RE: https://mastodon.social/@fgregg/111429385063753445

i revisit this problem every few years to learn something new.

this time i used PuLP.

my branch and bound rust version took 10 minutes to solve the problem at K=30 (or more accurately to prove it had found an optimal solution).

this takes less than a second.

https://gist.github.com/fgregg/45f0c6e7f991789cd3f10414958b0ef8

i'm pretty close to refactor of datasette to allow pluggable backends so duckdb can get a first class treatment (unlikely to ever make it upstream), but i'm vacillating on what the interface should be.

datasette has enough of an installed based that users expect to write sqlite-flavored sql

so should users write sqlite-flavored sql (full power, less magic), or i should users continue to write in the sqlite dialect, and then translate that to duckdb's dialect transparently.

thoughts?

here's a python rewrite of a small but powerful tool by juho snellman that has saved me many hours over the years.

ported to python to make it a bit easier to install in my typical workflow, and i also made some adjustments for ergonomics.

https://github.com/fgregg/json-to-multicsv

here's a package that lets you use postgresql's full text search as a search backend for haystack in your django project.

https://pypi.org/project/postgres-fts-backend/

why i might be interesting:

1. postgresql fts has gotten pretty good
2. any complicated enough search implementation will reinvent most of haystack
3. even if you ultimately want a dedicated search provider, it's nice to get started without and additional service to run.

#python #django

i wrote an explanation about what salesforce really is: a cursed way to write web applications.

https://bunkum.us/2025/12/18/what-is-salesforce