Algorithm::GDiffDelta - generate and apply GDIFF format binary deltas


    use Algorithm::GDiffDelta qw(
        gdiff_adler32 gdiff_delta gdiff_apply

    # Pass in two file handles for reading from and one to
    # writing the GDIFF binary delta to:
    gdiff_delta($orig, $changed, $delta);

    # Pass in file handles of original file and GDIFF delta
    # to read from, and file to write reconstructed changed
    # file to:
    gdiff_apply($orig, $delta, $changed);

    # A fast adler32 digest implementation is also available:
    my $adler32 = gdiff_adler32(1, 'some data');
    $adler32 = gdiff_adler32($adler32, 'some more data');


This module can be used to generate binary deltas describing the differences between two files. Given the first file and the delta the second file can be reconstructed.

A delta is equivalent to the output of the unix diff program, except that it can efficiently represent the differences between similar binary files, containing any sequences of bytes. These deltas can be used for updating files over a network (as rsync does) or for efficiently storing a revision history of changes to a file (as Subversion does).

There are several formats used for binary deltas. The one supported by this module is the GDIFF format, which is fairly simple and is documented as a W3C note (See the SEE ALSO section below).

This module generates and processes deltas using file handles. It supports both native Perl file handles (created with the built-in open format) and objects that support the right methods. For an object to work it must support at least the read, seek, and tell methods (if it is an input file) or the write method (if it is an output file). This allows strings to be used for input and output, by wrapping them in an IO::Scalar object or similar. A future version of this module might support reading and writing directly through references to scalars, because that should be much more efficient.

See the section ALGORITHM AND DELTA FORMAT below for some notes on the algorithm used by this module and how the GDIFF delta format works.


No functions are exported by default. Pass the function names to Exporter in the use line, or use the :all tag to import them all.

gdiff_adler32($initial_value, $string_data)

Generate an Adler32 digest of the bytes in $string_data, starting with a hash value of $initial_value. This function is provided only because it is used internally and so it might as well be made available. It isn't needed for generating or applying binary delta files.

The $initial_value should usually be 1. The result of calling this function (which is a 32-bit unsigned integer value) can be passed back in as a new initial value to checksum some more data. This allows a large file to be checksummed in separate chunks.

Another implementation of Adler-32 is provided in the Digest::Adler32 module.

The Adler-32 checksum algorithm is defined in RFC 1950, section 8.2. Sample code in C is also provided there.

gdiff_apply($file1, $file2, $delta_file)

Takes three file handles. The first two are read from, and it must be possible to seek in them. The third is written to.

This generates a binary delta describing the changes from $file1 to $file2. The delta will allow $file2 to be reconstructed from $file1 later.

No value is returned. Errors will cause this function to croak with a suitable error message.

gdiff_delta($file1, $delta_file, $file2)

Takes three file handles. The first two are read from, and it must be possible to seek in them. The third is written to.

The delta is used to reconstruct $file2 from $file1. The delta must be a valid GDIFF file.

No value is returned. Errors will cause this function to croak with a suitable error message.


The algorithm and some of the code used in this module was derived from the xdiff library. It has been adjusted to write GDIFF deltas, rather than the custom format used by xdiff (which was funtionally equivalent but a little simpler than GDIFF).

Notes about how the algorithm works should be added here once I figure it out myself.

A GDIFF file consists of a five byte header, followed by a sequence of commands. Each command has a one byte 'opcode' followed by some arguments. The delta file is terminated by opcode zero.

There are two types of commands. DATA commands have a chunk of literal data as one of their arguments, which is inserted into the output. COPY commands cause a chunk of data from the input file to be inserted into the output.

A GDIFF delta is one-way: if it generates $file2 from $file1 then it cannot be used to generate $file1 from $file2.


The xdiff library, upon which this module was based:

GDIFF specification (as a W3C note):

RFC 1950, which defines the Adler-32 checksum algorithm:


Parts of this library were derived from the code for libxdiff, by Davide Libenzi <>

The rest is Copyright 2004, Geoff Richards <>

This library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 2.1 of the License, or (at your option) any later version. A copy of the license is available at: