Use `TypedDict` instead of `BaseModel` tip

[Kludex]

I should add this tip.

[yohann84L]

@Kludex in reference to pydantic#use-typeddict-over-nested-models ?

[Kludex]

That's one reason yeah, the other was the NotRequired, but now we have MISSING in Pydantic...

[Jmandarino]

this would be lovely. I am working on a small presentation of performance improvements but need to test the following (Might be sligtly off topic but maybe some ideas of what to include if you have the answers already?)

• TypeAdapter
• OrJson
• SQL alchemy to pydantic can be slow (even with from_attributes=True) using large nested data sets.
• Large output in general takes pydantic struggles Long time between request and endpoint code get executed for large List data fastapi/fastapi#9136
• wanted to look into if nested models were super slow (e.g a Book has an FK to author and I return both of those in pydantic)

FWIW I love fastapi but if you want to eek out performance there are a lot of ways to do it wrong. your docs @Kludex have been super helpful :)

[Kludex]

Large output in general takes pydantic struggles fastapi/fastapi#9136

I think that's too old.

OrJson

I'm not sure this helps. I think the problem is that there's a layer of JSONResponse instead of using TypeAdapter with dumping directly into Response.

[Jmandarino]

I think that's too old.

that's fair but I think its still an issue.

I can try to post an example on fastapi but I still have similar issues
we have an unpaginated endpoint with 10k items (4.9MB uncompressed total data). most of endpoint time is taken via pydantic.

I'm not sure this helps. I think the problem is that there's a layer of JSONResponse instead of using TypeAdapter with dumping directly into Response.

Also fair! def spitballing ideas so feel free to take what you like if any 😄