Skip to main content

A tiny, zero-dependency replacement for Python's zipfile.ZipFile for creating reproducible/deterministic ZIP archives.

Project description


PyPI Conda Version conda-forge feedstock Supported Python versions tests codecov

A tiny, zero-dependency replacement for Python's zipfile.ZipFile library for creating reproducible/deterministic ZIP archives.

"Reproducible" or "deterministic" in this context means that the binary content of the ZIP archive is identical if you add files with identical binary content in the same order. It means you can reliably check equality of the contents of two ZIP archives by simply comparing checksums of the archive using a hash function like MD5 or SHA-256.

This Python package provides a ReproducibleZipFile class that works exactly like zipfile.ZipFile from the Python standard library, except that certain file metadata are set to fixed values. See "How does repro-zipfile work?" below for details.

You can also optionally install a command-line program, rpzip. See "rpzip command line program" below for more information.


repro-zipfile is available from PyPI. To install, run:

pip install repro-zipfile

It is also available from conda-forge. To install, run:

conda install repro-zipfile -c conda-forge


Simply import ReproducibleZipFile and use it in the same way you would use zipfile.ZipFile from the Python standard library.

from repro_zipfile import ReproducibleZipFile

with ReproducibleZipFile("", "w") as zp:
    # Use write to add a file to the archive
    zp.write("examples/data.txt", arcname="data.txt")
    # Or writestr to write data to the archive
    zp.writestr("lore.txt", data="goodbye")

Note that files must be written to the archive in the same order to reproduce an identical archive. Be aware that functions that like os.listdir, os.glob, Path.iterdir, and Path.glob return files in a nondeterministic order—you should call sorted on their returned values first.

See examples/ for an example script that you can run, and examples/ for a demonstration in contrast with the standard library's zipfile module.

For more advanced usage, such as customizing the fixed metadata values, see the subsections under "How does repro-zipfile work?".

rpzip command-line program


You can optionally install a lightweight command-line program, rpzip. This includes an additional dependency on the typer CLI framework. You can install it either directly or using the cli extra with repro-zipfile:

pip install rpzip
# or
pip install repro-zipfile[cli]

rpzip is designed to a partial drop-in replacement ubiquitous zip program. Use rpzip --help to see the documentation. Here are some usage examples:

# Archive a single file
rpzip examples/data.txt
# Archive multiple files
rpzip examples/data.txt
# Archive multiple files with a shell glob
rpzip examples/*.py
# Archive a directory recursively
rpzip -r examples

In addition to the fixed file metadata done by repro-zipfile, rpzip will also always sort all paths being written.

How does repro-zipfile work?

ZIP archives are not normally reproducible even when containing files with identical content because of file metadata. In particular, the usual culprits are:

  1. Last-modified timestamps
  2. File-system permissions (mode)

repro_zipfile.ReproducibleZipFile is a subclass of zipfile.ZipFile that overrides the write, writestr, and mkdir methods with versions that set the above metadata to fixed values. Note that repro-zipfile does not modify the original files—only the metadata written to the archive.

You can effectively reproduce what ReproducibleZipFile does with something like this:

from zipfile import ZipFile

with ZipFile("", "w") as zp:
    # Use write to add a file to the archive
    zp.write("examples/data.txt", arcname="data.txt")
    zinfo = zp.getinfo("data.txt")
    zinfo.date_time = (1980, 1, 1, 0, 0, 0)
    zinfo.external_attr = 0o644 << 16
    # Or writestr to write data to the archive
    zp.writestr("lore.txt", data="goodbye")
    zinfo = zp.getinfo("lore.txt")
    zinfo.date_time = (1980, 1, 1, 0, 0, 0)
    zinfo.external_attr = 0o644 << 16

It's not hard to do, but we believe ReproducibleZipFile is sufficiently more convenient to justify a small package!

See the next two sections for more details about the replacement metadata values and how to customize them.

Last-modified timestamps

ZIP archives store the last-modified timestamps of files and directories. ReproducibleZipFile will set this to a fixed value. By default, the fixed value is 1980-01-01 00:00 UTC, which is the earliest timestamp that is supported by the ZIP format specifications.

You can customize this value with the SOURCE_DATE_EPOCH environment variable. If set, it will be used as the fixed value instead. This should be an integer corresponding to the Unix epoch time of the timestamp you want to set, e.g., 1704067230 for 2024-01-01 00:00:00 UTC. SOURCE_DATE_EPOCH is a standard created by the Reproducible Builds project for software distributions.

File-system permissions

ZIP archives store the file-system permissions of files and directories. The default permissions set for new files or directories often can be different across different systems or users without any intentional choices being made. (These default permissions are controlled by something called umask.) ReproducibleZipFile will set these to fixed values. By default, the fixed values are 0o644 (rw-r--r--) for files and 0o755 (rwxr-xr-x) for directories, which matches the common default umask of 0o022 for root users on Unix systems. (The 0o prefix is how you can write an octal—i.e., base 8—integer literal in Python.)

You can customize these values using the environment variables REPRO_ZIPFILE_FILE_MODE and REPRO_ZIPFILE_DIR_MODE. They should be in three-digit octal Unix numeric notation, e.g., 644 for rw-r--r--.

Why care about reproducible ZIP archives?

ZIP archives are often useful when dealing with a set of multiple files, especially if the files are large and can be compressed. Creating reproducible ZIP archives is often useful for:

  • Building a software package. This is a development best practice to make it easier to verify distributed software packages. See the Reproducible Builds project for more explanation.
  • Working with data. Verify that your data pipeline produced the same outputs, and avoid further reprocessing of identical data.
  • Packaging machine learning model artifacts. Manage model artifact packages more effectively by knowing when they contain identical models.

Related Tools and Alternatives

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

repro_zipfile-0.3.1.tar.gz (20.7 kB view hashes)

Uploaded Source

Built Distribution

repro_zipfile-0.3.1-py3-none-any.whl (10.8 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page