Why no interest in python-native accelerators?

[Scott Gorlin]

Hi,

Seeing lots of interest in 'making beancount faster' over the years by porting to C++/Rust/etc.  While this might be a good idea for some, I have to assume many of us like this project because we are python-first developers, and leveraging tools like Numba, Cython, or similar high-performance native python tooling will be orders of magnitude simpler from a porting a python code base & maintenance PoV as well as more amenable to collaboration.  Going to an entirely different language feels, to me, like an existential crisis for a python user base!

I might be in a position to try some experiments over the next several months, would love to know if any of these strategies have been considered, or if there is a good strategy for why moving to other languages makes sense for this project.

(as a side note, I have found the lack of developer documentation for building and contributing to BC source a challenge - I have been coding in python for 20 years and never heard of meson... contributing guidelines would be appreciated!!)

[Martin Blais]

On Sat, Feb 7, 2026 at 2:26 PM Scott Gorlin <scott...@gmail.com> wrote:

Hi,

Seeing lots of interest in 'making beancount faster' over the years by porting to C++/Rust/etc.  While this might be a good idea for some, I have to assume many of us like this project because we are python-first developers, and leveraging tools like Numba, Cython, or similar high-performance native python tooling will be orders of magnitude simpler from a porting a python code base & maintenance PoV as well as more amenable to collaboration.  Going to an entirely different language feels, to me, like an existential crisis for a python user base!

Rust + Python has become the killer combo.

There are already some rewrites out these, see contrib document.

I might be in a position to try some experiments over the next several months, would love to know if any of these strategies have been considered, or if there is a good strategy for why moving to other languages makes sense for this project.

I think the general idea is, knock yourself out.

I've become too busy with the explosion in machine learning over the last few years and the appearance of children in my life to put a significant amount of cycles into it.

When retirement comes - who knows, maybe not too far - there will be a second wind.

I honestly think we're getting pretty close to a one shot conversion with AI, but me, well I'm particular about the output, so it'll take a few weeks probably to get something I like.

 

(as a side note, I have found the lack of developer documentation for building and contributing to BC source a challenge - I have been coding in python for 20 years and never heard of meson... contributing guidelines would be appreciated!!)

Jusrt ask Gemini or Claude and read the Makefile and spend 30 minutes reading around the codebase. It's a pretty small codebase...

 

--
You received this message because you are subscribed to the Google Groups "Beancount" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beancount+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/beancount/11fe70d9-99a8-4e4a-831b-f7670c3ca90bn%40googlegroups.com.