Parse @unsafe nonisolated conformance

[DougGregor]

A protocol conformance can have a conformance attribute (like @unsafe) along with nonisolated. Extend the syntax tree to allow type specifiers after the type attributes, and parse nonisolated in that position.

Fixes #3109.

[DougGregor]

@swift-ci please test

[DougGregor]

@swift-ci please test

[DougGregor]

@swift-ci please test

[DougGregor]

@swift-ci please test Windows

[DougGregor]

@swift-ci please test

[DougGregor]

@swift-ci please test Windows

[ahoppen]

Thank you. It’s a little sad that we are not consistent with the ordering of attributes and specifiers in front of types but I assume we can’t do anything about that anymore, can we? That’s probably also why I’m not a huge fan of the name lateSpecifiers but I don’t have a better suggestion, it’s about as clear as it gets.

Would like to hear @rintaro’s feedback as well.

[ahoppen]

Interesting formatting choice by swift-format here. swift-format seems to be happy with the following as well, which I’d prefer

Suggested change

	case .collection(_, collectionElementName: let childElt?, _, _, generateDeprecatedAddFunction: true) = child
	.kind
	case .collection(_, collectionElementName: let childElt?, _, _, generateDeprecatedAddFunction: true) =
	child.kind

[DougGregor]

Formatting fix in #3114 because I too eagerly pushed "merge"

[DougGregor]

Thank you. It’s a little sad that we are not consistent with the ordering of attributes and specifiers in front of types but I assume we can’t do anything about that anymore, can we? That’s probably also why I’m not a huge fan of the name lateSpecifiers but I don’t have a better suggestion, it’s about as clear as it gets.

Would like to hear @rintaro’s feedback as well.

An alternative approach here would be to handle the specifiers different for the places where we can have modifiers, i.e. parameters (where things like inout go before the type) and conformances (where nonisolated goes after the attributes). I suppose we could revisit the language design here to force specifiers like this to be first, i.e., nonisolated @preconcurrency. Parameters are inconsistent with declarations, so conformances could choose which one to align with.