By default (currently) the closure passed to a parameter with `@_inheritActorContext`
would only inherit isolation from `nonisolated`, global actor isolated or actor
context when "self" is captured by the closure. `always` changes this behavior to
always inherit actor isolation from context regardless of whether it's captured
or not.
We've been converging the implementations of educational notes and
diagnostic groups, where both provide category information in
diagnostics (e.g., `[#StrictMemorySafety]`) and corresponding
short-form documentation files. The diagnostic group model is more
useful in a few ways:
* It provides warnings-as-errors control for warnings in the group
* It is easier to associate a diagnostic with a group with
GROUPED_ERROR/GROUPED_WARNING than it is to have a separate diagnostic
ID -> mapping.
* It is easier to see our progress on diagnostic-group coverage
* It provides an easy name to use for diagnostic purposes.
Collapse the educational-notes infrastructure into diagnostic groups,
migrating all of the existing educational notes into new groups.
Simplify the code paths that dealt with multiple educational notes to
have a single, possibly-missing "category documentation URL", which is
how we're treating this.
Attempting to bypass the compiler and access runtime functions directly has
a long history of breaking in hard-to-predict ways, and there's usually a better
way. Put up a warning to try to flush out misuses of runtime functions to see
if we can turn this into an error.
Consider the following piece of code and what the isolation is of the closure
literal passed to doSomething():
```swift
func doSomething(_ f: sending () -> ()) { ... }
@MyCustomActor
func foo() async {
doSomething {
// What is the isolation here?
}
}
```
In this case, the isolation of the closure is @MyCustomActor. This is because
non-Sendable closures are by default isolated to their current context (in this
case @MyCustomActor since foo is @MyCustomActor isolated). This is a problem
since
1. Our closure is a synchronous function that does not have the ability to hop
to MyCustomActor to run said code. This could result in a concurrency hole
caused by running the closure in doSomething() without hopping to
MyCustomActor's executor.
2. In Region Based Isolation, a closure that is actor isolated cannot be sent,
so we would immediately hit a region isolation error.
To fix this issue, by default, if a closure literal is passed as a sending
parameter, we make its isolation nonisolated. This ensures that it is
disconnected and can be transferred safely.
In the case of an async closure literal, we follow the same semantics, but we
add an additional wrinkle: we keep support of inheritActorIsolation. If one
marks an async closure literal with inheritActorIsolation, we allow for it to be
passed as a sendable parameter since it is actually Sendable under the hood.
This attribute instructs the compiler that this function declaration
should be "import"ed from host environment. It's equivalent of Clang's
`__attribute__((import_module("module"), import_name("field")))`
This attribute instructs the compiler that this function declaration
should be "export"ed from this .wasm module. It's equivalent of Clang's
`__attribute__((export_name("name")))`
Make it clearer the statement about borrowed `self` and other parameters having
stable addresses for the duration of a method only holds for `@_rawLayout` types.
This attribute can be attached to a noncopyable struct to specify that its
storage is raw, meaning the type definition is (with some limitations)
able to do as it pleases with the storage. This provides a basis for
implementing types for things like atomics, locks, and data structures
that use inline storage to store conditionally-initialized values.
The example in `test/Prototypes/UnfairLock.swift` demonstrates the use
of a raw layout type to wrap Darwin's `os_unfair_lock` APIs, allowing
a lock value to be stored inside of classes or other types without
needing a separate allocation, and using the borrow model to enforce
safe access to lock-guarded storage.
Attribute @_silgen_name is today only allowed to be used on functions, this change allows usage on globals as well. The motivation for that is to be able to "forward declare" globals just like it's today possible to do with functions (for the cases where it's not practical or convenient to use a bridging header).
Separately, this change also adds a @_silgen_name(raw: ...) syntax, which simply avoids mangling the name (by using the \01 name prefix that LLVM uses). The motivation for that is to be able to reference the "magic Darwin linker symbols" that can be used to look up section bounds (in the current dylib/module) -- those symbols don't use the underscore prefix in their mangled names.
