Give Pyright what it wants (`alias` attributes everywhere)

[A5rocks]

I don't like pyright nor this behavior, but that doesn't mean users should have to care. We can add a test to prevent any sort of regression regarding this, forever.

cc @mikenerone cause this is from Gitter.

also @CoolCat467 I ran into issues with running the regenerate-files pre-commit hook and couldn't figure them out after about 15 minutes. It was complaining about attrs not being possible to import but unfortunately there's no way to get it to use the test-requirements.txt file. Have you run into this issue regarding it not picking up the virtual environment? (I'm using a seperate git client so that kind of makes sense, but now's the first time I ran into this...)

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 99.71616%. Comparing base (8850705) to head (ce806e9). Report is 443 commits behind head on main.

Additional details and impacted files

@@                 Coverage Diff                 @@
##                main       #3114         +/-   ##
===================================================
+ Coverage   99.58694%   99.71616%   +0.12922%     
===================================================
  Files            121         122          +1     
  Lines          18157       23605       +5448     
  Branches        3272        4022        +750     
===================================================
+ Hits           18082       23538       +5456     
+ Misses            52          48          -4     
+ Partials          23          19          -4     

Files with missing lines	Coverage Δ
src/trio/_core/_local.py	100.00000% <100.00000%> (ø)
src/trio/_core/_run.py	99.40000% <100.00000%> (+0.37275%)	:arrow_up:
src/trio/_core/_tests/tutil.py	100.00000% <ø> (ø)
src/trio/_tests/test_exports.py	99.68254% <100.00000%> (+0.06716%)	:arrow_up:

... and 51 files with indirect coverage changes

[jakkdl]

this makes me want to write a linter rule, but also feels ridiculous (and idk which flake8 plugin it would fit) to write an attrs+pyright-specific rule.

How slow/fast is the test? If it's anything but trivial I could rewrite it as an ast visitor.

[A5rocks]

this makes me want to write a linter rule, but also feels ridiculous (and idk which flake8 plugin it would fit) to write an attrs+pyright-specific rule.

How slow/fast is the test? If it's anything but trivial I could rewrite it as an ast visitor.

Yeah lint rule would make more sense but unfortunately a lint rule would miss context that's helpful (like whether a class is exported).

It takes 1.75 seconds to run locally, though that's with the overhead of starting pytest.

[A5rocks]

If test is slow, do we need to mark it with @slow or something?

Maybe? Here's the output at the end of pytest --durations=10 locally:

================================================ slowest 10 durations =================================================
3.66s call     src/trio/_tests/tools/test_gen_exports.py::test_process[from collections import Counter\n]
2.73s call     src/trio/_tests/tools/test_gen_exports.py::test_process[from collections import Counter\nimport os\n]
2.53s call     src/trio/_tests/tools/test_gen_exports.py::test_process[from typing import TYPE_CHECKING\nif TYPE_CHECKING:\n    from collections import Counter\n]
2.11s call     src/trio/_tests/test_socket.py::test_many_sockets
2.05s call     src/trio/_tests/test_socket.py::test_SocketType_connect_paths
2.04s call     src/trio/_tests/test_socket.py::test_address_in_socket_error
2.02s call     src/trio/_tests/test_highlevel_open_tcp_listeners.py::test_open_tcp_listeners_ipv6_v6only
1.73s call     src/trio/_tests/test_exports.py::test_pyright_recognizes_init_attributes
0.65s call     src/trio/_tests/test_exports.py::test_static_tool_sees_all_symbols[jedi-trio]
0.41s call     src/trio/_tests/tools/test_gen_exports.py::test_lint_failure

I don't think it's an appreciable slowdown. Maybe it's a good idea to put @slow on everything over 1 second? I can probably write the test to be faster (by not rescanning the tokenizations, not including tests in the glob). Actually I'll do just that.

EDIT: just the second optimization there changed it to 0.64s locally, so I think it's fine if not marked @slow.

[TeamSpen210]

There’s another approach we could use, but it’d might require starting an additional interpreter. Run a script which monkeypatches attrs.field to store whether alias was set into the metadata dict, import trio, then we can just loop through fields.

Unsure if that’d be faster though. We could avoid the new interpreter if the monkeypatch was done before any tests import trio, but that means it affects all tests. Probably not too much of an issue since metadata does nothing, and just wrapping the function should be fairly safe.

[A5rocks]

There’s another approach we could use, but it’d might require starting an additional interpreter. Run a script which monkeypatches attrs.field to store whether alias was set into the metadata dict, import trio, then we can just loop through fields.

Unsure if that’d be faster though. We could avoid the new interpreter if the monkeypatch was done before any tests import trio, but that means it affects all tests. Probably not too much of an issue since metadata does nothing, and just wrapping the function should be fairly safe.

It's not significantly faster:

(.venv) PS C:\Users\A5rocks\Documents\trio> hyperfine 'python -c \"import trio\"'
Benchmark 1: python -c "import trio"
  Time (mean ± σ):     410.5 ms ±  14.2 ms    [User: 45.3 ms, System: 22.9 ms]
  Range (min … max):   390.3 ms … 435.0 ms    10 runs

I think the main improvement would be that it's slightly more reliable. It sounds annoying to implement though.

[A5rocks]

@TeamSpen210 do you have any implementation ideas for your idea?

I was thinking about it just for the gain in reliability, but I think we would then get a flakey test because we delete and reimport trio to get all the warnings. I guess it's fine if we just patch attrs cause then the reimport will still be using the patched attrs?

