Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
pydantic
main.py					752
pydantic/_internal
_generate_schema.py
Project Total

This report was generated bypython-coverage-comment-action

CodSpeed Performance Report

Merging#11434 willimprove performances by 67.14%

_{ComparingMarkusSintonen:allow-generic-ref (7e7c6dd) withmain (ff3789d)}

Summary

⚡ 1 improvements
✅ 45 untouched benchmarks

Benchmarks breakdown

please review

@MarkusSintonen,

Woah, super small diff, incredible perf boost. We need to look into consequences here, but all tests passing seems quite promising. I've pinged@dmontagu to double check this (he did a lot of the initial work on generics), and@Viicos will also probably want to have a look.

I'm running third party tests to see if any integrations fail.

Third party tests passing seems like a promising start.

I'll just share that I wrote an (in-my-opinion) complex recursive generic and tried to get validation to pass when it should not, and was not able to:

from __future__importannotationsfrompydanticimportBaseModelfromtypingimportTypeVar,GenericT=TypeVar('T')S=TypeVar('S')classA(BaseModel,Generic[T,S]):t_a:T|None=Nones_a:S|None=Noneb:B[T,S]|None=NoneclassB(BaseModel,Generic[T,S]):t_b:T|None=Nones_b:S|None=Nonec:C[T,S]|None=NoneclassC(BaseModel,Generic[T,S]):t_c:T|None=Nones_c:S|None=Nonea:A[T,S]|None=Nonea1=A[B[int,str],C[bool,float]](t_a=B(t_b=2,s_b='b',c=C(t_c=2,s_c='b')),s_a=C(t_c=True,s_c=1.0,a=A(t_a=True,s_a=1.0)),)b1=B[B[int,str],C[bool,float]](t_b=a1.t_a,s_b=a1.s_a,c=C[B[int,str],C[bool,float]](a=a1))data= {'t_b': {'t_b':2,'s_b':'b','c': {'t_c':2,'s_c':'b','a':None}},'s_b': {'t_c':True,'s_c':1.0,'a': {'t_a':True,'s_a':1.0,'b':None}},'c': {'t_c':None,'s_c':None,'a': {'t_a': {'t_b':2,'s_b':'b','c': {'t_c':2,'s_c':'b','a':None}},'s_a': {'t_c':True,'s_c':1.0,'a': {'t_a':True,'s_a':1.0,'b':None}},'b':None,        },    },}b1.model_validate(data)

If I edited any of the stuff indata, it was producing the validation errors I expected, not letting some data through like I feared it might.

So I feel like whatever concerns I had might have been mitigated by improvements to our generics handling in the time since I was responsible for it 😅.

@MarkusSintonen I'm curious how much of a startup speed improvement you noticed on your own codebase. Was it significantly more/less than the codspeed benchmark here, or comparable? (Basically trying to get an intuition for how representative this codspeed example is of your usage patterns.)

this is awesome. Thank you@MarkusSintonen.

@MarkusSintonen I'm curious how much of a startup speed improvement you noticed on your own codebase. Was it significantly more/less than the codspeed benchmark here, or comparable? (Basically trying to get an intuition for how representative this codspeed example is of your usage patterns.)

It is quite close to what we see here from codspeed. I got about +50-60% in the full initialization time of the whole FastAPI stack utilizing generics with different complexities.

I'll just share that I wrote an (in-my-opinion) complex recursive generic and tried to get validation to pass when it should not, and was not able to

@dmontagu do you think it would make sense to add your test as part of the unit tests here if you are unsure if the case is covered?

do you think it would make sense to add your test as part of the unit tests here if you are unsure if the case is covered?

Yes, let's!

I thinktest_generic_recursive_models_complicated already covers this example, although it could be tweaked a bit?