mirror of
https://github.com/NixOS/nixpkgs.git
synced 2024-11-25 16:33:15 +00:00
251 lines
6.6 KiB
Python
Executable File
251 lines
6.6 KiB
Python
Executable File
#!/usr/bin/env nix-shell
|
|
#! nix-shell -i "python3 -I" -p "python3.withPackages(p: with p; [ rich structlog ])"
|
|
|
|
from abc import ABC, abstractmethod
|
|
from contextlib import contextmanager
|
|
from pathlib import Path
|
|
from structlog.contextvars import bound_contextvars as log_context
|
|
from typing import ClassVar, List, Tuple
|
|
|
|
import hashlib, logging, re, structlog
|
|
|
|
|
|
logger = structlog.getLogger("sha-to-SRI")
|
|
|
|
|
|
class Encoding(ABC):
|
|
alphabet: ClassVar[str]
|
|
|
|
@classmethod
|
|
@property
|
|
def name(cls) -> str:
|
|
return cls.__name__.lower()
|
|
|
|
def toSRI(self, s: str) -> str:
|
|
digest = self.decode(s)
|
|
assert len(digest) == self.n
|
|
|
|
from base64 import b64encode
|
|
|
|
return f"{self.hashName}-{b64encode(digest).decode()}"
|
|
|
|
@classmethod
|
|
def all(cls, h) -> "List[Encoding]":
|
|
return [c(h) for c in cls.__subclasses__()]
|
|
|
|
def __init__(self, h):
|
|
self.n = h.digest_size
|
|
self.hashName = h.name
|
|
|
|
@property
|
|
@abstractmethod
|
|
def length(self) -> int: ...
|
|
|
|
@property
|
|
def regex(self) -> str:
|
|
return f"[{self.alphabet}]{{{self.length}}}"
|
|
|
|
@abstractmethod
|
|
def decode(self, s: str) -> bytes: ...
|
|
|
|
|
|
class Nix32(Encoding):
|
|
alphabet = "0123456789abcdfghijklmnpqrsvwxyz"
|
|
inverted = {c: i for i, c in enumerate(alphabet)}
|
|
|
|
@property
|
|
def length(self):
|
|
return 1 + (8 * self.n) // 5
|
|
|
|
def decode(self, s: str):
|
|
assert len(s) == self.length
|
|
out = bytearray(self.n)
|
|
|
|
for n, c in enumerate(reversed(s)):
|
|
digit = self.inverted[c]
|
|
i, j = divmod(5 * n, 8)
|
|
out[i] = out[i] | (digit << j) & 0xFF
|
|
rem = digit >> (8 - j)
|
|
if rem == 0:
|
|
continue
|
|
elif i < self.n:
|
|
out[i + 1] = rem
|
|
else:
|
|
raise ValueError(f"Invalid nix32 hash: '{s}'")
|
|
|
|
return bytes(out)
|
|
|
|
|
|
class Hex(Encoding):
|
|
alphabet = "0-9A-Fa-f"
|
|
|
|
@property
|
|
def length(self):
|
|
return 2 * self.n
|
|
|
|
def decode(self, s: str):
|
|
from binascii import unhexlify
|
|
|
|
return unhexlify(s)
|
|
|
|
|
|
class Base64(Encoding):
|
|
alphabet = "A-Za-z0-9+/"
|
|
|
|
@property
|
|
def format(self) -> Tuple[int, int]:
|
|
"""Number of characters in data and padding."""