[TeamSpen210]

Pushed an implementation using monkeypatching here. Unfortunately a downside I realised is that the monkeypatching code has to be either outside the trio package, or directly in __init__ before any other imports. So either this test can't be run by people just installing the package, or we'll have to have a module outside there. I could set it up to just skip if the monkeypatch "plugin" isn't specified though.

[Zac-HD]

I do like the "monkeypatch, and skip that test if not monkeypatched" plan 🙂

[Zac-HD]

@TeamSpen210 @A5rocks I'd really like to get this merged so I can use it; is there anything I can do to help?

[A5rocks]

@TeamSpen210 @A5rocks I'd really like to get this merged so I can use it; is there anything I can do to help?

Nope! I've just been pushing off working on some trio stuff for a bit, I'll incorporate better testing strategy soon:tm: (probably just ~~using our already created pytest plugin~~ nvm I forgot we can't + copy.copy() of the kwargs instead of picking and choosing. Everything else sounds fine)

[TeamSpen210]

Would you like me to make the monkeypatch approach a competing PR, or PR to yours, or do you want to stick with the current approach?

[A5rocks]

If you wouldn't mind, you should be able to commit to this PR.

[TeamSpen210]

Committed, didn't want to do big changes to other people's PRs without permission. Though it is easy to revert so probably that's fine.

[A5rocks]

Should we move the plugin to the (adjacent to src) tests directory or something? Having it in src feels weird...

[TeamSpen210]

That should work, if it's in sys.path so pytest can import.

[A5rocks]

That should work, if it's in sys.path so pytest can import.

Nevermind! I didn't realize pytest's sys.path schenanigans are the way they are. I'm surprised the test passes in CI! Maybe you can explain to me? I'm obviously missing something, because my mental model is:

• this works locally because pytest adds src/ to sys.path. then pytest looks there for the plugin
• but on ci... how? when? where? why would it add src/ to sys.path?

---

It feels like this shouldn't work -- if this is a pytest bug, I literally see no other way of adding to the path. The pythonpath option only applies after plugins (why??). ... well, other than setting the PYTHONPATH environment variable which would be annoying locally!

[TeamSpen210]

Not really sure either, but it's working so don't touch it?

[A5rocks]

Ohhhhh nooooo, it's installing _trio_check_attrs_aliases.py as a module into site-packages as part of the single wheel we make. That's quite the footgun!!!!

As a knee-jerk reaction I think we should revert back to before using a plugin. Maybe you see a way to work around this? Probably setting PYTHONPATH=......

list of packages in `site-packages`

  + ls 'C:\hostedtoolcache\windows\Python\3.11.9\x64\Lib\site-packages\trio/../'
  30fcd23745efe32ce681__mypyc.cp311-win_amd64.pyd
  3204bda914b7f2c6f497__mypyc.cp311-win_amd64.pyd
  MarkupSafe-2.1.5.dist-info
  OpenSSL
  OpenSSL-stubs
  README.txt
  __pycache__
  _black_version.py
  _cffi_backend-stubs
  _cffi_backend.cp311-win_amd64.pyd
  _distutils_hack
  _pytest
  _trio_check_attrs_aliases.py
  alabaster
  alabaster-0.7.13.dist-info
  astor
  astor-0.8.1.dist-info
  astroid
  astroid-3.2.4.dist-info
  async_generator
  async_generator-1.10.dist-info
  attr
  attrs
  attrs-24.2.0.dist-info
  babel
  babel-2.16.0.dist-info
  black
  black-24.8.0.dist-info
  blackd
  blib2to3
  build
  build-1.2.2.post1.dist-info
  certifi
  certifi-2024.8.30.dist-info
  cffi
  cffi-1.17.1.dist-info
  cffi-stubs
  charset_normalizer
  charset_normalizer-3.3.2.dist-info
  click
  click-8.1.7.dist-info
  codespell-2.3.0.dist-info
  codespell_lib
  colorama
  colorama-0.4.6.dist-info
  coverage
  coverage-7.6.1.dist-info
  cryptography
  cryptography-43.0.1.dist-info
  dill
  dill-0.3.9.dist-info
  distutils-precedence.pth
  distutils-stubs
  docutils
  docutils-0.20.1.dist-info
  docutils-stubs
  idna
  idna-3.10.dist-info
  imagesize
  imagesize-1.4.1.dist-info
  imagesize.py
  iniconfig
  iniconfig-2.0.0.dist-info
  isort
  isort-5.13.2.dist-info
  jedi
  jedi-0.19.1.dist-info
  jinja2
  jinja2-3.1.4.dist-info
  markupsafe
  mccabe-0.7.0.dist-info
  mccabe.py

[TeamSpen210]

We could set PYTHONPATH I think yes? I was going to do that but didn't trust myself with changing the complicated shell script. Probably should put the plugin in its own folder to isolate. I don't think it matters that the test won't run other than in CI, since it's only going to fail if the source code changes, not due to environment etc.

Another approach would be to use usercustomize perhaps, copying the module into site-packages ourselves.

[TeamSpen210]

The plugin script can't be anywhere in the trio package, because then trying to import it implicitly imports trio first, before we hooked anything. That's why I gave it such a long name.

[A5rocks]

The plugin script can't be anywhere in the trio package, because then trying to import it implicitly imports trio first, before we hooked anything. That's why I gave it such a long name.

Yeah the issue with that commit was that we had -p trio._tests.pytest_plugin in addopts, which then messes things up because turns out pytest runs plugins in the order they're provided. Anyways we don't need that anymore, so I removed it.

---

nevermind, I should know better than make statements based on things running locally... turns out conftest always worked except for --run-slow and that's why we switched. :/ I completely forgot.