Clarify the semantics of enum discriminants

[nodakai]

cf. https://doc.rust-lang.org/error-index.html#E0082

The default type for enum discriminants is isize, but it can be adjusted by adding the repr attribute to the enum declaration.

It would be great if anyone could check my English.

[rust-highfive]

r? @steveklabnik

(rust_highfive has picked a reviewer for you, use r? to override)

[apasel422]

@pnkfelix "specify"?

[pnkfelix]

@apasel422 works for me

[Aatch]

The error explanation is somewhat misleading, we actually use the smallest valid integer type (with a preference for unsigned integers) when no repr is supplied. While typechecking treats the RHS as an isize for, well, checking, to say it "defaults" to it implies that it will be represented as such. Especially given the context.

We probably shouldn't replicate that.

[nodakai]

@Aatch Are you suggesting to fix error-index.html? Could you elaborate on what you meant by "replicate" ? I believe my patch isn't making the situation worse in terms of an ambiguity on what defaults to isize in enum definition.

[Aatch]

@nodakai it's the same issue here as in error-index.html. It says "the default type for enum discriminants is isize", which is kinda true, but it suggests that the enum discriminants will be represented using isize by default, which isn't true. Given that both this patch and error-index.html are talking about #[repr(X)] in the same place, it actually makes the semantics less clear.

[nodakai]

@Aatch So, what should I do? Are you suggesting me to fix error-index.html at the same time?

[Aatch]

@nodakai I think I may have been misremembering your wording earlier, confusing it with the error-index.html wording, so apologies for that. Your wording is actually fine in this respect. If you want to update the error index at the same time, go for it.

[steveklabnik]

I would agree with "specify"

r=me after

[nodakai]

Extended the scope of the patch to include error-index.html

Avoided the word "assign" in referring to enum declarations because "assign" often implies an update of a mutable state.

[steveklabnik]

@bors: r+ rollup

Thanks!

[bors]

📌 Commit 969d027 has been approved by steveklabnik