Attempt to reproduce #91 #92
Conversation
|
Try applying diff --git a/tests/test_samples.py b/tests/test_samples.py
index f853009..6111e7e 100644
--- a/tests/test_samples.py
+++ b/tests/test_samples.py
@@ -26,6 +26,7 @@ def test_samples(samplefile, mypy_cache_dir):
opts.show_traceback = True
opts.namespace_packages = True
opts.hide_error_codes = True
+ opts.warn_unreachable = True
opts.plugins = ['mypy_zope:plugin']
# Config file is needed to load plugins, it doesn't not exist and is not
# supposed to.To get the unreachable error |
|
Works for me locally: |
|
I spent a little time trying to track this down by sticking print statements into mypy. I'm a bit suspicious of the fact that mypy decides to return in this In this situation:
It seems a bit odd to conclude that |
|
Sorry for lack of attention to this, I was on my vacation for the last couple of weeks. Getting back to the usual rhythm now. And thanks for digging in. What looks alarming in that snippet is |
|
@DMRobertson I did a pretty crazy hack to work around this (see the diffs). I'm both proud of it and afraid of it at the same time :) I suspect it could cause some memory leaking in the long-running dmypy instances, but it is the best I could come up with so far. Would it be possible to check the fix on your codebase? |
|
Apologies for the delay in getting back to you. I'm trying this out on matrix-org/synapse#15313, as well as locally.
EDIT: I don't seem able to reproduce that outcome now... not sure what is happening! I haven't had the chance to dig into the remaining error yet; when I get time I'll see if I can reduce that to a minimal example for you. Before: After:
Afraid I don't have any information to report here: I don't use dmypy myself. I'm not sure if anyone else on the team does either (@clokep maybe?) |
I do not use it. Sorry. 😢 |
|
Erm, I have just reran the commands after which I saw the full suite of 11 errors. So I am not sure why I only saw one error in the sample above. Some kind of caching problem? (But I explicitly nuked the cache...) |
Hm, you mean changes in this branch doesn't make any difference for you? I wonder how can we ensure you are actually getting the right branch. Could you check out the branch in a separate directory and install it via |
It looks like it :( I had one mypy run where it seemed to make a difference, but I'm unable to reproduce that---even with an editable install of mypy-zope checked out on this branch. I reduced the test failures I saw in our project to a case. It fails for me on this branch of mypy-zope: do you see the same? diff --git a/tests/samples/dmr_unreachable.py b/tests/samples/dmr_unreachable.py
index e69de29..36f2b13 100644
--- a/tests/samples/dmr_unreachable.py
+++ b/tests/samples/dmr_unreachable.py
@@ -0,0 +1,31 @@
+from typing import Optional
+
+from zope.interface import Interface, implementer
+
+
+class IFoo(Interface):
+ pass
+
+
+@implementer(IFoo)
+class BaseFoo:
+ pass
+
+
+class ChildFoo(BaseFoo):
+ pass
+
+
+class IFooFactory(Interface):
+ def build() -> Optional[IFoo]:
+ pass
+
+
+def build_and_use_foo(client_factory: IFooFactory) -> None:
+ client_protocol = client_factory.build()
+ assert isinstance(client_protocol, ChildFoo)
+ print("Hello")
+
+
+"""
+"""
\ No newline at end of file |
|
@DMRobertson just for the sake of experiment, could you copy https://github.com/Shoobx/mypy-zope/blob/96518e2725401adb8ecbe177441971666037fd81/tests/samples/isinstance_impl.py into the working directory of your project and run mypy on that file? |
Certainly! This is matrix-org/synapse@ |
| class ChildFoo(BaseFoo): | ||
| pass |
There was a problem hiding this comment.
I think this is the "important" difference between this test case and isinstance_impl. On line 26 below we are asserting that something of type IFoo could be an instance of a subclass of a class implementing IFoo.
In isinstance_impl line 20, we do the same but without the additional subclass.
No description provided.