tdvutil

A grab bag of little Python helpers I kept copy-pasting between projects, so I finally stuck them in one place.

Heads up: this is mostly here for my own convenience. You're welcome to use it, but I make no promises about keeping the API stable. Things can and will get renamed, changed, or ripped out whenever it suits me, even in minor releases. If you depend on it, pin a version.

Installing

pip install tdvutil

Needs Python 3.9+.

What's in here

From tdvutil directly:

• ppretty: pretty-prints objects
• alintrospect / whatis: a very janky way to poke at an object's class hierarchy and see where its methods and attributes come from
• now / nowf: current time, with less typing
• pathfix: path cleanup
• sec_to_hms, sec_to_timecode, sec_to_shortstr, hms_to_sec, timecode_to_sec: converting between seconds and human-readable times

Plus a couple of submodules:

• tdvutil.argparse: CheckFile and NegateAction, for argparse
• tdvutil.llmcost: pricing data for a bunch of LLM models, and a way to figure out what a call cost

from tdvutil import sec_to_hms
from tdvutil.llmcost import get_model

sec_to_hms(3725)  # '01:02:05.000'

cost = get_model("gpt-4o").cost(input_tokens=10_000, output_tokens=2_000)
print(cost.total_cost)  # 0.045

Hacking on it

Uses uv. make localdev sets things up, make test runs the tests, and make dist builds.

GitHub Actions runs the tests and builds packages for PRs targeting main and pushes to main, without publishing. Maintainer approval is required to run outside contributors' PR workflows.

Pushing a v* tag makes GitHub Actions build it and publish to TestPyPI, then PyPI. Manually running the workflow on a v* tag does the same, including publishing to real PyPI. Manual branch runs just run the checks and save the built packages as workflow artifacts; they don't publish.

License

GPLv3, see LICENSE.txt.