Nullary variants should be considered rvalues, not static identifiers.

[nikomatsakis]

I think there's an existing issue, but I can't find it. Currently borrowck treats nullary variants (like None) as static identifies. This doesn't make sense for generic enums and I don't think it matches up with trans. The reason it doesn't make sense for generic enums is that None::<Foo> and None::<Bar> have totally different sizes etc., so there couldn't possible be a single static constant to refer to. This is a relatively simple fix to the mem_categorization code, but it's worth investigating how trans handles enum variants and make sure that everything will be compatible.

Nominating since it is a backwards compatibility thing.

[nikomatsakis]

I just checked trans and indeed it is always generated as an rvalue.

[pnkfelix]

(is this somehow related to #8229 ? Just curious...)

[pnkfelix]

Accepted for 1.0, P-backcompat-lang

[dmski]

Should this be just closed? It looks like a duplicate of #11681 (which is closed).

[flaper87]

@dmski looks like you're right. Tagging as needs test.

[dmski]

There is https://github.com/mozilla/rust/blob/master/src/test/compile-fail/regions-lifetime-of-struct-or-enum-variant.rs which looks like it tests exactly this. Do you think that another test for this can be added?

[flaper87]

@dmski looks sufficient. I had forgotten about that test :D