Is proc-macro-attribute application order defined?

[CAD97]

In code, if I write

#[strip_attrs_on_item]
#[this_attr_always_panics]
struct Foo;

Is it guaranteed that strip_attrs_on_item will receive the token stream # [ this_attr_always_panics ] struct Foo ; and that if it does what it says on the tin and emits struct Foo ;, this_attr_always_panics is not called?

As an addendum, is this_attr_always_panics even resolved, if it's transformed before application?

#565 is probably related. rust-lang/rust#63336 is this issue.

derive macros have a special path to say "I handle this attribute" so derive(serde::Serialize) implies that you can use #[serde(..)] afterwards, and the derive gives it meaning. If expansion order isn't guaranteed outer before inner, then replacing the derive with an attribute would remove this ability.

This is not #578, which is about attributes within the same crate but on different items.

[Centril]

cc @petrochenkov @eddyb

[Luro02]

This comment might be related rust-lang/rust#67839 (comment)