Python snapshots expose a subtle problem with reused randomness

Hood Chatham’s Python summit proposal would speed startup by restoring memory captured after interpreter initialization. But that also restores values meant to change between processes, including the random salt protecting hash tables from collision attacks. The discussion explores an explicit initialization or reinitialization step so the runtime and libraries can regenerate state that must not be shared.

Python Insider · Sep 30, 2026

Why it matters

Startup optimizations must preserve process-specific assumptions, not just restore working objects.

More context

Read the design discussion if you embed Python, use Pyodide or work on edge runtimes. Watch the proposal rather than treating it as an available CPython switch; the article also explains why lazy imports do not solve the same problem.

Keep in mind

This is a design discussion, not a released CPython feature. The reported startup measurement concerns a specific Pyodide example; it does not establish gains for arbitrary Python services. 1min has not reproduced it.

Original read