCaml derivative from Microsoft) also has first class support for units.

The Julia package Unitful.jl

Open questions and future work

• Fortress uses an in operator to apply units to a non-literal value, e.g x in meters. This has the advantage of being unambiguous regardless of parenthesization. In python this would conflcit with value in [a, b, c], but as is

• Test against various ipython and python versions

• Support standalone scripts through sys.meta_path

• Check units at parse time

• Unit type hints, maybe checked with @runtime_checkable. More Pint typechecking discussion

• Pint does not do the right thing when applied to generator expressions, e.g (a for a in range(0, 4)) meters

• Demo colab notebook: https://colab.research.google.com/drive/1PInyLGZHnUzEuUVgMsLrUUNdCurXK7v1#scrollTo=JszzXmATY0TV

Development

To regenerate the parser:

python -m pegen grammar.txt -o unit_literals/parser.py

Running tests:

 $ poetry install --with dev
 $ poetry run pytest

## Future work and open questions

 * Move to tree-sitter so there's a chance of getting syntax highlighting
 * Jupyter syntax checks
 * Typography of output
 * Test against various ipython and python versions
 * Support standalone scripts through sys.meta_path
 * Check units at parse time
 * Unit type hints, maybe checked with [@runtime_checkable](https://docs.python.org/3/library/typing.html#typing.runtime_checkable).  More Pint typechecking [discussion](https://github.com/hgrecco/pint/issues/1166) 
 * Does not do the right thing when applied to generator expressions, e.g `(a for a in range(0, 4)) meters`
 * Parenthisized units expressions
 * Demo colab notebook: https://colab.research.google.com/drive/1PInyLGZHnUzEuUVgMsLrUUNdCurXK7v1#scrollTo=JszzXmATY0TV