[SE-0461] Replace `@execution(...)` with `@concurrent` and `nonisolated(nonsending)` #80668

xedin · 2025-04-09T07:41:33Z

Add @concurrent declaration and type attributes to replace @execution(concurrent)
Add new nonsending modifier to nonisolated declaration modifier
Make it possible to use nonisolated(nonsending) in type and expression contexts (modeled via CallerIsolatedTypeRepr)
Replace @execution(caller) with nonisolated(nonsending) modifier
Remove all of the references to @execution attribute throughout the codebase
Remove ExecutionAttribute experimental flag

xedin · 2025-04-09T07:42:19Z

swiftlang/swift-syntax#3047
@swift-ci please test

xedin · 2025-04-10T18:56:23Z

swiftlang/swift-syntax#3047
@swift-ci please test

hamishknight

I have a question about the parsing logic in expression position, but otherwise this LGTM

include/swift/AST/Attr.h

lib/Parse/ParseType.cpp

lib/Sema/TypeCheckDecl.cpp

hamishknight · 2025-04-10T20:34:03Z

lib/Parse/ParseExpr.cpp

+  // Try to parse '@' sign, 'inout' or 'nonisolated' as a attributed typerepr.
+  if (Tok.isAny(tok::at_sign, tok::kw_inout) ||
+      Tok.isContextualKeyword("nonisolated")) {


Hmm, does this mean the following tries to parse as a nonisolated type?

func nonisolated(_ x: Int) {} nonisolated(0) print("hello")

I noticed the swift-syntax PR doesn't include similar logic, should this just be removed?

If I remove this condition C++ parser cannot parse - let _ = [nonisolated(nonsending) () async -> Void](), maybe there is a better way to do this I don't know about...

I tightened the condition in canParseNonisolatedAsTypeModifier and added your snippet. Rintaro pointed me to this spot initially for [nonisolated(nonsending) ...]() support, but if there is a better way please let me know.

I see! In that case, yes this is the correct place for the logic, I think you'll need similar logic in swift-syntax though, I suspect that will currently be parsing nonisolated(nonsending) as a function call in that position (the way we model 'async'? 'throws'? '->' in the parser is as a binary operator, and nonisolated(nonsending) () is a valid expression). I think it would be worth adding a couple more ASTGen tests to verify that's behaving correctly.

I tightened the condition in canParseNonisolatedAsTypeModifier and added your snippet

Thanks! In principle it's still source breaking if someone were passing an argument named nonsending, but I doubt that'll ever come up in practice :)

xedin · 2025-04-11T18:53:55Z

Curious: do we not need feature suppression for this new pair of attributes?

We don't need a feature for this since the proposal was been accepted, that's why I've removed ExecutionAttribute feature in this PR.

…ute` or `StringRef` It has been decided to split the attribute into `@concurrent` and `nonisolated(nonsending`. Adjusting diagnostics to accept the attribute makes the transition easier.

…(concurrent)` spelling

…dable`

AnthonyLatsis · 2025-04-11T19:20:48Z

We don't need a feature for this since the proposal was been accepted

I thought the practice was to ship syntactic features as suppressible for a while for short-term textual interface compatibility.

xedin · 2025-04-11T20:33:59Z

I don't think we make these backward compatibility guarantees (anymore?) based on conversation with @tshortli about -interface-compiler-version.

xedin · 2025-04-11T20:35:36Z

These new/old attribute and modifier are not going to be present in old interface files unconditionally anyway.

tshortli · 2025-04-11T20:35:44Z

No, Anthony is right. There is a trailing window of previous development versions of the compiler that textual interfaces must remain compatible with.

…execution(caller)`

Complete the transition from `@execution` to `@concurrent` and `nonisolated(nonsending)`

SE-0461 has been accepted and `@concurrent` and `nonisolated(nonsending)` can be make generally available now.

Accept it only if it's spelled `nonisolated(nonsending)` and nothing else to minimize any possible source compatibility impact.

…ure to match `@execution` attribute split

…ed(nonsending)` in swift interface files

xedin · 2025-04-11T23:48:22Z

For posterity: we discussed this offline and yes, we do need a feature flag here to indicate that new attributes are supported only by new compilers even though we have been inconsistent about this in the past.

xedin · 2025-04-11T23:49:49Z

swiftlang/swift-syntax#3047
@swift-ci please test

xedin · 2025-04-12T00:07:08Z

@swift-ci please test source compatibility

xedin · 2025-04-12T02:13:25Z

swiftlang/swift-syntax#3047
@swift-ci please test source compatibility

xedin · 2025-04-12T08:07:10Z

swiftlang/swift-syntax#3047
@swift-ci please test macOS platform

xedin requested review from xymus, jckarter, ahoppen, bnbarham, CodaFi, DougGregor, hamishknight, rintaro, ktoso, rjmccall, AnthonyLatsis, hborla, slavapestov, nkcsgexi and tshortli as code owners April 9, 2025 07:41

xedin mentioned this pull request Apr 9, 2025

[SwiftParser] Implement nonisolated(nonsending) specifier swiftlang/swift-syntax#3047

