Flask error handler caching leads to inconsistent behavior

[ngaya-ll]

When Flask finds an error handler for a given exception, it caches it in the handler map. This can lead to incorrect behavior for subclasses of the exception.

Here is some code demonstrating the problem:

from flask import Flask
from werkzeug.exceptions import InternalServerError


class E1(Exception):
    pass


class E2(Exception):
    pass


class E(E1, E2):
    pass


app = Flask(__name__)


@app.errorhandler(E2)
def handle_e2(e):
    return "E2", 500


@app.errorhandler(Exception)
def handle_exception(e):
    return "Exception", 500


@app.route("/<exception>", methods=['POST'])
def raise_exception(exception):
    exc = globals()[exception]
    raise exc


def test_errorhandler_precedence():
    client = app.test_client()

    response1 = client.post('/E1')
    assert response1.data == "Exception"

    response2 = client.post('/E')
    assert response2.data == "E2"

In this example, we have two exception classes, E1 and E2, and a third exception E that inherits from both. We register error handlers for E2 and Exception. When E is raised, Flask should invoke the E2 error handler as E2 comes before Exception in the MRO of E. But if E1 has been raised in the past, then the Exception error handler will be "cached" as the handler for E1, so that handler will be invoked instead.

[davidism]

Related to #1433