Fix #89: #[derive(Schema)] for StructVariants

[barafael]

Here comes a 4-commit sandwich.

The actual issue I believe is fixed with this. However, there are 2 open questions regarding the schema tests (see commit message bodies):

• Where should the schema tests be located? As of now, they force the schema module to be public because they are integration tests. A doc comment above mod schema; explicitly says not to make experimental features pub 😆
• The test for struct serialization does not pass. It seems it wants the array member to be schematized as a Seq, but the array member is schematized as a 10-element Tuple.

Let me know what you think :) I love using this crate.

[netlify]

✅ Deploy Preview for cute-starship-2d9c9b canceled.

Name	Link
🔨 Latest commit	6906f4b
🔍 Latest deploy log	https://app.netlify.com/sites/cute-starship-2d9c9b/deploys/66421acdcd942c0008f6de3f

[jamesmunns]

Hey @barafael, overall looks good!

1. can we use the scoped/conditional exports, e.g. postcard::experimental::schema instead of making postcard::schema public? If not, could you please slap it with a #[doc(hidden)] and a comment that says "this isn't API stable, just an impl detail"?
2. Re: 5d0e703, I dunno, I'll have to look, that looks very seq-y to me.

Thanks!

[barafael]

1.: maybe a simple conditional pub is good enough?

[barafael]

Re: 2, not sure what to do about it, @jamesmunns could you have a look?

[jamesmunns]

@barafael I fixed some merge conflicts, and looked at the test.

I believe the test as-is now is correct: fixed size arrays, such as [T; N], should be serialized as tuples, as the length is known, so the length is not sent on the wire. Slices (and Vecs) such as &[T] should be serialized as Seqs, as their length WILL be sent on the wire.

Thank you!