I got up to 15% speedup using Nuitta for a tool I have. It also seems very compatible, handling almost every code I throw at it.

Nuitka does actually use libpython, so it should be expected to be highly compatible.

A sample template project using Nuitka if you want to play around with it: https://github.com/zedr/aiohttp-nuitka-docker

To deploy a Python program as something that looks just like if a C program + a C library was deployed, is a nice option to have.

Nuitka provides this.

The speed is also better than a regular Python application.

Yes, and if they can keep the project healthy, we can expect it to continue to increase in speed, in fact according to the authors the speed improvements are just starting.

Earlier discussion (12 months ago):

https://news.ycombinator.com/item?id=15354613

There was an interesting podcast episode with Nuitka creator, Kay Hayden, a month ago on TalkPythonToMe : https://talkpython.fm/episodes/download/172/nuitka-a-full-py...

Kay Hayden does not bother to appear nice, and has some strong opinions on things, I liked it a lot

I just tried it yesterday on one Django project. The DB migration failed. It seems that when doing DB migration, Django actually looks for .py files which don't exist anymore with nuitka.

Yes; any framework or tooling that relies on runtime imports (eval is OK, just not eval-from-file) is not going to work with rather a lot of compiled-python programs.

It's possible to build shims that support "import module if running interpreted python; just point to already-loaded/compiled code (or eval a string, ew, but if you must) if not", but many large projects do not do that. It is likely not too painful to implement/send a PR for though.

Surprising since it advertises itself as "It is the extremely compatible Python compiler."

There's a bunch of discussion and links in this[1] SO question. It looks like numpy's size becomes an issue.

It appears that some people do use the two projects together, though.

[1] https://stackoverflow.com/questions/37713589/nuitka-error-ca...

Even if it the end it spits out an exe, it's 100s of MB. This makes it a non-starter for short lived numerical programs.

Are you using --standalone flag? Nuitka scans for dependencies recursively, that's why output bundle's so big.

I don't understand why you've that large .exe, a bunch of DLLs is what should be heavy.