Hashlib produces different results on Ubuntu and Windows via GitHub actions

Question

I have created a test, to ensure that the content of the license file does not change. The test looks like the following:

    license_file = context.root_dir.joinpath("LICENSE")

    with open(license_file, "rb") as license_fd:
        content = license_fd.read()
    license_hash = hashlib.sha224(content).hexdigest()

    # Compare to "sha224sum LICENSE" of initial commit
    # TODO: Figure out why Windows gives another hash
    # assert license_hash == "45fd0b382919a02f391b9ce13e70ed703b9569cce812332d03c514a2"

In GitHub I'm running the test in a matrix with the os set to "ubuntu-latest", "macos-latest", and "windows-latest". Version of Python is set to 3.8.

For both Ubuntu and MacOS the test is passing. But the assertion fails on Windows.

I have read that open replaces line endings on Windows, if the mode is set to r . But opening the file with rb should not alter the file. I have also tried to replace the mode with r+b . The tests pasases on both Unix systems, but not Windows.

Are hashlib working different on Windows? It shouldn't. Are the pipeline altering files? It shouldn't.

I don't have a Windows machine to validate which part is altering the file or if hashlib have a bug in the Windows implementation. Most likely the first part.

Am I missing the obvious?

Answer 1

Are the pipeline altering files? It shouldn't.

Following Iłya Bursov's advice showed that the checkout of source code changed file endings as default on Windows.

It's a known hurdle (but not a bug): https://github.com/actions/checkout/issues/135#issuecomment-602171132

Fixed using:

    steps:
      - name: Set git to use LF
        run: |
          git config --global core.autocrlf false

      - uses: actions/checkout@v2

However I didn't force line endings to be lf as posted in the answer on GitHub.