Possible bug: Serialization to JSON is not deterministic

[apirogov]

I'm using codemetapy as a linter with pre-commit and noticed that for some reason it fails sometimes because the codemeta.json file changed.

In my case, this is caused by my classifiers:

classifiers = [
    "Operating System :: POSIX :: Linux",
    "Development Status :: 3 - Alpha",
    "Intended Audience :: Science/Research",
    "Intended Audience :: Developers",
]

Then codemetapy will randomly orders them, sometimes resulting in:

    "audience": [
        {
            "@type": "Audience",
            "audienceType": "Science/Research"
        },
        {
            "@type": "Audience",
            "audienceType": "Developers"
        }
    ],

and sometimes in:

    "audience": [
        {
            "@type": "Audience",
            "audienceType": "Developers"
        },
        {
            "@type": "Audience",
            "audienceType": "Science/Research"
        }
    ],

Would it be possible to make sure that a consistent order is ensured?

As a workaround I currently add a global exclude: '^codemeta.json$' to my hook, which is not nice because it prevents my JSON linter to format the JSON.

Expected behavior

On the same input files exactly the same output file should be generated.

[proycon]

Thanks for the report, I agree that the output should be deterministic. It already is in the ordering of fields in a map but not yet in the ordering of objects in a sequence, I'll implement a solution for it.