Fix testing infrasucture for Pydantic 2 default

[pkalita-lbl]

Fixes #2040

Summary

• With Pydandic 2.x.x specified in our poetry.lock file, running tests directly via pytest or the tox py3x testenvs will use Pydantic v2. These changes replace the existing "special case" pydantic2 testenv with a pydantic1 testenv. The new testenv operates the same as the old one, except swapping Pydantic v1 for v2.
• Previously the main.yaml testing workflow had a pydantic-version variable in the configuration matrix. With the addition of Python 3.11 and 3.12 to the testing matrix (Add support for Python 3.12 #1704), the matrix permutations were starting to get a little unruly. Here I'm splitting off Pydantic 1 testing into a separate workflow (.github/workflows/test-pydantic1.yaml). I then updated the exclude combinations in main.yaml so that what we end up running is:
  • All supported Python version with Pydantic 2 on Ubuntu (main.yaml)
  • The oldest and newest supported Python versions with Pydantic 2 on Windows (main.yaml)
  • The latest supported Python version with Pydantic 1 on Ubuntu (test-pydantic1.yaml)
• The updated testing workflows caught a few issues we had missed with our previous configuration matrix. @sneakers-the-rat fixed a few issues in the Pydantic generator that crop up when using older Python versions. I updated a few compliance tests to accurately reflect the actual behavior of Pydantic on older Python versions.

[codecov]

Codecov Report

Attention: Patch coverage is 60.00000% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 80.64%. Comparing base (8a28750) to head (b5d0fb8).
Report is 1 commits behind head on main.

Files	Patch %	Lines
linkml/generators/pydanticgen/array.py	60.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2041      +/-   ##
==========================================
- Coverage   80.67%   80.64%   -0.03%     
==========================================
  Files         107      107              
  Lines       11943    11946       +3     
  Branches     3415     3416       +1     
==========================================
- Hits         9635     9634       -1     
- Misses       1743     1744       +1     
- Partials      565      568       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[pkalita-lbl]

Notes to self: seems like we've never actually tested the output of the Pydantic generator, targeting v2, on Python 3.8 or 3.9. The 3.8 tests are failing with a lot of

ImportError: cannot import name 'Annotated' from 'typing'

Makes sense, because that was introduced in Python 3.9 (https://docs.python.org/3/library/typing.html#typing.Annotated)

On 3.9 the errors are mostly

AttributeError: '_SpecialForm' object has no attribute '__name__'

They also seem to be related to array test cases. Needs further investigation.

[sneakers-the-rat]

oop how did i miss that. assigning myself bc these are my fault probs

[sneakers-the-rat]

fixed the array generation and version problems. unassigning myself bc that was the 'my fault' part.

the remaining test failures look like they're from pydantic coercion, where a model like

class C(BaseModel):
    s1: Optional[bool] = None

incorrectly validates with

inst = C(s1=1.1)

as

C(s1=True)

which is a pydantic coercion thing. idk what would be the desired behavior there - we could enable strict mode ( #1955 ) or we could mark the expected behavior as COERCES. i'll defer to y'all on what the fix should be.

weird that it's happening only in older python versions though, don't know why that would be.
edit: even weirder, only on older versions in linux, the windows tests pass.

[pkalita-lbl]

Much appreciated @sneakers-the-rat! I will take care of the last few failures.

[pkalita-lbl]

I think our best move here is to just mark those narrows cases in the compliance tests as having coerce as the expected behavior. I didn't find any issues in the Pydantic repo that directly address what we're seeing on Python < 3.10 even though it directly contradicts what they say in their docs.

The error is coming from Pydantic's Rust core. Because of that I'm not totally surprised that the issue seems to only effect certain Python versions on Ubuntu and not Windows.

[sierra-moxon]

not 3.12 right? (pydantic1 seems to be setting up a 3.12 environment)

[pkalita-lbl]

Right. These are excludes, and when we subtract them from the default permutations they leave us with only Python 3.8 and 3.12 on Windows.

[sierra-moxon]

I see, I missed the excludes part.

[sierra-moxon]

thank you for this doc.

[sneakers-the-rat]

Go team!!!