Repository navigation
Daemon should respect the rule that stubs are normal even with follow-imports=skip #5566
Description
Currently this test fails, presumably because the daemon applies --follow-imports=skip to stub files as well (while it shouldn't):
[case testFineFollowImportSkipInvalidatedOnAddedStub]
# flags: --follow-imports=skip --ignore-missing-imports
# cmd: mypy main.py
[file main.py]
import other
x: str = other.x
[file other.pyi.2]
x = 1
[stale main, other]
[rechecked main, other]
[out]
==
main:2: error: Incompatible types in assignment (expression has type "int", variable has type "str")
This was found in #5556
@msullivan do you have any ideas why this may happen?
Activity
ilevkivskyi commentedon Sep 4, 2018
It looks like same happens with
--bazelincremental build.@gvanrossum could you give some hints about why this might happen?
gvanrossum commentedon Sep 5, 2018
(The thing we saw internally with Bazel has a completely different cause, unrelated to this issue.)
ilevkivskyi commentedon Sep 5, 2018
Yes, I just tried to look a bit more at this, and it looks like the root cause is simple actually: daemon never supposed to work with
follow_imports=normal, but apparently it should always do this for stubs.@JukkaL what do you think about this?
Also it looks like I had some misunderstanding about how fine-grained tests work, it seems to me now that by default at every test step all the modules currently present are passed to
Server.check(). This means I have previously added many tests like[case testFineAddedMissingStubs]that don't test what I wanted.JukkaL commentedon Sep 5, 2018
@ilevkivskyi Hmm this is pretty unfortunate. On the other hand, this has been around for a long time before this got reported, so it doesn't seem that an urgent fix is needed. We can look at this in a few weeks when the core team will have a bug fixing spree.
msullivan commentedon Nov 5, 2018
I've looked into this some and it is no easier to properly support
follow_imports=normalfor stubs alone than it is to support it in general. That makes this a sub-issue of the newly opened #5870.msullivan commentedon Jun 18, 2020
This might be easier to fix now with the new follow-imports=normal stuff?
JukkaL commentedon Jun 19, 2020
This might be easier now, though I can't say how easy exactly without spending some time on this.
JukkaL commentedon Mar 5, 2021
This works now.