Use well-defined enum representation

[tiran]

Python 3.10 changed enum's object and string representation. PyCA
cryptography now uses a custom subclass of enum.Enum() will well-defined
__repr__ and __str__ from Python 3.9.

Related: https://bugs.python.org/issue40066
Fixes: #5995
Signed-off-by: Christian Heimes [email protected]

[alex]

The motivation here is that __repr__ changed in 3.10 and we have tests on that, right?

@reaperhulk opinions on doing this vs. just putting a branch in the tests?

[alex]

This doesn't need an openssl type, it can just use the default one.

[tiran]

So just {VERSION: "3.10-dev", TOXENV: "py310"}?

[alex]

Exactly

…

On Mon, May 10, 2021 at 8:32 AM Christian Heimes ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In .github/workflows/ci.yml <#6042 (comment)>: > @@ -30,6 +30,7 @@ jobs: - {VERSION: "3.9", TOXENV: "py39", OPENSSL: {TYPE: "libressl", VERSION: "3.1.5"}} - {VERSION: "3.9", TOXENV: "py39", OPENSSL: {TYPE: "libressl", VERSION: "3.2.5"}} - {VERSION: "3.9", TOXENV: "py39", OPENSSL: {TYPE: "libressl", VERSION: "3.3.2"}} + - {VERSION: "3.10-dev", TOXENV: "py310", OPENSSL: {TYPE: "openssl", VERSION: "1.1.1k"}} So just {VERSION: "3.10-dev", TOXENV: "py310"}? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6042 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAAGBHXZGHPHPRPOJQVAD3TM7G7JANCNFSM44Q644AQ> .

-- All that is necessary for evil to succeed is for good people to do nothing.

[tiran]

Correct, the object and string representation were changed during the alphas. IIRC it changed again between last alpha and first beta. PyCA's tests contain checks for representation of some objects, for example:

___________________________ TestTLSFeature.test_repr ___________________________

self = <tests.x509.test_x509_ext.TestTLSFeature object at 0x7faf75790d60>

    def test_repr(self):
        ext1 = x509.TLSFeature([x509.TLSFeatureType.status_request])
>       assert repr(ext1) == (
            "<TLSFeature(features=[<TLSFeatureType.status_request: 5>])>"
        )
E       AssertionError: assert '<TLSFeature(...us_request])>' == '<TLSFeature(...equest: 5>])>'
E         - <TLSFeature(features=[<TLSFeatureType.status_request: 5>])>
E         ?                       -                             ----
E         + <TLSFeature(features=[TLSFeatureType.status_request])>

tests/x509/test_x509_ext.py:150: AssertionError

Miro asked me to look into the matter to unblock Python 3.10 update for next Fedora release. https://bugzilla.redhat.com/show_bug.cgi?id=1952522

[reaperhulk]

I think I'm in favor of implementing __repr__ since we can't guarantee stability without that. Alternately, we could stop testing repr, but since we want it to be readable we should commit to a stable one.