python
(🐞) crash with `--pretty` from attrs raising an error in the wrong file
> git clone https://github.com/DataDog/dd-trace-py
> cd dd-trace-py
> pip install -e .
> pip install mypy
> mypy --strict --check-untyped-defs | grep threading.py
ddtrace/profiling/collector/threading.py:25: error: Need type annotation for "recorder" [var-annotated]
ddtrace/profiling/collector/threading.py:29: error: Function is missing a type annotation for one or more arguments [no-untyped-def]
ddtrace/profiling/collector/threading.py:316: error: Need type annotation for "tracer" [var-annotated]
> mypy --strict --check-untyped-defs --pretty
...
File "mypy/main.py", line 100, in main
File "mypy/main.py", line 182, in run_build
File "mypy/build.py", line 192, in build
File "mypy/build.py", line 266, in _build
File "mypy/build.py", line 2942, in dispatch
File "mypy/build.py", line 3340, in process_graph
File "mypy/build.py", line 3465, in process_stale_scc
File "mypy/errors.py", line 915, in file_messages
File "mypy/errors.py", line 885, in format_messages
IndexError: list index out of range
threading.py:316: error, threading only has 43 lines in it.
I am seeing what might be a related issue. For me it doesn't result in a crash, but I am getting errors for rows which aren't in my file. I'm getting a bunch of errors like:
my_file.py:119: error: X | Y syntax for unions requires Python 3.10 [syntax]
my_file.py:120: error: X | Y syntax for unions requires Python 3.10 [syntax]
my_file.py:121: error: X | Y syntax for unions requires Python 3.10 [syntax]
my_file.py:122: error: X | Y syntax for unions requires Python 3.10 [syntax]
my_file.py:123: error: X | Y syntax for unions requires Python 3.10 [syntax]
...
My file my_file.py has only 30 lines.
It happens on three different files in my codebase, and what they all have in common is that they all have the import from pydantic_settings import BaseSettings.
I'm using python 3.8, and in the file in pydantic-settings where BaseSettings is defined, lines 129-136 are the parameter list for BaseSettings.__init__(), and they are all type-annnotated as something | None.
It looks like somehow mypy thinks those lines are in my source file?
Another strange detail: on the second run (if I already have a .mypy_cache/) it doesn't fail. But if I delete the cache and run again the errors come back.
I can reproduce it like this:
from pydantic_settings import BaseSettings
class Foo(BaseSettings):
pass
$ mypy test.py
test.py:119: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:120: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:121: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:122: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:123: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:124: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:125: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:126: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:127: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:128: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:129: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:130: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:131: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:132: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:133: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:134: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:135: error: X | Y syntax for unions requires Python 3.10 [syntax]
test.py:136: error: X | Y syntax for unions requires Python 3.10 [syntax]
Found 18 errors in 1 file (checked 1 source file)
environment: mypy 1.11.0 python 3.8.19 pydantic-settings 2.3.4 pydantic 2.8.2 pydantic mypy plugin enabled
It doesn't replicate with the mypy plugin disabled, but it doesn't replicate with mypy 1.10.1 either.
I have the same problem. Python 3.9.19, mypy 1.11.0
A file that imports from pydantic_settings gets a bunch of errors related to 3.10 union syntax. That syntax is used in pydantic-settings, but it is not used in my file. Also, the line numbers stated in the mypy output are greater than the number of lines in my file. It somehow seems to think some pydantic_settings file is my file.
If the mypy cache exists, the error does not show up. If I delete the cache and re-run, the errors are back.
Calling mypy with --pretty changes the listed errors to a crash (otherwise it behaves the same with regards to crashing).
Traceback (most recent call last):
File "REDACTED/venv/bin/mypy", line 8, in <module>
sys.exit(console_entry())
File "REDACTED/venv/lib/python3.9/site-packages/mypy/__main__.py", line 15, in console_entry
main()
File "mypy/main.py", line 103, in main
File "mypy/main.py", line 187, in run_build
File "mypy/build.py", line 193, in build
File "mypy/build.py", line 268, in _build
File "mypy/build.py", line 2950, in dispatch
File "mypy/build.py", line 3348, in process_graph
File "mypy/build.py", line 3471, in process_stale_scc
File "mypy/errors.py", line 923, in file_messages
File "mypy/errors.py", line 883, in format_messages
IndexError: list index out of range
Will this problem be tracked in this issue? (It was not clear to me from the title that this was the same problem I had.)
@brianmedigate I am witnessing the behavior you described in files with BaseSettings imported.
@KotlinIsland does your use case include pydantic? I wonder if we should cross-post this to the pydantic issues page?
Hitting the same issue, and I'm using Pydantic's BaseSettings where I'm hitting this.
Flagged it to pydantic-settings here: https://github.com/pydantic/pydantic-settings/issues/349
Pydantic dev here — I'll note that I can reproduce the error with Python 3.9.19 and Mypy 1.11.0, but the issue goes away if I use python >=3.10 or mypy <=1.10.1.
I do think it has something to do with the fact that pydantic-settings uses the X | Y union syntax and from __future__ import annotations, given the presumably spurious errors being produced without the --pretty flag that reference this syntax, and the fact that upgrading past 3.9 fixes it.
Perhaps it is an issue with detecting the from __future__ import annotations? (Technically we are using from __future__ import annotations as _annotations at the top of the relevant file in pydantic-settings.)
@KotlinIsland does your use case include pydantic? I wonder if we should cross-post this to the pydantic issues page?
I found it from the primer output in a basedmypy pull request, the primer is much more comprehensive in basedmypy, so it hasn't been detected in mypy
OK, I looked at it and in both cases it is a bug in the plugins, not sure why they are surfaced only now. For attrs plugin I can just fix it, because it is hosted in mypy. For pydantic I am not sure what to do, the problem is that all those errors are coming from (aparrently generated) Foo.__init__ (in the example above). It looks like the plugin simply copies types from the __init__ of supertype including all metadata, like line/column number and whether it was created using new syntax, etc. I can try disabling this error for plugin generated methods, but no guarantees, ultimately plugin authors may need to fix the plugin.
Looking at commit history, the pydantic issue was likely triggered by https://github.com/python/mypy/pull/17371 cc @hauntsaninja The default copy behavior for union is now to carry the uses_pep604_syntax flag, which is actually somewhat questionable.
@ilevkivskyi if you end up needing to modify the attrs plugin, I am happy to take responsibility (or at least to try to take responsibility 😅) for adapting the same fix to the pydantic plugin. I would just appreciate if you could share a link to any such PR on this issue (or vice versa).
To be clear, the pydantic plugin was adapted from the dataclasses plugin some time ago, and every so often I try to update it to reflect updates/improvements in the dataclasses plugin; if the dataclasses and/or attrs plugins now copy the "correct" subset of metadata data (instead of all supertype method metadata), I can try to modify the pydantic plugin to handle it correctly, it would just be helpful to have some kind of reference to a spot where this is being done "properly" from your perspective.
I made a change in basedmypy to prevent the crash if the requested source code doesn't exist
@ilevkivskyi @dmontagu any update on this? Is there an issue for this on the pydantic issue tracker?
After some discussions, we decided that the real underlying problem is using types as error context, and this needs to be prohibited altogether. This is however a relatively large refactoring, so will take some time. I may have time to work on this (or on the quick fix I mentioned above) but no guarantees.