Implement experimental @abi attribute #76878

beccadax · 2024-10-04T22:02:22Z

This is the beginning of an implementation of the @abi attribute as recently pitched. It is intended to subsume the use of @_silgen_name in ABI compatibility hacks, providing a tool for handling API replacements and enhancements that is both more capable and usable by mere mortals.

This PR also replaces all standard library uses of @_silgen_name with a mangled Swift name with @abi uses. This is about a quarter of all standard library @_silgen_name uses (the remaining uses are to import or export runtime functions or compiler intrinsics by simple known names). (Note: Standard library adoption will be deferred until the type checking rules are more stable.)

There's a lot left to do here, ~~but one of the most important things is making @abi on functions suppressible by providing a fallback to @_silgen_name in the module interface~~.

SwiftSyntax PR: swiftlang/swift-syntax#2882

ktoso · 2024-10-05T05:54:12Z

stdlib/public/Distributed/DistributedActor.swift

  // and even though this is @AEIC, the symbol name would conflict.
-  @_silgen_name("$s11Distributed0A5ActorPAAE20whenLocalTypedThrowsyqd__Sgqd__xYiYaYbqd_0_YKXEYaqd_0_YKs8SendableRd__s5ErrorRd_0_r0_lF")
+  @abi(
+    public nonisolated func __api_whenLocal<T: Sendable, E>(


Probably this would be nicer if it stayed using the explicit "explaining why this method is different" name like we did in the silgen name:

Suggested change

public nonisolated func __api_whenLocal<T: Sendable, E>(

public nonisolated func whenLocalTypedThrows<T: Sendable, E>(

There's actually three different variants. Here's what I think I'll do here:

@abi( // We need to @abi here because the signature is the same as // `__separately_compiled_typed_throws_whenLocal(_:)`, and even though this // is @AEIC, the symbol name would conflict. public nonisolated func __typed_throws_whenLocal<T: Sendable, E>( _ body: @Sendable (isolated Self) async throws(E) -> T ) async throws(E) -> T? ) @_alwaysEmitIntoClient public nonisolated func whenLocal<T: Sendable, E>( _ body: @Sendable (isolated Self) async throws(E) -> T ) async throws(E) -> T? { /* snip */ } // ABI: This is a workaround when in Swift 6 this method was introduced // in order to support typed-throws, but missed to add @_aeic. // In practice, this method should not ever be used by anyone, ever. @abi( public nonisolated func whenLocal<T: Sendable, E>( _ body: @Sendable (isolated Self) async throws(E) -> T ) async throws(E) -> T? ) @usableFromInline nonisolated func __separately_compiled_typed_throws_whenLocal<T: Sendable, E>( _ body: @Sendable (isolated Self) async throws(E) -> T ) async throws(E) -> T? { /* snip */ } // ABI: Historical whenLocal, rethrows was changed to typed throws `throws(E)` @abi( public nonisolated func whenLocal<T: Sendable>( _ body: @Sendable (isolated Self) async throws -> T ) async rethrows -> T? ) @usableFromInline nonisolated func __rethrows_whenLocal<T: Sendable>( _ body: @Sendable (isolated Self) async throws -> T ) async rethrows -> T? { /* snip */ }

This looks good, thank you @beccadax! I'm really enjoying this way to spell these

lorentey

Thank you for working on this -- the core stdlib parts look good to me. (Great, in fact!)

lorentey · 2024-10-07T21:25:00Z

stdlib/public/core/Optional.swift

 @_spi(SwiftStdlibLegacyABI) @available(swift, obsoleted: 1)
 @usableFromInline
-internal func ?? <T>(
+internal func _legacy_abi_optionalNilCoalescingOperator <T>(


Ooh, nice catch

beccadax · 2024-10-09T21:08:26Z

@swift-ci please test

beccadax · 2024-10-09T21:12:44Z

I think the implementation here is getting towards something mergeable, if not fully usable, but I expect the feature to continue to evolve, including in some ways that will break source compatibility with the current version. (For example, I may ban access control keywords in an @abi attribute and instead inherit them from the attribute it's attached to.) For that reason, I'm planning to move the standard library adoption into its own PR before this comes out of draft and gets merged. The stdlib PR will continue to evolve as the compiler implementation improves.

beccadax · 2024-10-15T00:21:26Z

The macOS test failure, at least, was caused by SwiftSyntax not being able to parse the new syntax, so I'm adding support for that.

