feat: discover schema with lifetimes and const generics

[Calum4]

Fixed as issue where schema with lifetimes were ignored due to the overly eager .generics.params.is_empty() found here and here.

Disclaimer that I'm not at all familiar with the inner workings of the crate so there may be unintended consequences I've overlooked.

[DenuxPlays]

Please also run cargo fmt to fix the ci

[Calum4]

Apologies! I presumed fmt wasn't in use after running it and seeing a litany of changed files in my diff. Little did I know it was git converting the files to CRLF, sigh.

Should be sorted now

[DenuxPlays]

I still do not get it.

How does your change add functionality?

You filter the list so it gets smaller how does this is better then a simple .is_empty() on the same unfiltered list?

[Calum4]

I have created a new branch with the fix removed to demonstrate the problem.

For your convenience, the output of running tests:

thread 'default_features::test::test_lifetimes' panicked at utoipauto\tests\default_features\test.rs:146:14:
no components
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    default_features::test::test_lifetimes

test result: FAILED. 9 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

error: test failed, to rerun pass `-p utoipauto --test test`

As you can see from this line, if there are any generics on a struct the value of is_generic is true. This includes lifetimes which are considered generic.

Struct with lifetimes

#[derive(ToSchema)]
pub struct LifetimeStructSchema<'a> {
    foo: &'a str,
}

Output of dbg!(&s.generics.params);

[utoipauto-core\src\discover.rs:89:17] &s.generics.params = [
    GenericParam::Lifetime(
        LifetimeParam {
            attrs: [],
            lifetime: Lifetime {
                apostrophe: #143 bytes(7503..7574),
                ident: Ident {
                    ident: "a",
                    span: #143 bytes(7503..7574),
                },
            },
            colon_token: None,
            bounds: [],
        },
    ),
]

Now if we look at the definition of parse_from_attr(), the struct from the example above reaches this evaluation however due to the value of is_generic being true, does not meet the condition and therefore the schema is not discovered.

In order to fix this, my solution is to filter out of s.generics.params any lifetimes. I do this because it is my understanding rather than broadly checking for all generics, your intention is instead to check for type and const generics.

Perhaps I should also rename is_generic to clarify the updated meaning.

[DenuxPlays]

Ahh okay now I understand.

But then I would rather check if we have any GenericParam::Type as this is the Type of Generic we want.
We do not want to include Lifetimes or Consts.

[Calum4]

I'll take a further look at this when I get time, hopefully tomorrow. Currently this does not compile for whatever reason

error[E0107]: missing generics for struct `ConstGenericStructSchema`
  --> utoipauto\tests\default_features\test.rs:62:1
   |
62 | #[utoipauto(paths = "./utoipauto/tests/default_features")]
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected 1 generic argument
   |
note: struct defined here, with 1 generic parameter: `N`
  --> utoipauto\tests\default_features\const_generics.rs:4:12
   |
4  | pub struct ConstGenericStructSchema<const N: usize> {
   |            ^^^^^^^^^^^^^^^^^^^^^^^^ --------------
   = note: this error originates in the attribute macro `utoipauto` (in Nightly builds, run with -Z macro-backtrace for more info)
help: add missing generic argument
   |
62 | #[utoipauto(paths = "./utoipauto/tests/default_features")]<N>
   |                                                           +++

[DenuxPlays]

Very weird

I am also enhancing the acceptance tests to cover generics.
If you need any help let me know. :)

[Calum4]

I've managed to track down the issue.

#[derive(utoipa::ToSchema)]
pub struct ConstGenericStructSchema<const N: usize> {
    foo: [u16; N],
}

The struct above once discovered is internally represented as crate::ConstGenericStructSchema. This once inserted into the openapi attribute is problematic, demonstrated below.

Current output which fails to compile:

#[openapi(components(schemas(crate::ConstGenericStructSchema)))]

Modified output which successfully compiles:

#[openapi(components(schemas(crate::ConstGenericStructSchema<0>)))]

Best I can see, the value of N is totally disregarded in the final openapi representation:

{
  "properties": {
    "foo": {
      "items": {
        "format": "int32",
        "minimum": 0,
        "type": "integer"
      },
      "type": "array"
    }
  },
  "required": [
    "foo"
  ],
  "type": "object"
}

I do not understand the implications of my proposed solution, however I would suggest we instead internally represent the struct as crate::ConstGenericStructSchema<x> where x is the default value of const generic's type. In this case usize::default() -> crate::ConstGenericStructSchema<0>.

I'm happy to code this up, if you deem it an acceptable solution

[DenuxPlays]

No not at all

Generic schemas are meant to used with the aliases attribute.
This is mentioned in utopia.

[DenuxPlays]

OpenApi does not support generic Schemas afaik.
So we have to hard code all the generics via the alias macro sadly.

[Calum4]

Ah well, it appears my lack of knowledge on utoipauto and utoipa in general has shown itself.

I'll see what I can figure out with this new info, thanks

[DenuxPlays]

No I find it very confusing too.
You keep saying that you have limited knowledge but you figured out quit a bit 👍

[DenuxPlays]

Your last commit looks really good on the first glance.
I am working on an acceptance test for generics.

Also I am reworking how special imports are processed.

I have to think how to develop everything in parallel and merge them with one semi-big pr. (Otherwise the acceptance tests will fail!)

But that's a task for tomorrow let me know when you finish this pr.

For now:
Thank you for all your work :)

[Calum4]

Thank you for all your work :)

No problem at all, thanks for your work on the crates! This has presented me a good excuse to finally learn proc macros

I've managed to get lifetimes, const and type generics all to work nicely in isolation, however when mixing lifetimes with either results in a compile error

From my initial looks it might be a difficult one to solve, but I'll take a stab at it and let you know how it goes

[Calum4]

Okay, tests pass and I believe this is good to go. Thanks for your help on this!

[DenuxPlays]

No thank you for doing this :)

I'll take a look when I find more time.

[DenuxPlays]

Looks good to me.

[DenuxPlays]

@ProbablyClem Would be great if you can look at this too.
Maybe I missed something

[DenuxPlays]

Okay
The unit tests work for now.
I will later add an acceptance test.

@Calum4 Thank you for your work :)