Enum validator improvements

[samuelcolvin]

Change Summary

While looking through how we do enum validation in preparation for moving some of it to Rust, I saw a few things that could be improved. Hopefully this provides a "cleaner" schema, and should improve performance a bit.

Performance gain is small, but not insignificant:

before

In [1]: import enum
   ...: from pydantic import TypeAdapter
   ...: 
   ...: class Foo(enum.IntEnum):
   ...:     x = 1
   ...:     y = 2
   ...: 
   ...: 
   ...: ta = TypeAdapter(Foo)
   ...: 
   ...: %timeit ta.validate_python(2)
   ...: %timeit ta.validate_json('2')
   ...: 
550 ns ± 2.37 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)
560 ns ± 2.29 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

After

In [1]: import enum
   ...: from pydantic import TypeAdapter
   ...: 
   ...: class Foo(enum.IntEnum):
   ...:     x = 1
   ...:     y = 2
   ...: 
   ...: 
   ...: ta = TypeAdapter(Foo)
   ...: 
   ...: %timeit ta.validate_python(2)
   ...: %timeit ta.validate_json('2')
   ...: 
508 ns ± 0.84 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)
517 ns ± 1.95 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

Checklist

• The pull request title is a good summary of the changes - it will be used in the changelog
• Unit tests for the changes exist
• Tests pass on CI
• Documentation reflects the changes where applicable
• My PR is ready to review, please add a comment including the phrase "please review" to assign reviewers

[cloudflare-pages]

Deploying pydantic-docs with    Cloudflare Pages

Latest commit:	8bd46c3
Status:	✅  Deploy successful!
Preview URL:	https://11ebbd3f.pydantic-docs2.pages.dev
Branch Preview URL:	https://enum-validator-improvements.pydantic-docs2.pages.dev

View logs

[codspeed-hq]

CodSpeed Performance Report

Merging #9045 will not alter performance

_{Comparing enum-validator-improvements (8bd46c3) with main (09f4456)}

Summary

✅ 10 untouched benchmarks

[samuelcolvin]

output is the same, but it's more correct to let pydantic-core take care of replacing the variables in the message template.

[samuelcolvin]

I've replaced lots of use of core_schema.chain_schema with core_schema.no_info_after_validator_function which should be more efficient, and is more concise.

[samuelcolvin]

attrgetter is nearly twice as fast as a lambda:

In [1]: import enum

In [2]: class Foo(enum.Enum):
   ...:     x = 'a'
   ...:     y = 'b'
   ...: 

In [3]: with_lambda = lambda thing: thing.x

In [4]: %timeit with_lambda(Foo)
39.1 ns ± 0.161 ns per loop (mean ± std. dev. of 7 runs, 10,000,000 loops each)

In [5]: from operator import attrgetter

In [6]: using_attrgetter = attrgetter('x')

In [7]: %timeit using_attrgetter(Foo)
24.5 ns ± 0.473 ns per loop (mean ± std. dev. of 7 runs, 10,000,000 loops each)

[sydney-runkle]

Should we be using attrgetter more broadly?

[sydney-runkle]

Looks great! This is much more simple than the existing behavior!

[sydney-runkle]

Should we be using attrgetter more broadly?