Matches work in swiftlang/swift#76878.

beccadax · 2024-10-16T00:49:54Z

With swiftlang/swift-syntax#2882

@swift-ci please test

Matches work in swiftlang/swift#76878.

beccadax · 2024-10-29T18:48:05Z

With swiftlang/swift-syntax#2882

@swift-ci please test

bnbarham · 2024-10-31T23:38:42Z

test/IDE/complete_decl_attribute_feature_requirement.swift

@@ -0,0 +1,154 @@
+// RUN: %target-swift-ide-test -code-completion -source-filename %s -code-completion-token=KEYWORD2 | %FileCheck %s -check-prefixes=KEYWORD2,KEYWORD2_DISABLED


Note that you can use %batch-code-completion here instead. It has the benefit of not needing the many RUN lines + can re-use the ast context. Thanks for adding these by the way, it's very much appreciated!

I had to modify the tool with a feature that could add suffixes to the check prefixes, but this definitely cleaned things up quite a bit!

Oh nice, thank you! CC @rintaro

Matches work in swiftlang/swift#76878.

beccadax · 2024-12-06T01:11:48Z

@swift-ci please test

rintaro

Parser and ASTGen related changes LGTM!

lib/AST/Decl.cpp

lib/ASTGen/Sources/ASTGen/DeclAttrs.swift

beccadax · 2024-12-13T02:25:38Z

@swift-ci test

lib/Serialization/Deserialization.cpp

beccadax · 2024-12-13T23:09:21Z

lib/Sema/TypeCheckAttr.cpp

+
+public:
+  void visitABIAttr(ABIAttr *attr) {
+    Decl *AD = attr->abiDecl;


Reviewers: In this PR, I'm mostly just trying to make sure I can find the right decls to check against one another. The exact rules I'm enforcing are extremely rough and incomplete placeholders; properly defining them will be the subject of my next PR on this feature.

xedin

Sema changes look good to me!

xedin · 2024-12-18T23:44:21Z

include/swift/AST/ASTContext.h

+  /// `ABIAttr::abiDecl` to the declaration the `ABIAttr` is attached to.
+  ///
+  /// \seeAlso \c recordABIAttr()
+  llvm::DenseMap<Decl *, Decl *> ABIDeclCounterparts;


Have you considered turning this into a cached request instead?

This would be difficult to compute lazily because there's no way (other than this table) to walk up from an ABI-only declaration to its counterpart; you have to start from the counterpart and then walk down to the ABI-only decl. The most reliable way I've found to do that is to eagerly populate a table while parsing or deserializing. I suppose I could still create a dummy request whose evaluate() function returns the default output and then make recordABIAttr() manually cache non-default outputs, but I'm not sure if that would really buy us anything over what I'm doing here.

I am suggesting it because we have a similar pattern with other requests that are cached manually after parse and deserialization, the primary advantage is use a common mechanism and storage.

Changed in 413c673.

Changes dump output used in one test; otherwise NFC.

The `@differentiable` and `@derivative` attributes need a parent pointer. Move the code to populate it from Parser to AST so it can be more easily shared between the parsers. Done in preparation for similar code to be added for `@abi`.

This attribute will allow you to specify an alternate version of the declaration used for mangling. It will allow minor adjustments to be made to declarations so long as they’re still compatible at the calling convention level, such as refining isolation or sendability, renaming without breaking ABI, etc. The attribute is behind the experimental feature flag `ABIAttribute`.

@abi

The new `DECL_ATTR_FEATURE_REQUIREMENT` macro in DeclAttr.def can be used to declare that an attribute should only be available when a related language feature is enabled. Effects: • `#if hasAttribute(someAttr)` will return `false` unless the required feature is enabled. • Code completion will not include the attribute unless the required feature is enabled. • `TypeChecker::checkDeclAttributes()` diagnoses non-implicit uses of the attribute. Add this mechanism and use it to tie @abi to the ABIAttribute feature. Also design tests for it.

