jwt.encode produces not constant JWT header in Python 3

After updating to 1.0.0 I found some inconsistency: When running the following test:

def test_encode_jwt():
    import jwt
    from sys import version_info
    claims_set = {
        'sub': 'user'
    }
    key = 'secret'
    token = jwt.encode(claims_set, key)
    if version_info >= (3, 0, 0):
        token = token.decode(encoding='UTF-8')

    assert token == 'eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ1c2VyIn0.' \
                    '8jVjALlPRYpE03sMD8kuqG9D4RSih5NjiISNZ-wO3oY'

In Python 2.7 this test passes always. In Python 3.4 this test sometimes passes and sometimes not. The reason is, that in Python 3 jwt.encode sometimes produce a header like:

{
  "alg": "HS256",
  "typ": "JWT"
}

This is the same like in Python 2.

But sometimes it produces a header like:

{
  "typ": "JWT",
  "alg": "HS256"
}

I didn’t find out why this happens.

Regards, Henri

[fergyfresh]

We have a 3rd party integration that hardcoded the order of the keys 😦. I’m glad I was validated here though by you guys confirming that the order doesn’t matter and isn’t specified in the RFC either.

[henri-hulski]

I understand. So I will close this issue.