diffoscope

Compare two files

$ diffoscope [path/to/file1] [path/to/file2]

$ diffoscope --no-progress [path/to/file1] [path/to/file2]

$ diffoscope --html [path/to/outfile|-] [path/to/file1] [path/to/file2]

$ diffoscope --exclude [pattern] [path/to/directory1] [path/to/directory2]

$ diffoscope --exclude-directory-metadata [auto|yes|no|recursive] [path/to/directory1] [path/to/directory2]

diffoscope [OPTION]... file1 file2

-h, --help
    show this help message and exit

--version
    show program's version number and exit

--list-tools
    list external tools used by diffoscope

--html[=FILE]
    write HTML report to FILE (default: diffoscope.html)

--html-title=TITLE
    set HTML title

--css=URL
    set CSS URL in HTML report

--new-file=NAME
    use NAME as label for first file

--old-file=NAME
    use NAME as label for second file

--ignore=METADATA
    ignore specific metadata differences (e.g. timestamp,overlay)

--max-diff-block-lines=LINES
    truncate unified diffs longer than LINES

--max-report-size=SIZE
    truncate report if larger than SIZE bytes

--no-report-size-limit
    do not limit report size

--unicode
    use Unicode line-drawing characters

--verbosity=LEVEL
    verbosity level (0-3)

--profile=PROFILE
    enable profiling (wall-clock or cpu-clock)

diffoscope is a powerful tool for comparing files, directories, and archives byte-by-byte and in human-readable form. Unlike diff(1), it dissects structured files like ELF binaries, tarballs, PDFs, images, and more using external helpers (e.g., binutils, pdftohtml, squashfs-tools). It generates detailed reports showing differences in metadata, contents, and subcomponents.

Ideal for reproducible builds verification in Debian and other distros, debugging container images, or spotting subtle changes in firmware. Output formats include text, HTML, JSON, and side-by-side diffs. It handles symlinks, devices, FIFOs via emulation and supports exclusions for noise reduction.

Requires many external tools for full functionality; install via package manager. Resource-intensive on large files/directories. Some formats need root for devices/FIFOs.

Available in most distros: apt install diffoscope, dnf install diffoscope. Or pip install diffoscope.

Run diffoscope --list-tools to see required tools; auto-detects most.

Developed since 2014 by Chris Lamb for Debian Reproducible Builds effort. Now maintained by Reproducible Builds project, supports 100+ formats, Python-based with active upstream.

diff(1), cmp(1), vbindiff(1), xxd(1), binwalk(1)