@abi

Sema now type-checks the alternate ABI-providing decls inside of @abi attributes. Making this work—particularly, making redeclaration checking work—required making name lookup aware of ABI decls. Name lookup now evaluates both API-providing and ABI-providing declarations. In most cases, it will filter ABI-only decls out unless a specific flag is passed, in which case it will filter API-only decls out instead. Calls that simply retrieve a list of declarations, like `IterableDeclContext::getMembers()` and friends, typically only return API-providing decls; you have to access the ABI-providing ones through those. As part of that work, I have also added some basic compiler interfaces for working with the API-providing and ABI-providing variants. `ABIRole` encodes whether a declaration provides only API, only ABI, or both, and `ABIRoleInfo` combines that with a pointer to the counterpart providing the other role (for a declaration that provides both, that’ll just be a pointer to `this`). Decl checking of behavior specific to @abi will come in a future commit. Note that this probably doesn’t properly exercise some of the new code (ASTScope::lookupEnclosingABIAttributeScope(), for instance); I expect that to happen only once we can rename types using an @abi attribute, since that will create distinguishable behavior differences when resolving TypeReprs in other @abi attributes.

Check for: • Matching decl kinds • Matching PBD shapes (does every VarDecl on both sides have a counterpart?) • Matching function effects • Matching function arity (roughly)

@abi

What’s implemented now is actually *far* more thorough than what the surface syntax can currently express, mainly because I can’t apply @abi to nominal types yet.

Use `Decl::attachParsedAttrs()` instead of `Decl::setAttrs()` to attach attributes to a declaration in ASTGen. This causes the common attribute-setup logic there to be run. NFC in this commit because none of the attributes that have special setup logic are currently implemented in ASTGen. Prepares to add support for `@abi` in a future commit.

beccadax · 2024-12-20T05:23:34Z

@swift-ci please test

ktoso reviewed Oct 5, 2024

View reviewed changes

lorentey approved these changes Oct 7, 2024

View reviewed changes

beccadax force-pushed the abi-changed-your-name branch 4 times, most recently from f3a443a to 6c8eebd Compare October 9, 2024 21:03

nkcsgexi self-requested a review October 9, 2024 22:57

beccadax force-pushed the abi-changed-your-name branch 2 times, most recently from 7b715b7 to 446ecbc Compare October 16, 2024 00:38

beccadax added a commit to beccadax/swift-syntax that referenced this pull request Oct 16, 2024

Parse the @abi attribute in SwiftSyntax

de2dc9f

Matches work in swiftlang/swift#76878.

beccadax mentioned this pull request Oct 16, 2024

SwiftSyntax support for experimental @abi attribute swiftlang/swift-syntax#2882

Open

beccadax added a commit to beccadax/swift-syntax that referenced this pull request Oct 28, 2024

Parse the @abi attribute in SwiftSyntax

f7dfc81

Matches work in swiftlang/swift#76878.

beccadax force-pushed the abi-changed-your-name branch 2 times, most recently from d332e60 to a007f36 Compare October 29, 2024 18:47

bnbarham reviewed Oct 31, 2024

View reviewed changes

beccadax added a commit to beccadax/swift-syntax that referenced this pull request Nov 5, 2024

Parse the @abi attribute in SwiftSyntax

54ce286

Matches work in swiftlang/swift#76878.

beccadax force-pushed the abi-changed-your-name branch 2 times, most recently from 99ddda0 to 7a96bdd Compare November 7, 2024 20:32

beccadax mentioned this pull request Nov 21, 2024

Partial SwiftSyntax support for experimental @abi attribute swiftlang/swift-syntax#2902

Merged

beccadax force-pushed the abi-changed-your-name branch 2 times, most recently from 7fd234b to c2cd7b7 Compare December 6, 2024 01:11

beccadax marked this pull request as ready for review December 6, 2024 01:12

beccadax requested review from ahoppen and hamishknight as code owners December 6, 2024 01:12

beccadax force-pushed the abi-changed-your-name branch from 4c63dd8 to 6c8adfb Compare December 9, 2024 20:07

rintaro reviewed Dec 9, 2024

View reviewed changes

lib/AST/Decl.cpp Outdated Show resolved Hide resolved

rintaro reviewed Dec 10, 2024

View reviewed changes

lib/ASTGen/Sources/ASTGen/DeclAttrs.swift Outdated Show resolved Hide resolved

beccadax force-pushed the abi-changed-your-name branch 2 times, most recently from 1293a50 to 3483a6c Compare December 13, 2024 02:20

