CDX Fork of warc-tools

This is a fork of Hanzo Archives' warc tools package. It was forked after changeset 1897e2bc9d29 and has not yet been merged upstream.

This fork contains changes to support creating CDX indexes of warc files used by the Internet Archive's Wayback Machine. The CDX indexing script is here.


  • ArchiveRecord base class now contains a compressed_record_size field: changesets 941b79f83d6c and f0a98baef67b
  • The WARC-Payload-Digest is fabricated if not present. This allows us to include the sha1 hash in cdx files without having to store the entire payload in memory. Changeset 0bedbca70d7c and 73242eb3913d
  • Do not store payloads larger than 5MB due to memory issues: changesets f412b98c5575, d3767f736601, and 00f9a2879469
  • Speed up parsing of large payloads: changeset a84ad1ec7377
  • Bugfix for parsing warc headers with a CRLF at 1K compressed chunk boundary: changesets 25595bef06b0 and af549e5e3586
  • Handle Alexa arc files with missing content-type header or negative content length: changesets 683b833935b9 and a070fd062ce1

dependencies setuptools unittest2 python 2.6

hanzo warc tools:

    returns 0 if the arguments are all valid arc/warc files
    non zero on error

warcdump.py - writes human readable summary of warcfiles:
    usage: python warcdump.py foo.warc foo.warc.gz
    autodetects input format when filenames are passed
    i.e recordgzip vs plaintext, warc vs arc

    assumes uncompressed warc on stdin if no args

    python warcfilter.py pattern file file file
        searches all headers for regex pattern
    use -i to invert search
    use -U to constrain to url
    use -T to constrain to record type
    use -C to constrain to content-type

    autodetects and stdin like warcdump

    prints out a warc format by default.

    python warc2warc <input files>

    autodetects compression on file
    args, assumes uncompressed stdin if none

    use -Z to write compressed output

    i.e warc2warc -Z input > input.gz

    should ignore buggy records in input

    creates a crappy warc file from arc files on input
    a handful of headers are preserved
    use -Z to write compressed output
    i.e arc2warc -Z input.arc > input.warc.gz

    spits out an index like this:
#WARC filename offset warc-type warc-subject-uri warc-record-id content-type content-length
warccrap/mywarc.warc 1196018 request /images/slides/hanzo_markm__wwwoh.pdf <urn:uuid:fd1255a8-d07c-11df-b125-12313b0a18c6> application/http;msgtype=request 193
warccrap/mywarc.warc 1196631 response http://www.hanzoarchives.com/images/slides/hanzo_markm__wwwoh.pdf <urn:uuid:fd2614f8-d07c-11df-b125-12313b0a18c6> application/http;msgtype=response 3279474
    not great, but a start


arc2warc uses the conversion rules from the earlier arc2warc.c
as a starter for converting the headers

I haven't profiled the code yet (and don't plan to until it falls over)

warcvalid barely skirts some of the iso standard:
    missing things: strict whitespace, required headers check...
mime quoted printable header encoding
treating headers as utf8

things left to do (in no order): lots more testing. supporting pre 1.0 warc files add more documentation support more commandline options for output and filenames s3 urls

-- tef thomas.figg@hanzoarchives.com