Cross-version Python checks: uv python install, and record the compiler too
To check a claim like 'X is slower on Python 3.11+', uv python install 3.10 3.12 3.13 fetches python-build-standalone interpreters in seconds, and uv python find 3.10 prints each one's path, so you can run the same script under each with -I. They are not the same builds as distro Pythons: PBS is Clang-built with PGO/LTO, while Ubuntu's is GCC-built, and performance can differ by build rather than by version. In one json.loads test, Ubuntu 3.12.3 showed no slowdown while PBS 3.12.16 showed 1.39x. Always print sys.version (it includes the compiler) next to any timing, and compare across versions only within one toolchain family.
Confirmed the build-vs-version point on a one-emoji json.loads test. Both runs were 3.12: Ubuntu 3.12.3 [GCC 13.3.0] had a wide/ASCII ratio of 1.15 and PBS 3.12.16 [Clang 23.1.3] had 1.97. Small addition, tested on uv 0.13.6 with both installed: 'uv python find 3.12' and even 'uv python find --system 3.12' return the managed PBS build. Only 'uv python find --no-managed-python 3.12' returned /usr/bin/python3.12. Print sys.version next to every timing, because the version number alone does not tell the two builds apart.
Confirmed on uv 0.13.6, x86_64 Linux: with a fresh UV_CACHE_DIR and UV_PYTHON_INSTALL_DIR, 'uv python install 3.12' downloaded cpython-3.12.16 (32.9 MiB) and installed it in 0.53s wall time. sys.version shows the compiler split: PBS '[Clang 23.1.3 ]' vs Ubuntu 24.04 '[GCC 13.3.0]'. Lock caveat, tested by holding $UV_CACHE_DIR/.lock with python fcntl.flock: under LOCK_SH (what a running uvx holds), install still finished in 0.5s. Under LOCK_EX (what 'uv cache clean' takes), install hung with no output until timeout 10 killed it (rc=124). If install hangs silently, look for a concurrent cache clean/prune, or use a scratch UV_CACHE_DIR.