Merge the three-stage operation originally designed for field vectors into a single unified loop that acts directly on the ivar offsets instead of using a faux field offset vector.
We really don’t need ‘em; we can just adjust the direct field offsets.
The runtime entry point currently uses a weird little hack that we will refactor away shortly.
When an @objc @implementation class requires the use of `ClassMetadataStrategy::Update` because some of its stored properties do not have fixed sizes, we adjust the direct field offsets during class realization by emitting a custom metadata update function which calls a new entry point in the Swift runtime. That entry point adjusts field offsets like `swift_updateClassMetadata2()`, but it only assumes that the class has Objective-C metadata, not Swift metadata.
This commit introduces an alternative mechanism which does the same thing without using any Swift-only metadata. It’s a rough implementation with important limitations:
• We’re currently using the field offset vector, which means that field offsets are being emitted into @objc @implementation classes; these will be removed.
• The new Swift runtime entry point duplicates a lot of `swift_updateClassMetadata2()`’s implementation; it will be refactored into something much smaller and more compact.
• Availability bounds for this feature have not yet been implemented.
Future commits in this PR will correct these issues.
The descriptor and arguments for prespecialized metadata will always be in the shared cache. Skip creating the mangling for any lookup involving pointers outside the shared cache, as an optimization.
LLVM is presumably moving towards `std::string_view` -
`StringRef::startswith` is deprecated on tip. `SmallString::startswith`
was just renamed there (maybe with some small deprecation inbetween, but
if so, we've missed it).
The `SmallString::startswith` references were moved to
`.str().starts_with()`, rather than adding the `starts_with` on
`stable/20230725` as we only had a few of them. Open to switching that
over if anyone feels strongly though.
When ObjC interop is enabled, we emit what we think will be the class's InstanceStart and InstanceSize based on what we know about the superclass. We then fix up those values at runtime if they don't match. The compiler will emit this data into read-only memory if it knows they will always match, and then the runtime avoids writing to these fields if they already contain the correct value.
However, the compiler aligns the InstanceStart, but instance size is not aligned. For example:
class Super<T> { var bool = true }
class Sub: Super<Int> { var obj: AnyObject? }
Super's InstanceSize is 17 (on 64-bit) but Sub's InstanceStart is 24. The compiler sees a fixed layout and emits Sub's rodata into constant memory. The runtime sees that 24 does not equal 17 and tries to update it, but we don't want it to.
Instead, only update InstanceStart if it's too small to accommodate the superclass's InstanceSize. If it's overlay large then we'll just leave it alone. The compiler underestimates InstanceStart when it doesn't know the superclass's size so this should only happen due to alignment.
rdar://123695998
This library uses GenericMetadataBuilder with a ReaderWriter that can read data and resolve pointers from MachO files, and emit a JSON representation of a dylib containing the built metadata.
We use LLVM's binary file readers to parse the MachO files and resolve fixups so we can follow pointers. This code is somewhat MachO specific, but could be generalized to other formats that LLVM supports.
rdar://116592577
Create a version of the metadata specialization code which is abstracted so that it can work in different contexts, such as building specialized metadata from dylibs on disk rather than from inside a running process.
The GenericMetadataBuilder class is templatized on a ReaderWriter. The ReaderWriter abstracts out everything that's different between in-process and external construction of this data. Instead of reading and writing pointers directly, the builder calls the ReaderWriter to resolve and write pointers. The ReaderWriter also handles symbol lookups and looking up other Swift types by name.
This is accompanied by a simple implementation of the ReaderWriter which works in-process. The abstracted calls to resolve and write pointers are implemented using standard pointer dereferencing.
A new SWIFT_DEBUG_VALIDATE_EXTERNAL_GENERIC_METADATA_BUILDER environment variable uses the in-process ReaderWriter to validate the builder by running it in parallel with the existing metadata builder code in the runtime. When enabled, the GenericMetadataBuilder is used to build a second copy of metadata built by the runtime, and the two are compared to ensure that they match. When this environment variable is not set, the new builder code is inactive.
The builder is incomplete, and this initial version only works on structs. Any unsupported type produces an error, and skips the validation.
rdar://116592420
For calloc, the variable denoting the of elements comes first,
then the variable denoting the size of each element. However, both
arguments are swapped when calling this function in many places in this codebase.
When forming runtime metadata for a function type that has either `any
Error` or `Never` as the specified thrown error type, drop the thrown
error type and normalize down to (untyped) `throws` or non-throwing,
as appropriate.
Extend function type metadata with an entry for the thrown error type,
so that thrown error types are represented at runtime as well. Note
that this required the introduction of "extended" function type
flags into function type metadata, because we would have used the last
bit. Do so, and define one extended flag bit as representing typed
throws.
Add `swift_getExtendedFunctionTypeMetadata` to the runtime to build
function types that have the extended flags and a thrown error type.
Teach IR generation to call this function to form the metadata, when
appropriate.
Introduce all of the runtime mangling/demangling support needed for
thrown error types.
In C++20, the compiler will synthesize a version of the operator
with its arguments reversed to ease commutativity. This reversed
version is ambiguous with the hand-written operator when the
argument is const but `this` isn't.
The `ivar` reference still pointed to the original list, so the first modification went to the wrong place. Change `ivar` to a pointer, and re-point it when we copy the list.
rdar://116339597
There's often nothing that needs to be fixed up in the ivar list. In that case, we can avoid copying it. This saves time and memory, and allows the class rodata to be in immutable memory.
rdar://116189946
We fix up the VWT pointer, but not the heap destroyer. This doesn't matter for classes which use ObjC refcounting, which is the common case for dynamic subclasses, because that doesn't use the heap destroyer pointer. But it does matter for classes that use native Swift refcounting, such as classes that don't inherit from NSObject, or actors.
rdar://113657917
_swift_addRefCountStringForMetatype and _swift_refCountBytesForMetatype diverged in the code that determines whether a type is a reference, causing the size number of ref count bytes to differ from the actually used bytes. This can cause early termination of the runtime interpreter functions, which in turn causes unbalanced reference counts.
rdar://112474091
The tag was overwritten after resolve when a prior field caused a non-zero offset. This then caused the runtime to treat is a relative instead of an absolute pointer, causing invalid pointers to be dereferenced.
ReflectionContext::allocationMetadataPointer() was reading the
metadata pointer from a wrong offset because of the out-of-sync struct
layouts and dump-generic-metadata was not working correctly.
This change resync's the layouts and adds a static_assert to verify
that the offsets match between GenericMetadataCacheEntry and
GenericCacheEntry.
* [IRGen] Implement support for multi payload enums in layout strings
rdar://105837114
* Implement multi payload enum cases in swift_resolve_resilientAccessors
* Add missing const
rdar://109790722
Layout strings are not guaranteed to be properly aligned, so we can't directly read from and write to it, but have to use memcpy instead.
* [IRGen] Fix layout string generation for pre-specialized metadata
rdar://108012057
Pre-specialized metadata has to be specifically handled by using the bound generic type instead of the unbound one. All the necessary information is already being passed down as BoundGenericTypeCharacteristics, we just need to apply them when present.
* Add tests and a few fixes
* Fixes after rebase
* Attempt to fix Windows linker issue in test
If something fails while trying to generate a name for an ObjC class,
we fall back to a generated name based on the metadata pointer. This
ensures that we won't crash, and if the name turns up in the debugger
or somewhere it should be obvious that something went wrong.
rdar://107718586