Merged

xedin force-pushed the se-0461-renamings branch from 22a69b6 to 9bde623 Compare April 9, 2025 21:26

hamishknight approved these changes Apr 10, 2025

View reviewed changes

xedin changed the title ~~[SE-0461] Replace @execution(...) with @concurrent and `nonisolated(nonsending)~~ [SE-0461] Replace @execution(...) with @concurrent and nonisolated(nonsending) Apr 10, 2025

xedin force-pushed the se-0461-renamings branch from 9bde623 to c61382f Compare April 11, 2025 00:51

xedin requested review from zoecarver, egorzhdan, beccadax, ian-twilightcoder, Xazax-hun, j-hui, fahadnayyar and susmonteiro as code owners April 11, 2025 00:51

xedin added 5 commits April 11, 2025 12:08

[Diagnostics] Adjust @execution(...) diagnostic to take `DeclAttrib…

1a3af5c

…ute` or `StringRef` It has been decided to split the attribute into `@concurrent` and `nonisolated(nonsending`. Adjusting diagnostics to accept the attribute makes the transition easier.

[AST/ASTGen] Introduce @concurrent attribute to replace `@execution…

2704ab7

…(concurrent)` spelling

[Parse] Remove warning about @concurrent being alternative to `@Sen…

e3cbdaa

…dable`

[AST/Sema] Support for @concurrent attribute in type context

318c415

[AST/Sema] Replace @execution(concurrent) with @concurrent

4b8c8e7

xedin added 9 commits April 11, 2025 15:57

[AST/Sema] Intoduce nonisolated(nonsending) as a replacement for `@…

b1ffa06

…execution(caller)`

[AST/Parse] Implement nonisolated(nonsending) type modifier

54b62ae

[AST/ASTGen/Sema/Serialization] Remove @execution attribute

07ff063

Complete the transition from `@execution` to `@concurrent` and `nonisolated(nonsending)`

[Sema/SILGen] NFC: Remove all mentions of @execution from comments

9da457d

[AST] Remove ExecutionAttribute experimental feature

f1b3c7b

SE-0461 has been accepted and `@concurrent` and `nonisolated(nonsending)` can be make generally available now.

[AST] NFC: Add a convenient way to create implicit NonisolatedAttrs

3906980

[Parse] Tighten constraints in canParseNonisolatedAsTypeModifier

f00d808

Accept it only if it's spelled `nonisolated(nonsending)` and nothing else to minimize any possible source compatibility impact.

[Diagnostics] Update educational note for AsyncCallerExecution feat…

2b4ba84

…ure to match `@execution` attribute split

[Frontend] Add a feature to guard use of @concurrent and `nonisolat…

734b1f1

…ed(nonsending)` in swift interface files

xedin force-pushed the se-0461-renamings branch from b337552 to 734b1f1 Compare April 11, 2025 23:29

tshortli approved these changes Apr 11, 2025

View reviewed changes

xedin merged commit 0caa9b5 into swiftlang:main Apr 12, 2025
7 checks passed

xedin mentioned this pull request Apr 14, 2025

[6.2][SE-0461] Replace @execution(...) with @concurrent and nonisolated(nonsending) #80807

Merged

DataCorrupted added a commit to DataCorrupted/swift that referenced this pull request Apr 15, 2025

[IDE] Also add missing tests from swiftlang#80668

a93b880

DataCorrupted mentioned this pull request Apr 15, 2025

[IDE] Fix a test in IDE #80804

Merged

[SE-0461] Replace @execution(...) with @concurrent and nonisolated(nonsending) #80668

[SE-0461] Replace @execution(...) with @concurrent and nonisolated(nonsending) #80668

Uh oh!

Conversation

xedin commented Apr 9, 2025

Uh oh!

xedin commented Apr 9, 2025

Uh oh!

xedin commented Apr 10, 2025

Uh oh!

hamishknight left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hamishknight Apr 10, 2025

Choose a reason for hiding this comment

Uh oh!

xedin Apr 10, 2025

Choose a reason for hiding this comment

Uh oh!

xedin Apr 11, 2025

Choose a reason for hiding this comment

Uh oh!

hamishknight Apr 11, 2025

Choose a reason for hiding this comment

Uh oh!

xedin commented Apr 11, 2025

Uh oh!

AnthonyLatsis commented Apr 11, 2025

Uh oh!

xedin commented Apr 11, 2025

Uh oh!

xedin commented Apr 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tshortli commented Apr 11, 2025

Uh oh!

xedin commented Apr 11, 2025

Uh oh!

xedin commented Apr 11, 2025

Uh oh!

xedin commented Apr 12, 2025

Uh oh!

xedin commented Apr 12, 2025

Uh oh!

xedin commented Apr 12, 2025

Uh oh!

Uh oh!

Uh oh!

[SE-0461] Replace `@execution(...)` with `@concurrent` and `nonisolated(nonsending)` #80668

[SE-0461] Replace `@execution(...)` with `@concurrent` and `nonisolated(nonsending)` #80668

xedin commented Apr 11, 2025 •

edited

Loading