Interop is injecting escapability annotations for the STL and doing a
limited inference for aggregates. Let's reuse the same facilities in the
AST when we calculate the safety of the foreign types.
SafeInterop was guarding whether we import certain foreign types as
unsafe. Since these attrbutes are only considered when an opt-in strict
language mode is on, this PR removes this feature flag. We still rely on
the presence of the AllowUnsafeAttribute flag to add the unsafe
attributes to the imported types and functions.
Map the lifetime dependencies described in terms of the formal AST-level parameters
to the correct parameter(s) in the lowered SIL function type. There can be 0, 1,
or many SIL parameters per formal parameter because of tuple exploding. Also,
record which dependencies are on addressable parameters (meaning that the dependency
includes not only the value of the parameter, but its specific memory location).
Repeatedly lookup up a key from a dictionary can be justified whenever
the content of the dictionary might change between the lookups (so any
references into the dictionary might get invalidated). We had a couple
of instances where as far as I can tell no such modifications should
happen between two lookups with identical keys. This PR simplifies the
code to remove the extra lookups. It also removes a dictionary that was
completely unused.
…at least in the specific case of initializers.
Normally, clang validates that a SwiftNameAttr’s specified name seems to have the right number of argument labels for the declaration it’s applied to; if it doesn’t, it diagnoses and drops the attribute. However, this checking isn’t perfect because clang doesn’t know all of Clang Importer’s throwing rules. The upshot is that it’s possible to get a mismatched SwiftNameAttr into Clang Importer if the last parameter looks like an out parameter. This trips an assertion in asserts compilers, but release compilers get past that and don’t seem to notice the mismatch at all.
Add code to the compiler to tolerate this condition, at least in initializers, by modifying the Swift name to have the correct number of argument labels and deprecating the declaration with a message explaining the problem.
Fixes rdar://141124373.
Initializers should always have Swift names that have the special `DeclBaseName::createConstructor()` base name. Although there is an assertion to this effect in the constructor for ConstructorDecl, ClangImporter did not actually reject custom Swift names for initializers that violated this rule. This meant that asserts compilers would crash if they encountered code with an invalid `swift_name` attribute, while release compilers would silently accept them (while excluding these decls from certain checks since lookups that were supposed to find all initializers didn’t find them).
Modify ClangImporter to diagnose this condition and ignore the custom Swift name.
C++ code can return values that depend on the storage that backs the
references that were passed in as argument. Thus, swift should not
introdue temporary copies of that storage before invoking those
functions as they could result in lifetime issues.
This commit removes the guardrails in ImportDecl.cpp:SwiftDeclConverter
that prevent it from importing non-public C++ members. It also
accordingly adjusts all code that assumes generated Swift decls should
be public. This commit does not import non-public inherited members;
that needs its own follow-up patch.
Note that Swift enforces stricter invariants about access levels than C++.
For instance, public typealiases cannot be assigned private underlying types,
and public functions cannot take or return private types. Meanwhile,
both of these patterns are supported in C++, where exposing private types
from a class's public interface is considered feature. As far as I am aware,
Swift was already importing such private-containing public decls from C++
already, but I added a test suite, access inversion, that checks and
documents this scenario, to ensure that it doesn't trip any assertions.
Support adding safe wrappers for APIs returning std::span depending on
the this object. This also fixes an issue for APIs with 0 parameters.
rdar://139074571
This passes along the noescape attribute to @_SwiftifyImport as
.noescape(pointer: .param(X)). This allows importing parameters as Span,
MutableSpan and RawSpan.
ClangImporter will generate value and default initializers for certain
structs imported from C++. These generated initializers have no
associated lifetime dependence information so they will trigger spurious
errors for non-escapable types. This patch makes sure these are marked
as unsafe so the type checker will not generate errors for them.
Moreover, the generated default initializer would trigger a crash for
non-escapable types as the builtin to zero initialize an object does not
support non-escapable types yet.
rdar://143040862
In strict safe mode we should consider all C++ APIs with non-escapable
parameters unsafe unless they have their lifetimes annotated. This can
be done using [[clang::lifetimebound]], [[clang::lifetime_capture_by]],
or [[clang::noescape]].
Consider code like:
```
// Foo.h
typealias NSString * FooKey NS_EXTENSIBLE_TYPED_ENUM;
// Foo.swift
extension FooKey { … }
```
When Swift binds the extension to `FooKey`, that forces ClangImporter to import `FooKey`. ClangImporter’s newtype logic, among other things, checks whether the underlying type (`Swift.String` here) is Objective-C bridgeable and, if so, makes `FooKey` bridgeable too.
But what happens if this code is actually *in* Foundation, which is where the `extension String: _ObjectiveCBridgeable` lives? Well, if the compiler has already bound that extension, it works fine…but if it hasn’t, `FooKey` ends up unbridgeable, which can cause both type checking failures and IRGen crashes when code tries to use its bridging capabilities. And these symptoms are sensitive to precise details of the order Swift happens to bind extensions in, so e.g. adding empty files to the project can make the bug appear or disappear. Spooky.
