Skip to content
Viscosity

Python

Viscosity reads a numeric subset of Python:

  • @dataclass classes with annotated fields
  • lists built with [] and .append(Record(...))
  • for loops over range(...) or over a list, while and if
  • functions, global, arithmetic, math.*

Files outside the subset are left exactly as they are (this is not an error).

Both engines read Python: the TypeScript CLI and the Rust CLI compile .py files with the same results, byte for byte.

It lowers to the same IR and runs the same analyses as JavaScript. Typed array('d') columns are used only for fields that provably always hold floats: in Python an int would come back as a float and print differently.

Why it is opt-in: in CPython, reading an array('d') element creates a new float object each time, so columns measured slower per element than dataclass objects (see Benchmarks). They used about 5× less memory. Annotate a list with # @viscosity soa when memory matters, or pass --python-auto to transform every safe list.