MesaPy: A Memory-Safe Python Implementation based on PyPy (2018) | Hacker News Reader

MesaPy: A Memory-Safe Python Implementation based on PyPy (2018)

github.com

119 pointsospider17 comments

I gave this a spin a while back but, unfortunately didn’t get very far since the software and dependencies have grown long in the byte. Since then, I’ve found RustPython [0] which is progressing toward feature parity with CPython but entirely written in Rust (!). A side benefit is that it compiles to Web Assembly, so if you could sandbox it without too much extra overhead.

[0] https://github.com/RustPython/RustPython

Since a lot of CPython’s power comes from its C API (think for example NumPy), this is unfortunately pretty far from feature parity with CPython, unless I’m missing something.

This. Native python is very slow so almost any compute heavy python work will offload work to external native code. While the a RustPython would help on the python side, this wouldn't help the external native calls which end up doing most of the work.

> Since then, I’ve found RustPython [0] which is progressing toward feature parity with CPython but entirely written in Rust (!). A side benefit is that it compiles to Web Assembly, so if you could sandbox it without too much extra overhead.

It's now possible to run JupyterLab entirely within a browser with jyve (JupyterLab + pyodide) https://github.com/iodide-project/pyodide/issues/431

Pyodide:

> Pyodide brings the Python runtime to the browser via WebAssembly, along with the Python scientific stack including NumPy, Pandas, Matplotlib, parts of SciPy, and NetworkX. The packages directory lists over 35 packages which are currently available.

Is the RustPython WASM build more performant or otherwise preferable to brython or pyodide?

So did they actually find any memory leaks in the C python libs or are they just assuming they exist because it's written in C?

It's pretty much the latter, as far as I can tell?

This is part of Baidu X-Lab's efforts to make MesaLock Linux (https://github.com/mesalock-linux). They are creating memory safe versions of tools for their distribution.

Baidu has even made a memory safe version of TPM (https://medium.com/baiduxlab/mesatee-security-formal-verific...).

How does this handle “pip install” and third party code? It seems the memory safety guarantees depend on reimplementing things in Rust (for stdlib modules), which would be interesting in and of itself, as well as formal verification.

Looks like it’s only python 2.7? Is there any plans to move this to 3.7+?

PyPy is still not fully Py3K ready. I'd expect MesaPy to only have built on a stable version of PyPy.

It is, insofar as any PyPy is ever "fully ready."

The latest PyPy3 release targets 3.5.3.

PyPy 3.6.1 was released and is very usable, the next release will be 3.6.9

3.6 releases are "beta quality," which I agree is very usable but not "fully ready." as per grandparent.

Too bad there's nothing about threaded / parallel execution.

It probably has the same limitations as PyPy.

Interesting project. Who’s sponsoring this?

The mesalock project is run by Baidu's X-lab.

Reply on news.ycombinator.com