`getLoweredLocalCaptures` seems the best place to do it because
during Sema there is a chicken and an egg situation where isolation
depends on captures and in this case captures depend on isolation.
Also, make it more tolerant to instructions and builtins, which are not explicitly handled.
This avoids crashes when new instructions are added. We got lucky that this didn't happen so far.
It derives the address of the first element of a vector, i.e. a `Builtin.FixedArray`, from the address of the vector itself.
Addresses of other vector elements can then be derived with `index_addr`.
Some notes:
1. In most cases, I think we were getting lucky with this by just inferring the
closure's isolation from its decl context. In the specific case that we were
looking at here, this was not true since we are returning from an @concurrent
async function a nonisolated(nonsending) method that closes over self. This
occurs since even when NonisolatedNonsendingByDefault we want to start importing
objc async functions as nonisolated(nonsending).
2. I also discovered that in the ActorIsolationChecker we were not visiting the
inner autoclosure meaning that we never set the ActorIsolation field on the
closure. After some discussion with @xedin about potentially visiting the
function in the ActorIsolationChecker, we came to the conclusion that this was
likely to result in source stability changes. So we put in a targeted fix just
for autoclosures in this specific case by setting their actor isolation in the
type checker.
3. Beyond adding tests to objc_async_from_swift to make sure that when
NonisolatedNonsendingByDefault is disabled we do the right thing, I noticed that
we did not have any tests that actually tested the behavior around
objc_async_from_swift when NonisolatedNonsendingByDefault is enabled. So I added
the relevant test lines so we can be sure that we get correct behavior in such a
case.
rdar://150209093
In Swift 6.1, we introduced `SWIFT_RETURNS_RETAINED` and
`SWIFT_RETURNS_UNRETAINED` annotations for C++ APIs to explicitly
specify the ownership convention of `SWIFT_SHARED_REFERENCE` type return
values.
Currently the Swift compiler emits warnings for unannotated C++ APIs
returning `SWIFT_SHARED_REFERENCE` types. We've received some feedback
that people are finding these warnings useful to get a reminder to
annotate their APIs. While this improves correctness , it also imposes a
high annotation burden on adopters — especially in large C++ codebases.
This patch addresses that burden by introducing two new type-level
annotations:
- `SWIFT_RETURNED_AS_RETAINED_BY_DEFAULT`
- `SWIFT_RETURNED_AS_UNRETAINED_BY_DEFAULT`
These annotations allow developers to specify a default ownership
convention for all C++ APIs returning a given
`SWIFT_SHARED_REFERENCE`-annotated type, unless explicitly overridden at
the API by using `SWIFT_RETURNS_RETAINED` or `SWIFT_RETURNS_UNRETAINED`.
If a C++ class inherits from a base class annotated with
`SWIFT_RETURNED_AS_RETAINED_BY_DEFAULT` or
`SWIFT_RETURNED_AS_UNRETAINED_BY_DEFAULT`, the derived class
automatically inherits the default ownership convention unless it is
explicitly overridden. This strikes a balance between safety/correctness
and usability:
- It avoids the need to annotate every API individually.
- It retains the ability to opt out of the default at the API level when
needed.
- To verify correctness, the user can just remove the
`SWIFT_RETURNED_AS_(UN)RETAINED_BY_DEFAULT` annotation from that type
and they will start seeing the warnings on all the unannotated C++ APIs
returning that `SWIFT_SHARED_REFERENCE` type. They can add
`SWIFT_RETURNS_(UN)RETAINED` annotation at each API in which they want a
different behaviour than the default. Then they can reintroduce the
`SWIFT_RETURNED_AS_(UN)RETAINED_BY_DEFAULT` at the type level to
suppress the warnings on remaining unannotated APIs.
A global default ownership convention (like always return
`unretained`/`unowned`) was considered but it would weaken the
diagnostic signal and remove valuable guardrails that help detect
use-after-free bugs and memory leaks in absence of
`SWIFT_RETURNS_(UN)RETAINED` annotations. In the absence of these
annotations when Swift emits the unannotated API warning, the current
fallback behavior (e.g. relying on heuristics based on API name such as
`"create"`, `"copy"`, `"get"`) is derived from Objective-C interop but
is ill-suited for C++, which has no consistent naming patterns for
ownership semantics.