|
|
i, k = divmod(self.n, 3)
|
|
return 4 * i + (0 if k == 0 else k + 1), (3 - k) % 3
|
|
|
|
@property
|
|
def length(self):
|
|
return sum(self.format)
|
|
|
|
@property
|
|
def regex(self):
|
|
data, padding = self.format
|
|
return f"[{self.alphabet}]{{{data}}}={{{padding}}}"
|
|
|
|
def decode(self, s):
|
|
from base64 import b64decode
|
|
|
|
return b64decode(s, validate = True)
|
|
|
|
|
|
_HASHES = (hashlib.new(n) for n in ("SHA-256", "SHA-512"))
|
|
ENCODINGS = {h.name: Encoding.all(h) for h in _HASHES}
|
|
|
|
RE = {
|
|
h: "|".join(
|
|
(f"({h}-)?" if e.name == "base64" else "") + f"(?P<{h}_{e.name}>{e.regex})"
|
|
for e in encodings
|
|
)
|
|
for h, encodings in ENCODINGS.items()
|
|
}
|
|
|
|
_DEF_RE = re.compile(
|
|
"|".join(
|
|
f"(?P<{h}>{h} = (?P<{h}_quote>['\"])({re})(?P={h}_quote);)"
|
|
for h, re in RE.items()
|
|
)
|
|
)
|
|
|
|
|
|
def defToSRI(s: str) -> str:
|
|
def f(m: re.Match[str]) -> str:
|
|
try:
|
|
for h, encodings in ENCODINGS.items():
|
|
if m.group(h) is None:
|
|
continue
|
|
|
|
for e in encodings:
|
|
s = m.group(f"{h}_{e.name}")
|
|
if s is not None:
|
|
return f'hash = "{e.toSRI(s)}";'
|
|
|
|
raise ValueError(f"Match with '{h}' but no subgroup")
|
|
raise ValueError("Match with no hash")
|
|
|
|
except ValueError as exn:
|
|
logger.error(
|
|
"Skipping",
|
|
exc_info = exn,
|
|
)
|
|
return m.group()
|
|
|
|
return _DEF_RE.sub(f, s)
|
|
|
|
|
|
@contextmanager
|
|
def atomicFileUpdate(target: Path):
|
|
"""Atomically replace the contents of a file.
|
|
|
|
Guarantees that no temporary files are left behind, and `target` is either
|
|
left untouched, or overwritten with new content if no exception was raised.
|
|
|
|
Yields a pair `(original, new)` of open files.
|
|
`original` is the pre-existing file at `target`, open for reading;
|
|
`new` is an empty, temporary file in the same filder, open for writing.
|
|
|
|
Upon exiting the context, the files are closed; if no exception was
|
|
raised, `new` (atomically) replaces the `target`, otherwise it is deleted.
|
|
"""
|
|
# That's mostly copied from noto-emoji.py, should DRY it out
|
|
from tempfile import NamedTemporaryFile
|
|
|
|
try:
|
|
with target.open() as original:
|
|
with NamedTemporaryFile(
|
|
dir = target.parent,
|
|
prefix = target.stem,
|
|
suffix = target.suffix,
|
|
delete = False,
|
|
mode="w", # otherwise the file would be opened in binary mode by default
|
|
) as new:
|
|
tmpPath = Path(new.name)
|
|
yield (original, new)
|
|
|
|
tmpPath.replace(target)
|
|
|
|
except Exception:
|
|
tmpPath.unlink(missing_ok = True)
|
|
raise
|
|
|
|
|
|
def fileToSRI(p: Path):
|
|
with atomicFileUpdate(p) as (og, new):
|
|
for i, line in enumerate(og):
|
|
with log_context(line = i):
|
|
new.write(defToSRI(line))
|
|
|
|
|
|
_SKIP_RE = re.compile("(generated by)|(do not edit)", re.IGNORECASE)
|
|
_IGNORE = frozenset({
|
|
"gemset.nix",
|
|
"yarn.nix",
|
|
})
|
|
|
|
if __name__ == "__main__":
|
|
from sys import argv
|
|
|
|
logger.info("Starting!")
|
|
|
|
def handleFile(p: Path, skipLevel = logging.INFO):
|
|
with log_context(file = str(p)):
|
|
try:
|
|
with p.open() as f:
|
|
for line in f:
|
|
if line.strip():
|
|
break
|
|
|
|
if _SKIP_RE.search(line):
|
|
logger.log(skipLevel, "File looks autogenerated, skipping!")
|
|
return
|
|
|
|
fileToSRI(p)
|
|
|
|
except Exception as exn:
|
|
logger.error(
|
|
"Unhandled exception, skipping file!",
|
|
exc_info = exn,
|
|
)
|
|
else:
|
|
logger.info("Finished processing file")
|
|
|
|
for arg in argv[1:]:
|
|
p = Path(arg)
|
|
with log_context(arg = arg):
|
|
if p.is_file():
|
|
handleFile(p, skipLevel = logging.WARNING)
|
|
|
|
elif p.is_dir():
|
|
logger.info("Recursing into directory")
|
|
for q in p.glob("**/*.nix"):
|
|
if q.is_file():
|
|
if q.name in _IGNORE or q.name.find("generated") != -1:
|
|
logger.info("File looks autogenerated, skipping!")
|
|
continue
|
|
|
|
handleFile(q)
|