little-things/warc-peek

#!/usr/bin/env python3

# `warc-peek.py` is a small script to help looking into gzipped WARC files without decompressing the entire file.
# It searches a window in the file for gzip's magic bytes `1F 8B`, attempts decompression, compares the result to the expected beginning of a WARC record, and prints all valid offsets.
# These can then be used with e.g. `tail` and `zless` to actually look at the records.
#
# Usage: warc-peek.py WARCFILE OFFSET LENGTH
#  Opens `WARCFILE`, reads `LENGTH` bytes starting at `OFFSET` (zero-based), and prints valid WARC record offsets to stdout (one integer per line).
#
# Caveats
# - This script only works with WARCs in which each record is compressed individually.
#   This is what the specification recommends and what most tools should generate by default, but there definitely exist valid compressed WARCs which can't be processed in this way.
# - When you want to use `tail -c+OFFSET WARCFILE | zless` to look at the records, keep in mind that `tail` uses one-based indices, i.e. you will have to add one to the indices returned by `warc-peek.py`.
# - `warc-peek.py` will miss valid record offsets in the last 512 bytes of the window.
#   This is because a certain length of the compressed data is necessary to be able to decompress it. `warc-peek.py` uses 512 bytes for this and will therefore
#   not attempt decompression when `1F 8B` is found in the last 512 bytes of the window. You can increase `LENGTH` to compensate for this if necessary.

import argparse
import logging
import zlib


logger = logging.getLogger('warc-peek')


def finditer(b, sub):
	pos = 0
	while True:
		pos = b.find(sub, pos)
		if pos < 0:
			break
		yield pos
		pos += 1


def find_offsets(warcfile, offset, length):
	with open(warcfile, 'rb') as fp:
		fp.seek(offset)
		buffer = fp.read(length)

	logger.debug('Buffer length: {:d}'.format(len(buffer)))
	for pos in finditer(buffer, b'\x1f\x8b'):
		logger.debug('Trying relative offset {:d}'.format(pos))
		if pos > len(buffer) - 512: # 512 bytes might be a bit too much, but at least it ensures that the decompression will work.
			break
		try:
			dec = zlib.decompressobj(zlib.MAX_WBITS | 32).decompress(buffer[pos:pos+512])
		except:
			continue
		logger.debug('First 100 bytes of decompressed data: {!r}'.format(dec[:100]))
		if dec.startswith(b'WARC/1.0\r\n') or dec.startswith(b'WARC/1.1\r\n'):
			yield offset + pos


if __name__ == '__main__':
	parser = argparse.ArgumentParser()
	parser.add_argument('--debug', action = 'store_true', help = 'Enable debug output')
	parser.add_argument('warcfile', help = 'A .warc.gz file')
	parser.add_argument('offset', type = int, help = 'Zero-based byte offset of the window')
	parser.add_argument('length', type = int, help = 'Length in bytes of the window')
	args = parser.parse_args()

	if args.debug:
		logging.basicConfig(
			format = '{asctime}  {levelname}  {name}  {message}',
			style = '{',
			level = logging.DEBUG,
		)
	for offset in find_offsets(args.warcfile, args.offset, args.length):
		print(offset)