Several codebases are expected to have project-specific conventions,
such as defaulting to unretained except for factory methods and
constructors. A type-level default seems like the most precise and
scalable mechanism to support such patterns. It integrates cleanly with
existing `SWIFT_SHARED_REFERENCE` usage and provides a per-type opt-in
mechanism without global silencing of ownership diagnostics.
This addition improves ergonomics while preserving the safety benefits
of explicit annotations and diagnostics.
rdar://145453509
Similarly to how https://github.com/swiftlang/swift/pull/70564 configures 'ClangImporter's 'CodeGenerator' using Swift's compilation target triple, we must use the versioned version of the 'isWeakImported' query to determine linkage for imported Clang symbols.
Instead of passing in the substituted type, we pass in the
InFlightSubstitution. This allows the substituted type to be
recovered if needed, but we can now skip computing it for
the common case of LookUpConformanceInSubstitutionMap.
In case of ObjectiveC classes, the runtime type can differ from its declared type.
Therefore a cast between (compile-time) unrelated classes may succeed at runtime.
rdar://149810124
This replaces the oddly-named mapIntoTypeExpansionContext() method
on SubstitutionMap itself in favor of a global function, just like
the ones that take Type and ProtocolConformanceRef.
I made a mistake in 47156e006b. There was a
call to call to forAbstract() in SILTypeSubstituter that passed in the
wrong subject ype.
This call was inside SILTypeSubstituter's own implementation of replacing
opaque types with underlying types in a substitution map. This duplicates
an existing utility method in SubstitutionMap anyway, so let's just use
that instead.
Fixes rdar://149353285.
We should pass foreign references as Direct_Unowned calling convention,
this happened for non-const pointers. Unfortunately, we passed the const
variants as Indirect_In. This resulted in SILGen trying to insert a
non-sensical conversion. This triggered an assertion failure or a
miscompilation depending on how the compiler was compiled.
rdar://149398905
Inlinability doesn’t affect the mangling except in function specializations, which are applied after the fact and should never mangle in information from an ABI-only decl. That means we can simply ban these from `@abi` instead of inferring them.
Also adds some assertions to help double-check that SIL never tries to directly mangle or retrieve inlinability info from an ABI-only decl.
If the method is a default witness methods (`selfType` != nil) it has generic self type.
In this case the generic self parameter is at depth 0 and the actual generic parameters of the substitution map are at depth + 1, e.g:
```
@convention(witness_method: P) <τ_0_0><τ_1_0 where τ_0_0 : GenClass<τ_1_0>.T>
^ ^
self params of substitution map at depth + 1
```
Store specialize witness tables in a separate lookup table in the module. This allows that for a normal conformance there can exist the original _and_ a specialized witness table.
Also, add a boolean property `isSpecialized` to `WitnessTable` which indicates whether the witness table is specialized or not.
These functions already have special code generation that keeps them
in the caller's isolation context, so there is no behavior change here.
Resolves: rdar://145672343
This corrects how we were dealing with dispatch thunks -- mostly be
removing a lot of special casing we did but doesn't seem necessary and
instead we correct and emit all the necessary information int TBD.
This builds on https://github.com/swiftlang/swift/pull/74935 by further refining how we fixed that issue, and adds more regression tests. It also removes a load of special casing of distributed thunks in library evolution mode, which is great.
Resolves and adds regression test for for rdar://145292018
This is also a more proper fix to the previously resolved but in a not-great-way which caused other issues:
- resolves rdar://128284016
- resolves rdar://128310903
A trivial store is allowed to occur on an existing live value, and should not
trigger an attempt to destroy the original value completely. Fixes rdar://147791932.
Simply omit the 'nocapture' attribute on the parameter.
Fixes rdar://148039510 ([nonescapable] IRGen: lower addressable
params to LLVM: captures(ret: address, provenance))
