WildAi
f769a95a1f
The tree fragment is now matched as a path segment, so an identifier that merely carries the same letters (load_device, __device__) cannot trip the check, and the POSIX home roots are checked outside the vendored upstream tree. Two self-tests prove the tripwire catches a POSIX home path and does not fire on device identifiers.
133 lines
4.9 KiB
Python
133 lines
4.9 KiB
Python
"""Repo-wide tripwire: no tracked file may carry a machine-local path.
|
|
|
|
The rule is the owner's: the published package must not name a local drive, a
|
|
local directory tree or a local interpreter. The forbidden fragments are built
|
|
from parts, because this file is itself tracked and would otherwise trip the
|
|
very check it performs.
|
|
|
|
The scan covers the **tracked** files only: the internal development documents
|
|
live under the untracked ``.dev`` tree and are never part of the package.
|
|
|
|
Two fragment sets are checked. ``FORBIDDEN`` applies to the whole tracked
|
|
tree; ``POSIX_FORBIDDEN`` applies outside the vendored tree only, because the
|
|
single POSIX literal inside it is upstream code that this repository does not
|
|
own.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import subprocess
|
|
from pathlib import Path
|
|
from typing import Iterable, Iterator
|
|
|
|
import pytest
|
|
|
|
REPO_ROOT = Path(__file__).resolve().parent.parent
|
|
|
|
#: The forbidden fragments, built from parts so that this file does not
|
|
#: contain them literally. A local tree is matched as a **path segment**: the
|
|
#: separators belong to the fragment, so an identifier that merely carries the
|
|
#: same letters - a ``load_device`` call, a ``__device__`` kernel attribute -
|
|
#: is not a hit.
|
|
FORBIDDEN = (
|
|
"C:" + "\\",
|
|
"C:" + "/",
|
|
"/" + "_" + "Dev" + "/",
|
|
"/" + "_" + "dev" + "/",
|
|
"\\" + "_" + "Dev" + "\\",
|
|
"\\" + "_" + "dev" + "\\",
|
|
)
|
|
|
|
#: POSIX home roots. They are forbidden outside the vendored tree only (see
|
|
#: ``VENDORED_PREFIX``): the one literal inside it is upstream code.
|
|
POSIX_FORBIDDEN = (
|
|
"/" + "home/",
|
|
"/" + "Users/",
|
|
"/" + "root/",
|
|
)
|
|
|
|
#: The vendored third-party tree. It is upstream code that this repository
|
|
#: does not own, so the POSIX check skips it.
|
|
VENDORED_PREFIX = "modules/deps/"
|
|
|
|
#: Suffixes that hold binary content and are skipped.
|
|
BINARY_SUFFIXES = frozenset(
|
|
{".png", ".jpg", ".jpeg", ".gif", ".webp", ".ico", ".pth", ".pt", ".ckpt", ".bin", ".safetensors"}
|
|
)
|
|
|
|
|
|
def _tracked_files() -> list[str]:
|
|
"""The tracked file list (``git ls-files``), relative to the repository root."""
|
|
try:
|
|
result = subprocess.run(
|
|
["git", "ls-files", "-z"],
|
|
cwd=REPO_ROOT,
|
|
capture_output=True,
|
|
text=True,
|
|
check=True,
|
|
)
|
|
except (OSError, subprocess.CalledProcessError) as exc: # pragma: no cover
|
|
pytest.skip(f"git is not usable in this checkout: {exc}")
|
|
return [name for name in result.stdout.split("\0") if name]
|
|
|
|
|
|
def _tracked_texts() -> Iterator[tuple[str, str]]:
|
|
"""Yield ``(name, text)`` for every tracked file that can be read as text."""
|
|
for name in _tracked_files():
|
|
path = REPO_ROOT / name
|
|
if path.suffix.lower() in BINARY_SUFFIXES:
|
|
continue
|
|
try:
|
|
text = path.read_text(encoding="utf-8", errors="ignore")
|
|
except OSError: # pragma: no cover - a tracked file that cannot be read
|
|
continue
|
|
yield name, text
|
|
|
|
|
|
def _offenders(fragments: Iterable[str], texts: Iterable[tuple[str, str]]) -> list[str]:
|
|
"""The tracked files whose text carries one of ``fragments``."""
|
|
return [
|
|
f"{name} contains {fragment!r}"
|
|
for name, text in texts
|
|
for fragment in fragments
|
|
if fragment in text
|
|
]
|
|
|
|
|
|
@pytest.mark.unit
|
|
def test_no_tracked_file_carries_a_machine_local_path():
|
|
"""No tracked file names a local drive, a local tree or the local interpreter."""
|
|
offenders = _offenders(FORBIDDEN, _tracked_texts())
|
|
assert not offenders, "machine-local paths in tracked files:\n" + "\n".join(offenders)
|
|
|
|
|
|
@pytest.mark.unit
|
|
def test_no_tracked_file_outside_the_vendored_tree_carries_a_posix_path():
|
|
"""No tracked file of ours names a POSIX home root."""
|
|
texts = ((name, text) for name, text in _tracked_texts() if not name.startswith(VENDORED_PREFIX))
|
|
offenders = _offenders(POSIX_FORBIDDEN, texts)
|
|
assert not offenders, "POSIX machine paths in tracked files:\n" + "\n".join(offenders)
|
|
|
|
|
|
@pytest.mark.unit
|
|
def test_a_posix_machine_path_is_caught():
|
|
"""A tracked file that names a POSIX home root makes the tripwire fail."""
|
|
text = 'model_dir = "' + "/" + "home" + "/someone" + '/models"'
|
|
assert _offenders(POSIX_FORBIDDEN, [("example.py", text)])
|
|
|
|
|
|
@pytest.mark.unit
|
|
def test_a_device_identifier_is_not_a_machine_local_path():
|
|
"""An identifier that carries the same letters is not a local tree."""
|
|
identifiers = ("load_device", "offload_device", "__device__", "get_torch_device", "is_dev_only")
|
|
texts = [(f"example_{index}.py", f"value = {name}") for index, name in enumerate(identifiers)]
|
|
assert not _offenders(FORBIDDEN, texts)
|
|
|
|
|
|
@pytest.mark.unit
|
|
def test_the_internal_documents_are_not_tracked():
|
|
"""The development documents live outside the repository's index."""
|
|
tracked = _tracked_files()
|
|
offenders = [name for name in tracked if name.startswith("docs/") or name.startswith(".dev")]
|
|
assert not offenders, "internal documents must not be tracked:\n" + "\n".join(offenders)
|