Add a narrow hack to ClangImporter (only active for types in Foundation) to *assume* that `String` is bridgeable even if the extension declaring this hasn’t been bound yet.
Fixes rdar://142693093.
Swift imports template specializations as a standalone type (not as an
instantiation of a generic) so unsafety is not propagated from the
template arguments to the specialization. This PR propagates this
information explicitly.
This type is triggering modularization issues in libc++:
```
error: definition of '__builtin_new_deleter' must be imported from module 'std_private_memory_builtin_new_allocator' before it is required
```
This is a workaround to keep things building.
rdar://142576799
On Apple platforms, a system module `simd` declares a `namespace simd`
under `#if defined(__cplusplus)`. This namespace defines C++ overlays of
the simd types, but these types are already refined for Swift
separately, so it's not necessary to import this namespace.
This is the same issue previously encountered for the `os` module, work
around it in the same way.
rdar://143007477
On Apple platforms, a system module `os` declares a `namespace os` under `#if defined(__cplusplus)`. This causes ClangImporter to import it as `enum os` when C++ interop is enabled. This causes name lookup ambiguity (module os vs namespace os) which is resolved in namespace's favor, breaking existing usages.
rdar://119044493
* Import __counted_by for function return values
Instead of simply passing a parameter index to _SwiftifyInfo, the
_SwiftifyExpr enum is introduced. It currently has two cases:
- .param(index: Int), corresponding to the previous parameter index
- .return, corresponding to the function's return value.
ClangImporter is also updated to pass this new information along to
_SwiftifyImport, allowing overloads with buffer pointer return types to
be generated. The swiftified return values currently return Span when
the return value is marked as nonescaping, despite this not being sound.
This is a bug that will be fixed in the next commit, as the issue is
greater than just for return values.
* Fix Span variant selection
There was an assumption that all converted pointers were either
converted to Span-family pointers, or UnsafeBufferPointer-family
pointers. This was not consistently handled, resulting in violating the
`assert(nonescaping)` assert when the two were mixed. This patch removes
the Variant struct, and instead each swiftified pointer separately
tracks whether it should map to Span or UnsafeBufferPointer.
This also fixes return pointers being incorrectly mapped to Span when
marked as nonescaping.
AvailableAttr::Kind and AvailabilityDomain are designed to replace
PlatformAgnosticAvailabilityKind, allowing AvailableAttr to more flexibly model
availability for arbitrary domains. For now, the new constructor just
translates its inputs into inputs for the existing constructor. Once all of the
callers of the existing AvailableAttr constructor have been updated to use the
new constructor, the representation of AvailableAttr will be updated to store
the new properties.
A previous PR already added support to the SwiftifyImport macro to
generate safe wrappers. This PR makes ClangImporter emit the macro to do
the transformation.
decl being accessed is correct. When this assumption fails due to a deserialization error
of its members, the use site accesses the layout with a wrong field offset, resulting in
UB or a crash. The deserialization error is currently not caught at compile time due to
LangOpts.EnableDeserializationRecovery being enabled by default to allow for recovery of some
of the deserialization errors at a later time. In case of member deserialization, however,
it's not necessarily recovered later on.
This PR tracks whether member deserialization had an error by recursively loading members and
checking for deserialization error, and fails and emits a diagnostic. It provides a way to
prevent resilience bypassing when the deserialized decl's layout is incorrect.
Resolves rdar://132411524
Put AvailabilityRange into its own header with very few dependencies so that it
can be included freely in other headers that need to use it as a complete type.
NFC.
Protocol conformances have a handful attributes that can apply to them
directly, including @unchecked (for Sendable), @preconcurrency, and
@retroactive. Generalize this into an option set that we carry around,
so it's a bit easier to add them, as well as reworking the
serialization logic to deal with an arbitrary number of such options.
Use this generality to add support for @unsafe conformances, which are
needed when unsafe witnesses are used to conform to safe requirements.
Implement general support for @unsafe conformances, including
producing a single diagnostic per missing @unsafe that provides a
Fix-It and collects together all of the unsafe witnesses as notes.
* Make pointer bounds non-experimental
* Rename @PointerBounds to @_SwiftifyImport
* Rename filenames containing PointerBounds
* Add _PointerParam exception to stdlib ABI test
* Add _PointerParam to stdlib API changes
* Rename _PointerParam to _SwiftifyInfo
`recordObjCOverride()` records semantic overrides for imported Obj-C methods.
Since these methods are imported from a different language, it doesn't make
sense to enforce Swift's member import visibility rules when performing lookups
to find overridden methods. Doing so caused the Constrain Solver to lack
important information needed to eliminate overloads, resulting in erroneous
ambiguities.
Resolves rdar://141636723.
The C++ span should be a non-escapable type but is imported as escapable
for backward compatibility reason. This is inherently unsafe, so make
sure std::span is imported as such. In the future, we plan to generate
safe overloads using Swift's Span and that will be the preferred way of
using the API.
