Binging on EuroRust content right now, ran into this talk by Charlie Marsh about UV. Really great talks, with some surprises about how UV works and performance tweaks

Finally watched this! • The cache design makes total sense and it drives me nuts CPython doesn't have reflink support yet. • I'm feel like I'm missing something fundamental because I thought the view from the pip side was that using a sat solver was 'impossible' due to the dynamic metadata. (Sure there isn't a simple answer to backtracking vs sat approaches) • I'm surprised by how much they credit "small" optimizations. (per byte memory layout, etc)

I learned the power of small, frequent optimizations at my first job many moons ago. One of the guys sitting next to me was messing around with the custom string class this company made (this was an old, old, old C/C++ product), and for our use case, it turned out that a small 1 line tweak to our string compare was an 80-90% speed improvement or something equally crazy. And given the speed of computers, AND how frequently huge strings were compared, this actually was a genuine perf improvement that could be felt in some parts of the app