Support #[serde(with = "...")] for struct fields

[escritorio-gustavo]

Goal

Ensure the #[serde(with = "...")] attribute is properly handled in a way that prevents users from accidentally generating wrong TS type definitions by forgetting to add #[ts(as = "...")]/#[ts(type = "...")],

Why?

The #[serde(with = "...")] attribute changes how the type is serialized, therefore, it changes what type will be received by TypeScript, so the type generated by default will most likely be incorrect. The user must use #[ts(as = "...")] or #[ts(type = "...")] to ensure their type definition matches the type generated by serde

Why not make #[serde(with = "...")] an alias for #[ts(as = "...")]

#[ts(as = "...")] only accepts types, where as #[serde(with = "...")] accepts either a type or a path to a module containing both a serialize and a deserialize function, so the attributes are not compatible and thus cannot just alias each other

Closes #275

Changes

When the serde-compat feature is enabled, using #[serde(with = "...")] will demand the use of either #[ts(as = "...")] or #[ts(type = "...")], resulting in a compiler error if neither is used

Checklist

• I have followed the steps listed in the Contributing guide.
• If necessary, I have added documentation related to the changes made.
• I have added or updated the tests related to the changes made.

Todo

• #[serde(with = "...")] is a valid variant attribute, but I'm not yet clear on how to implement it, as neither #[ts(as = "...")] or #[ts(type = "...")] are valid variant attributes.
  • This is being explored in Support #[ts(as = "...")] and #[ts(type = "...")] on enum variants #284
• Find a way to test the compiler error case (Maybe compiletest_rs or trybuild?)

[NyxCode]

Huh, this is much less complicated than I thought it might end up being, nice!

#[serde(with = "...")] is a valid variant attribute, but I'm not yet clear on how to implement it, as neither #[ts(as = "...")] or #[ts(type = "...")] are valid variant attributes.

Hm. I guess we could at support for them for variants, or leave this edge-case for now. Will have to look into how complicated supporting these attributes for variants would be first, though I think it might be pretty straight forward.

Find a way to test the compiler error case (Maybe compiletest_rs or trybuild?)

I think that's supported in doctests with "```compile_fail ....`". Not super clean, but it's simple, so maybe that's enough?

[escritorio-gustavo]

I think that's supported in doctests with "```compile_fail ....`". Not super clean, but it's simple, so maybe that's enough?

It is, but I'm not sure if a doctest in the tests directory would be executed by cargo test

[escritorio-gustavo]

Hm. I guess we could at support for them for variants, or leave this edge-case for now. Will have to look into how complicated supporting these attributes for variants would be first, though I think it might be pretty straight forward.

That might require two new PRs, one for #[ts(as = "...")]/#[ts(type = "...")] and another for #[serde(with = "...")]. Otherwise I think this one would get too large

[NyxCode]

Agreed!

[NyxCode]

Didn't find the time to take a closer look earlier, sorry about that.
Everything looks good to me here! I was still wishing that we could actually support #[serde(with = "..")], but this is still better than the status quo and is actually robust. I'm sold!

[LZQCN]

Good PR, Please release a new version that includes these updates. I would really like to use them in my project. Thank you!

[escritorio-gustavo]

Hello @LZQCN! These changes will eventually be released with version 9 of the library, but we still need to complete #284, which covers this same feature on enum variants. It hasn't been merged yet because I couldn't quite figure out a way to test it and compare the results with serde's output.