How to make a custom comparison in pytest style?

[heckad]

Hi everybody.

I have two python builtin object instance without __eq__ method and I want to write a custom assert equal for this case. In the documentation, I found pytest_assertrepr_compare hook which does what I want but they call only when test return False. Does pytest have a hook which called instead of build-in __eq__ method? For example something like

foo.py

# It only has the build in __cmp__ method!
class Foo:
    def __init__(self, val):
        self.val = val

conftest.py

# If exists hoock
def pytest_asserttest(op, left, right):
   if isinstance(left, Foo) and isinstance(right, Foo) and op == "==":
      # do some cmp work
   # If return None then do build in __cmp__

test.py

def test_something()
    f1 = Foo(1)
    f2 = Foo(2)
    assert f1 == f2  # should be cal pytest_asserttest

[symonk]

I am confused by your use case, it seems overly complex. Would you not consider implementing dunder eq (and Dunder hash) etc in your object explicitly and then use simple asserting to check the fact, passing the logic to the hook seems weird to me, not to mention non reusable

[RonnyPfannschmidt]

Because magic behaviour hacks make debugging hell

[nicoddemus]

For the sake of discussion, currently pytest rewrites:

assert <expr>

Into:

if not <expr>:
    # magic to generate a nice message, by breaking <expr> into its components
    raise AssertionError(msg)

Which is a pretty harmless translation, because we don't really change any semantics.

To support the proposal here, it would need to break the expression even further in order to call hooks and decide to fail the assertion. At first glance seems complex and a potential source for bugs, but I might be wrong.

[heckad]

Pytest has the pytest_assertrepr_compare wich do that I want but they call after eval <expr> and used only to show a reason for a failed test.

[RonnyPfannschmidt]

I believe it's fine to introduce a assertrepr_valur hook for only the result valve

It's not fin to reuse the compare hook

[nicoddemus]

Pytest has the pytest_assertrepr_compare wich do that I want but they call after eval and used only to show a reason for a failed test.

Indeed, but my point was that we don't really try to do anything until the expression is False, at which point we would get an assertion error anyway. AFAIU to include a new hook we would need to break the expression first, because we want to overwrite the original value of <expr>.

I don't like this idea, because this seemingly innocent change might yield different results now:

assert <expr>

To:

if not <expr>:
    # do something else
    assert False

With the hook in place, both examples will behave very differently: in the first example <expr> will be True (it will call the hook), but will not call the hook in the second example and will evaluate to False.

Seems like we are trying to abuse the assertion rewrite to inject new logic into things; injecting new logic via mocking is not be a problem in general for "normal" Python objects, but is indeed a problem for C extension code, but I don't think we should use the assertion rewriting mechanism to circumvent that.

So far I'm 👎 on the idea.