Add missing attribute for struct fields

[teohhanhui]

Closes #32

[alexjg]

Left a few comments. Regarding testing, I prefer an end-to-end test where you define a type which uses the attribute in question and then check that it behaves the way we expect when hydrating. Something like:

#[derive(Debug, Hydrate)]
struct MaybeString {
    #[autosurgeon(missng=std::default::Default)]
    value: Option<String>
}

#[test]
fn hydrate_missing() {
    let mut doc = automerge::AutoCommit::new();
    let value: MaybeString = hydrate(&doc).unwrap();
    assert!(value.is_none());
}

[alexjg]

I think we'll want to use quote_spanned here to make sure that the compile error produced is comprehensible. It may be worth adding a trybuild based UI test to test the errors as described in https://ferrous-systems.com/blog/testing-proc-macros/#diagnostics-error-paths (I didn't do that previously because this was my first major procedural macro project).

[teohhanhui]

Sorry, I'm still not sure what is the right span to use when...

For example, should I use field.span() here: https://github.com/automerge/autosurgeon/pull/39/files#diff-39105dca315265219e34078932ac265ac7c14d4a5c46958808d175aa7b2de123L265-L276

If I'd like to do something similar to https://github.com/automerge/autosurgeon/pull/39/files#diff-39105dca315265219e34078932ac265ac7c14d4a5c46958808d175aa7b2de123R278-R285

[alexjg]

What do you mean by "do anything about the key"?

I don't think we need support for enums (provided we support variant fields).

[teohhanhui]

What do you mean by "do anything about the key"?

As in, migrating the key, from no key to some key. But perhaps that's not a possible scenario and I'm simply mistaken.

EDIT: Now that I think about it, that doesn't seem like something a default would help with anyway...

I don't think we need support for enums (provided we support variant fields).

Okay, so I'll have to add support for variant fields at least. 🙈

[teohhanhui]

#[derive(Clone, Debug, PartialEq, Eq)]
struct Inner(u64);

#[derive(Clone, Debug, PartialEq, Eq, Hydrate)]
#[autosurgeon(hydrate = "hydrate_outer", missing = "Default::default")]
struct Outer(Inner);

This should be a parse error, but isn't... I'm not sure how to fix this without a refactoring of the parsing code.

Apparently we skip everything if there's hydrate_with at the container level:

autosurgeon/autosurgeon-derive/src/hydrate.rs

Lines 25 to 27 in 4d9ad30

	if let Some(hydrate_with) = container_attrs.hydrate_with() {
	return proc_macro::TokenStream::from(on_hydrate_with(&input, &hydrate_with));
	}

[teohhanhui]

ping @alexjg

[alexjg]

Apologies, I missed this question. Will take a look tomorrow, maybe we will need to refactor a little. To be clear, the impact right now is that if you have a hydrate attribute the missing attribute is ignored?

[teohhanhui]

To be clear, the impact right now is that if you have a hydrate attribute the missing attribute is ignored?

Only on the container, but we already ignore everything else if hydrate (or with) is present on the container.

Currently that would not produce parse errors, so it might be a bit confusing for the user... Their attributes are just silently ignored. Perhaps we could fix that in another PR?

[alexjg]

Yeah let's fix the parse errors in another PR

[alexjg]

This is now published as 0.8.2. Thanks @teohhanhui !

[teohhanhui]

I think you forgot to push the v0.8.2 tag?

[alexjg]

Good spot, done.