Skip to content

Commit 5674df0

Browse files
committed
Add CLAUDE.md describing the development checks
Claude-Session: https://claude.ai/code/session_01JZGNmz4nWXtZMaKkbMEfPy
1 parent 295ea83 commit 5674df0

1 file changed

Lines changed: 42 additions & 0 deletions

File tree

‎CLAUDE.md‎

Lines changed: 42 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,42 @@
1+
# CLAUDE.md
2+
3+
Guidance for AI coding agents (and human contributors) working in this repository.
4+
5+
## Setup
6+
7+
The project uses [uv](https://docs.astral.sh/uv/). `uv sync` creates the environment
8+
(`uv sync --group docs` adds the Sphinx toolchain).
9+
10+
## Checks to run before considering a change done
11+
12+
```sh
13+
uv run ruff format . # formatting
14+
uv run ruff check . # lint (ruff's default rule set)
15+
uv run ty check # type checking
16+
uv run vulture # dead code (min_confidence = 80, see pyproject.toml)
17+
uv run pytest # tests (runs in parallel with xdist)
18+
uv run sphinx-build -W docs docs/_build # docs, if docstrings or docs/ changed
19+
```
20+
21+
CI runs the same checks. `pre-commit install` runs ruff on each commit.
22+
23+
### Tests must stay fast
24+
25+
The whole suite should finish in seconds. Slow tests are a bug, not a cost of doing
26+
business: use small problem sizes and a capped `maxiter`, and assert the same
27+
properties. `conftest.py` pins BLAS to one thread because the test matrices are tiny and
28+
multithreaded BLAS makes them dramatically slower. Warnings are errors
29+
(`filterwarnings = ["error", ...]`); fix the cause, or add a narrowly-scoped,
30+
commented ignore for third-party warnings.
31+
32+
### Dead code -- vulture
33+
34+
Anything vulture reports at 80%+ confidence should be removed. Public methods and
35+
functions (the library API) intentionally have no in-repo callers and are not
36+
reported below that threshold; do not delete them just because they are unused here.
37+
38+
## Conventions
39+
40+
- Docstrings are NumPy style (they feed Sphinx autodoc).
41+
- Missing data is represented by `NaN`; solvers must ignore it, not impute it.
42+
- Public API lives in `tensorpack/__init__.py` (`__all__`).

0 commit comments

Comments
 (0)