AstroM3Dataset / utils /parallelzipfile.py
MeriDK's picture
Added splits by seed abd subsample
9fa663b
raw
history blame
8.77 kB
"""
Like Python's zipfile, but using mmap instead of a regular file.
Helpful resources:
http://web.archive.org/web/20210219055401/https://blog.yaakov.online/zip64-go-big-or-go-home/
http://web.archive.org/web/20210310084602/https://users.cs.jmu.edu/buchhofp/forensics/formats/pkzip.html
http://web.archive.org/web/20210225050454/https://pkware.cachefly.net/webdocs/casestudies/APPNOTE.TXT
"""
import bz2
import lzma
import mmap
import struct
import zlib
from typing import IO, Dict, List, Optional, no_type_check
class ZipInfo:
"""Contains information about a file stored in a ParallelZipFile."""
filename: str
header_offset: int
CRC: int
compress_size: int
file_size: int
def __init__(
self,
filename: str,
header_offset: int,
CRC: int,
compress_size: int,
file_size: int,
) -> None:
self.filename = filename
self.header_offset = header_offset
self.CRC = CRC
self.compress_size = compress_size
self.file_size = file_size
def is_dir(self) -> bool:
"""Directories in a zip file should end with '/'."""
return self.filename.endswith("/")
def _read_eocd_mmap(m: mmap.mmap) -> Dict[str, ZipInfo]:
# Read end-of-central-directory (EOCD) from mmaped zipfile.
# TODO Can zip64 EOCDs be larger?
max_eocd_size = 22 + 65536
end = m[-max_eocd_size:]
# Scan backwards until EOCD signature is found
# TODO this could fail if a comment contains the EOCD signature.
# Should employ sanity check to verify that an actual EOCD was found.
offset32 = end.rfind(b"PK\5\6")
assert offset32 >= 0
eocd = struct.unpack("<4sHHHHIIH", end[offset32 : offset32 + 22])
(
signature,
num_disks,
num_disks2,
num_files,
num_files2,
directory_size,
directory_offset,
comment_length,
) = eocd
assert signature == b"PK\5\6"
# Read zip64 end of central directory if locator exists
locator_offset = end.rfind(b"PK\6\7")
if locator_offset != -1:
offset, = struct.unpack("<Q", end[locator_offset + 8: locator_offset + 16])
eocd64_data = m[offset: offset + 56]
eocd64 = struct.unpack("<4sQHHII4Q", eocd64_data)
(
eocd64_signature,
eocd_size,
version,
min_version,
num_disks,
num_disks2,
num_files,
num_files2,
directory_size,
directory_offset,
) = eocd64
assert eocd64_signature == b"PK\6\6"
# Read central directory headers which hold information about stored files
# as long as the signature matches
files: Dict[str, ZipInfo] = {}
mmap_offset = directory_offset
for _ in range(num_files):
header = m[mmap_offset : mmap_offset + 46]
mmap_offset += 46
(
signature,
version,
min_version,
unused0,
compression,
time,
date,
crc32,
compressed_size,
uncompressed_size,
filename_length,
extra_length,
comment_length,
unused1,
attributes0,
attributes1,
offset,
) = struct.unpack("<4s6H3I5HII", header)
assert signature == b"PK\1\2"
filename_bytes = m[mmap_offset : mmap_offset + filename_length]
mmap_offset += filename_length
extra = m[mmap_offset : mmap_offset + extra_length]
mmap_offset += extra_length + comment_length
for encoding in ["utf-8", "windows-1252", "CP437"]:
try:
filename = filename_bytes.rstrip(b"\0").decode(encoding)
break
except UnicodeDecodeError:
pass
else:
raise ValueError(f"Could not decode filename " + str(filename_bytes))