beccadax force-pushed the abi-changed-your-name branch 2 times, most recently from da21faf to 1fd92db Compare December 13, 2024 20:58

xymus reviewed Dec 13, 2024

View reviewed changes

lib/Serialization/Deserialization.cpp Outdated Show resolved Hide resolved

beccadax force-pushed the abi-changed-your-name branch from 1fd92db to 287e0bc Compare December 13, 2024 22:23

beccadax commented Dec 13, 2024

View reviewed changes

xedin approved these changes Dec 18, 2024

View reviewed changes

beccadax added 15 commits December 19, 2024 15:47

[ClangImporter] Try to warn about missing types

2d0b54b

[NFC] Dump an ASTScopeImpl’s parent nodes

9f0130b

[NFC] Improve lookup flag printing facilities

9b947f9

[NFC-ish] Fix some crashes from early Decl::dump()

73377fe

Changes dump output used in one test; otherwise NFC.

[NFC] Extract autodiff parsing code to Decl.cpp

c9f539e

The `@differentiable` and `@derivative` attributes need a parent pointer. Move the code to populate it from Parser to AST so it can be more easily shared between the parsers. Done in preparation for similar code to be added for `@abi`.

Add some basic validation of vars and funcs

c91cf69

Check for: • Matching decl kinds • Matching PBD shapes (does every VarDecl on both sides have a counterpart?) • Matching function effects • Matching function arity (roughly)

Use @abi attribute in mangling

0f3cf62

What’s implemented now is actually *far* more thorough than what the surface syntax can currently express, mainly because I can’t apply @abi to nominal types yet.

Print @abi into module interface (suppressibly)

0d34b4d

Diagnose use of @_silgen_name and @abi together

443bd5d

[ASTGen] Lower @abi in ASTGen

9c1e00c

[NFC] Requestify ABI role computation

413c673

beccadax force-pushed the abi-changed-your-name branch from 287e0bc to 413c673 Compare December 20, 2024 05:22

beccadax merged commit a6df4ef into swiftlang:main Dec 20, 2024
5 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement experimental @abi attribute #76878

Implement experimental @abi attribute #76878

beccadax commented Oct 4, 2024 •

edited

Loading

ktoso Oct 5, 2024

beccadax Oct 8, 2024 •

edited

Loading

ktoso Oct 9, 2024

lorentey left a comment

lorentey Oct 7, 2024

beccadax commented Oct 9, 2024

beccadax commented Oct 9, 2024

beccadax commented Oct 15, 2024

beccadax commented Oct 16, 2024

beccadax commented Oct 29, 2024

bnbarham Oct 31, 2024

beccadax Dec 6, 2024

bnbarham Dec 6, 2024

beccadax commented Dec 6, 2024

rintaro left a comment •

edited

Loading

beccadax commented Dec 13, 2024

beccadax Dec 13, 2024

xedin left a comment

xedin Dec 18, 2024

beccadax Dec 19, 2024

xedin Dec 19, 2024

beccadax Dec 20, 2024

beccadax commented Dec 20, 2024

	public nonisolated func __api_whenLocal<T: Sendable, E>(
	public nonisolated func whenLocalTypedThrows<T: Sendable, E>(

		@@ -0,0 +1,154 @@
		// RUN: %target-swift-ide-test -code-completion -source-filename %s -code-completion-token=KEYWORD2 \| %FileCheck %s -check-prefixes=KEYWORD2,KEYWORD2_DISABLED

Implement experimental @abi attribute #76878

Implement experimental @abi attribute #76878

Conversation

beccadax commented Oct 4, 2024 • edited Loading

Choose a reason for hiding this comment

beccadax Oct 8, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lorentey left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

beccadax commented Oct 9, 2024

beccadax commented Oct 9, 2024

beccadax commented Oct 15, 2024

beccadax commented Oct 16, 2024

beccadax commented Oct 29, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

beccadax commented Dec 6, 2024

rintaro left a comment • edited Loading

Choose a reason for hiding this comment

beccadax commented Dec 13, 2024

Choose a reason for hiding this comment

xedin left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

beccadax commented Dec 20, 2024

beccadax commented Oct 4, 2024 •

edited

Loading

beccadax Oct 8, 2024 •

edited

Loading

rintaro left a comment •

edited

Loading