* Add @_used and @_section attributes for global variables and top-level functions
This adds:
- @_used attribute that flags as a global variable or a top-level function as
"do not dead-strip" via llvm.used, roughly the equivalent of
__attribute__((used)) in C/C++.
- @_section("...") attribute that places a global variable or a top-level
function into a section with that name, roughly the equivalent of
__attribute__((section("..."))) in C/C++.
This attribute will, in the near future, be used to inform IRGen that a nominal type that conforms to such protocol must have its type metadata always emitted into the binary, regardless of whether it is used/public.
* [Distributed] dist actor always has default executor (currently)
* [Distributed] extra test for missing makeEncoder
* [DistributedDecl] Add DistributedActorSystem to known SDK types
* [DistributedActor] ok progress on getting the system via witness
* [Distributed] allow hop-to `let any: any X` where X is DistActor
* [Distributed] AST: Add an accessor to determine whether type is distributed actor
- Classes have specialized method on their declarations
- Archetypes and existentials check their conformances for
presence of `DistributedActor` protocol.
* [Distributed] AST: Account for distributed members declared in class extensions
`getConcreteReplacementForProtocolActorSystemType` should use `getSelfClassDecl`
otherwise it wouldn't be able to find actor if the member is declared in an extension.
* [Distributed] fix ad-hoc requirement checks for 'mutating'
[PreChecker] LookupDC might be null, so account for that
* [Distributed] Completed AST synthesis for dist thunk
* [Distributed][ASTDumper] print pretty distributed in right color in AST dumps
* wip on making the local/remote calls
* using the _local to mark the localCall as known local
* [Distributed] fix passing Never when not throwing
* fix lifetime of mangled string
* [Distributed] Implement recordGenericSubstitution
* [Distributed] Dont add .
* [Distributed] dont emit thunk when func broken
* [Distributed] fix tests; cleanups
* [Distributed] cleanup, move is... funcs to DistributedDecl
* [Distributed] Remove SILGen for distributed thunks, it is in Sema now!
* [Distributed] no need to check stored props in protocols
* remote not used flag
* fix mangling test
* [Distributed] Synthesis: Don't re-use AST nodes for `decodeArgument` references
* [Distributed] Synthesis: Make sure that each thunk parameter has an internal name
* [Distributed/Synthesis] NFC: Add a comment regarding empty internal parameter names
* [Distributed] NFC: Adjust distributed thunk manglings in the accessor section test-cases
* cleanup
* [Distributed] NFC: Adjust distributed thunk manglings in the accessor thunk test-cases
* review follow ups
* xfail some linux tests for now so we can land the AST thunk
* Update distributed_actor_remote_functions.swift
Co-authored-by: Pavel Yaskevich <xedin@apache.org>
SE-0338 changed the execution of non-actor async functions
so that they always hop to the generic executor, but some
functions need a way to suppress this so that they inherit
the caller's executor.
The right way to implement this is to have the caller pass
down the target executor in some reliable way and then
switch to it in all the appropriate places in the caller.
We might reasonably be able to build this on top of isolated
parameters, using some sort of default argument, or we might
need a wholly novel mechanism.
But those things are all ABI-breaking absent some sort of
guarantee about switching that we probably don't want to make,
and unfortunately we have functions in the library which we
need to export that need to inherit executors. So in the
short term, we need some unsafe way of getting back to the
previous behavior.
Store a list of argument effects in a function, which specify if and how arguments escape.
Such effects can be specified in the Swift source code (for details see docs/ReferenceGuides/UnderscoredAttributes.md) or derived in an optimization pass.
For details see the documentation in SwiftCompilerSources/Sources/SIL/Effects.swift.
The core part of the check runs as follows
1. Are we in an async context
2. Does the decl we're calling have the unavailableFromAsync attribute
3. Emit a diagnostic
There are a couple challenges that muddy the implementation. First,
single-expression closures are typechecked differently than
multi-expression closures. The single-expression closure is never added
to the closure stack. We have to record it separately upon entry to
verify separately. This is necessary for `_ = { foo() }()` where `foo`
is unavailable from async, and the surrounding context is async.
The second challenge is regarding when things get typechecked. A type
must be assigned to AbstractClosureExprs to determine whether they are
an async context or not. Unfortunately, availability checking runs
before types are fully assigned in some cases. This specifically happens
when working with result builders in specific conditions. I haven't been
able to figure out how to reduce the issue further.