# TODO Figure out what those bytes mean.
# TODO Parse extra header correctly
extra = extra[4:]
if uncompressed_size == 0xFFFFFFFF:
assert len(extra) >= 8
uncompressed_size = struct.unpack("<Q", extra[:8])[0]
extra = extra[8:]
if compressed_size == 0xFFFFFFFF:
assert len(extra) >= 8
compressed_size = struct.unpack("<Q", extra[:8])[0]
extra = extra[8:]
if offset == 0xFFFFFFFF:
assert len(extra) >= 8
offset = struct.unpack("<Q", extra[-8:])[0]
extra = extra[8:]
info = ZipInfo(
filename,
offset,
crc32,
compressed_size,
uncompressed_size,
)
files[filename] = info
return files
def read_files(filename: str) -> Dict[str, ZipInfo]:
"""Read ZipInfo from zip file given its file path."""
with open(filename, "rb") as f:
with mmap.mmap(f.fileno(), length=0, access=mmap.ACCESS_READ) as m:
return _read_eocd_mmap(m)
class ParallelZipFile:
"""Like Python's zipfile.ZipFile, but uses mmap instead of a file object for
reading."""
filename: str
files: Dict[str, ZipInfo]
f: IO[bytes]
m: mmap.mmap
def __init__(
self, file: str, mode: str = "r", files: Optional[Dict[str, ZipInfo]] = None
) -> None:
assert mode == "r"
f = open(file, "rb")
m = mmap.mmap(f.fileno(), length=0, access=mmap.ACCESS_READ)
if files is None:
files = _read_eocd_mmap(m)
self.filename: str = file
self.files = files
self.f = f
self.m = m
def __contains__(self, filename: str) -> bool:
return filename in self.files
def __enter__(self) -> "ParallelZipFile":
return self
@no_type_check
def __exit__(self, exception_type, exception_value, exception_traceback) -> None:
self.close()
def close(self) -> None:
"""Close internal file and mmap objects. Will be called automatically
when using context manager, i.e. "with" statement."""
if not isinstance(self.m, bytes):
self.m.close()
self.f.close()
def namelist(self) -> List[str]:
"""Get file names for each file stored in zip file."""
return list(self.files.keys())
def infolist(self) -> List[ZipInfo]:
"""Get list of ZipInfo objects for each file stored in zip file."""
return list(self.files.values())
def read(self, filename: str) -> bytes:
"""Get bytes for file stored in zip file given its filename."""
files = self.files
if filename not in files:
raise ValueError(f"{filename} does not exist")
fileinfo = files[filename]
offset = fileinfo.header_offset
m = self.m
header = m[offset : offset + 30]
if len(header) < 30:
error_message = f"Header for {filename} too small ({len(header)} bytes, but must be at least 30 bytes)"
raise ValueError(error_message)
(
signature,
version,
unused,
compression,
time,
date,
crc32,
compressed_size,
uncompressed_size,
filename_length,
extra_length,
) = struct.unpack("<IHHHHHIIIHH", header)
# TODO Is this legal/the indented way to do this?
if compressed_size != fileinfo.compress_size:
compressed_size = fileinfo.compress_size
offset += 30 + filename_length + extra_length
compressed = m[offset : offset + compressed_size]
assert signature == 0x4034B50
if compression == 0:
# No compression
return compressed
elif compression == 8:
# DEFLATE compression
decompress = zlib.decompressobj(-zlib.MAX_WBITS)
return decompress.decompress(compressed)
elif compression == 12:
return bz2.decompress(compressed)
elif compression == 14:
# LZMA compression
_, size = struct.unpack("<HH", compressed[:4])
assert len(compressed) >= 4 + size
filt = lzma._decode_filter_properties(lzma.FILTER_LZMA1, compressed[4:4 + size])
decompress = lzma.LZMADecompressor(lzma.FORMAT_RAW, filters=[filt])
decompressed = decompress.decompress(compressed[4 + size:])
# Decompresses too much data sometimes
decompressed = decompressed[:uncompressed_size]
return decompressed
else:
error_message = f"Compression method {compression} not implemented"
raise NotImplementedError(error_message)