Derive macro configuration does not support `cfg_attr` that expands to multiple attributes

[Veykril]

#16789

[Dessix]

I believe I may have just run into this with sea-orm in its latest version.

Rust-Analyzer is reporting the following error when rustc accepts the code when using multi-predicate cfg_attrs:

Missing macro attribute \`rs_type\`

For anyone hitting this on conditional code like the following:

#[cfg_attr(
  feature = "sea-query",
  derive(EnumIter, DeriveActiveEnum),
  sea_orm(
    db_type = "String(StringLen::N(64))",
    rs_type = "String",
  ),
)]
pub enum MyEnum {
  // ...
}

Manually expanding to two separate predicative attributes resolves the issue:

#[cfg_attr(feature = "sea-query", derive(EnumIter, DeriveActiveEnum))]
#[cfg_attr(feature = "sea-query", sea_orm(rs_type = "String", db_type = "String(StringLen::N(64))"))]
pub enum MyEnum {
  // ...
}

I am unsure why this started cropping up only in the most recent version of SeaORM or rust-analyzer for me, but the Rust reference documentation shows that this attribute accepts multiple attributes in its Attrs grammar rule. I hope this is what this issue was about, but the linked thread didn't go into much detail about what "handl[ing] multiple attributes" entailed. I'm posting this for searchability, as I spent a while digging through sea-orm instead of RA trying to track this one down. Being that I commonly use this syntax for 6+ serde attributes, I'm surprised I haven't run into this sooner.

[Veykril]

Ah, you are running into something differently here. I believe the case where cfg_attr expands to multiple things is fixed already (what this issue was about). What you are running into is a very special case that we haven't considered, namely where the derive is within a cfg_attr that expands to more things. When we expand the derive, we have to strip the derive invocation before passing the item as the attribute, but we do that pre-cfg_attr expansion so we strip the entire cfg_attr in your first example and which is why the second snippet works fine.

[Veykril]

#8434 has some more context

[Dessix]

My apologies for posting it to the wrong issue then- but I'm glad I helped catch a subtle bug ^^

[wyatt-herkamp]

@Veykril and @Dessix

This issue already exists here. #10110

I meant to fix it. However, I have not had time recently.