Type inference and hasattr

[JukkaL]

I just encountered code like this:

if hasattr(x, 'initialize'):
     x.initialize()

The type of x was an ABC but it doesn't include initialize. This can be easily worked around by rewriting it like this:

if hasattr(x, 'initialize'):
     cast(Any, x).initialize()

However, mypy could do better, plausibly. For example:

• If x has a union type, infer only union item types with attribute initialize after the hasattr check. So if type of x is Union[str, X] and X has initialize, infer type of x to be X in the if body.
• Allow specifying "potentially undefined" attributes in types. Accessing these requires a hasattr check (a little like Optional[...] requiring something like an is not None check). Not sure what the syntax for this would be like. It would be nice to support these in ABCs as well.

[timabbott]

I ran into an instance of this in the Zulip test runner code, where we add an attribute to a test method via a decorator to track the fact it's expected to be slow, and then check it like this:

def enforce_timely_test_completion(test_method, test_name, delay):                                                                                                     
    # type: (Callable[[], None], str, float) -> None                                                                                                                   
    if hasattr(test_method, 'expected_run_time'):                                                                                                                      
        # Allow for tests to run 50% slower than normal due                                                                                                            
        # to random variations.                                                                                                                                        
        max_delay = 1.5 * test_method.expected_run_time

See https://github.com/zulip/zulip/blob/master/zerver/lib/test_runner.py for the full code.

[gvanrossum]

In general i despise hasattr() checks; it's usually much better to add a
class level initialization to None and check for that.

On Saturday, June 4, 2016, Tim Abbott [email protected] wrote:

I ran into an instance of this in the Zulip test runner code, where we add
an attribute to a test method via a decorator to track the fact it's
expected to be slow, and then check it like this:

def enforce_timely_test_completion(test_method, test_name, delay):
# type: (Callable[[], None], str, float) -> None
if hasattr(test_method, 'expected_run_time'):
# Allow for tests to run 50% slower than normal due
# to random variations.
max_delay = 1.5 * test_method.expected_run_time

See https://github.com/zulip/zulip/blob/master/zerver/lib/test_runner.py
for the full code.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1424 (comment), or mute
the thread
https://github.com/notifications/unsubscribe/ACwrMgWW3Km5ug5KWyGjSWJtCKdAsPpuks5qIme4gaJpZM4INlY-
.

--Guido (mobile)

[timabbott]

Just to be clear, these are attributes monkey-added to a function, not a class, so I don't think that solution would work here.

[gvanrossum]

So there's no place to declare their type anyway. Maybe using three-arg getattr is better?

[jhance]

If x has a union type, infer only union item types with attribute initialize after the hasattr check. So if type of x is Union[str, X] and X has initialize, infer type of x to be X in the if body.

Likely has false positives if one of the elements on the union is an ABC

[ikelos]

I was wondering whether the protocol/structural subtyping work could be used on a hasattr call to in some way to coerce the type into one that accepts the attribute within that scope?

[isidentical]

This also affects using features from future python versions (I'm aware that we can do sys.version_info compare but IMHO this case shouldn't give errors)

import ast
if hasattr(ast, "unparse"):
    print(ast.unparse)
else:
    print("no")

[TornaxO7]

This might be interesting for some people who encounters this problem. You could use the getattr function to get the value and set the default value to None (or any other value to check) and add an if condition afterwards to do some other action if this attribute doesn't exist. Here's a little example:

unknown = getattr(<object>, 'valname', None)
if unknown is None:
    # your action if it's None

[jenstroeger]

I came across this issue when using multiple inheritance in the context of SQLAlchemy’s Mixins. Take the Mixing in Columns example code which adds a property created_at, and which mypy then fails to identify as an attribute.

For both,

if hasattr(self, "create_at"):
    create_at = self.create_at

and

try:
    create_at = self.create_at
except AttributeError:
    # Do something else.

mypy reports the "Foo" has no attribute "bar" error, although both are safe at runtime. @TornaxO7’s suggestion above to use getattr() works well for any attribute value but None.

For the time being I’ve chosen to # type: ignore the guarded assignment above, but I wonder if a plugin for this particular case would make sense? (I’ve not written a plugin for mypy before 🤓)

[antonagestam]

@jenstroeger Hmm, it sounds like something off, why is the definition of create_at not inherited from the mixin?

[alexmojaki]

Based on when the fix above was merged, it should be in mypy 0.982, right? It still doesn't seem to be working, e.g. this:

class C:
    pass

c = C()

setattr(c, "foo", 42)
if hasattr(c, "foo"):
    print(c.foo)

still leads to error: "C" has no attribute "foo"

[JelleZijlstra]

It's not in 0.982. In the b29051c page you can see the tags that include the commit, and there aren't any. (Example commit that was in 0.981/2: 0ec789d)

[BlackHC]

Any updates on this? When will the fix make it into mypy? Thanks so much!

[jenshnielsen]

@BlackHC if you follow @JelleZijlstra's instructions above you can see that it should be in mypy 0.991

[BlackHC]

Thanks! Upgraded my package dependencies and it works perfectly now!

(You made me realize that poetry update was not doing what I thought it was doing and that I had to install the poetry up plugin. Apologies for my confusion around this!)