workloads

Unsupported & Fallback

How Pyvorin handles code it cannot compile.

Published May 30, 2026

Execution Tiers

Tier	Meaning
`COMPILED_FULL`	Entire function compiled to native machine code.
`COMPILED_PARTIAL`	Some regions compiled; others execute in CPython.
`COMPATIBILITY_EXECUTED`	Entire function ran in CPython (fallback).
`UNSUPPORTED`	Could not compile; compatibility execution used.

Currently Unsupported Constructs

Construct	Fallback Behaviour
`eval` / `exec` / `compile`	Dynamic code generation falls back to CPython.
Custom metaclasses	Compatibility execution.
Complex multiple inheritance	Compatibility execution.
Direct mutation of C extension internals	Calls into NumPy, Pandas, etc. fall back gracefully.
GPU training / CUDA	Not applicable; use your ML framework directly.
Network-bound services	Compatibility execution (Python overhead is not the bottleneck).

How Fallback Works

Before compilation, the AST is scanned for unsupported nodes. If any are found, the function is marked for fallback and runs in CPython transparently. Fallback is never silent — Pyvorin records the reason in the execution status.

Minimising Fallback

Use supported types: int, float, list, dict, str, tuple.
Avoid unsupported constructs in hot paths.
Keep types stable across loop iterations.
Run pyvorin scan your_script.py to identify risks.