I wrote out this whole analysis of why different existential types
might have the same logical content, and then I turned around and
immediately uniqued existential shapes purely by logical content
rather than the (generalized) formal type. Oh well. At least it's
not too late to make ABI changes like this.
We now store a reference to a mangling of the generalized formal
type directly in the shape. This type alone is sufficient to unique
the shape:
- By the nature of the generalization algorithm, every type parameter
in the generalization signature should be mentioned in the
generalized formal type in a deterministic order.
- By the nature of the generalization algorithm, every other
requirement in the generalization signature should be implied
by the positions in which generalization type parameters appear
(e.g. because the formal type is C<T> & P, where C constrains
its type parameter for well-formedness).
- The requirement signature and type expression are extracted from
the existential type.
As a result, we no longer rely on computing a unique hash at
compile time.
Storing this separately from the requirement signature potentially
allows runtimes with general shape support to work with future
extensions to existential types even if they cannot demangle the
generalized formal type.
Storing the generalized formal type also allows us to easily and
reliably extract the formal type of the existential. Otherwise,
it's quite a heroic endeavor to match requirements back up with
primary associated types. Doing so would also only allows us to
extract *some* matching formal type, not necessarily the *right*
formal type. So there's some good synergy here.
This pipes the `-static` flag when building a static library into IRGen.
This should have no impact on non-Windows targets as the usage of the
information simply removes the `dllexport` attribute on the generated
interfaces. This ensures that a library built with `-static` will not
re-export its interfaces from the consumer. This is important to ensure
that the consumer does not vend the API surface when it statically links
a library. In conjunction with the removal of the force load symbol,
this allows the generation of static libraries which may be linked
against on Windows. However, a subsequent change is needed to ensure
that the consumer does not mark the symbol as being imported from a
foreign module (i.e. `dllimport`).
A "accessible" function that can be looked up based on a string key,
and then called through a fully-abstracted entry point whose arguments
can be constructed in code.
This was a relict from the -sil-serialize-all days. This linkage doesn't make any sense because a private function cannot be referenced from another module (or file, in case of non-wmo compilation).
Previously, a LinkEntity for an AST async function pointer was built by
passing an AbstractFunctionDecl. Later, decl was used to construct a
SILDeclRef.
That arrangement meant that clients could not construct such a
LinkEntity whose SILDeclRef::Kind could not be inferred from the dynamic
type of the decl from which the SILDeclRef was constructed. In
particular, clients could not construct a LinkEntity for the initializer
corresponding to a ConstructorDecl.
Here, the arrangment is changed so that the LinkEntity for an AST async
function pointer is built by passing a SILDeclRef.
When adding an AsyncFunctionPointer to the TBD, a LinkEntity of kind
AsyncFunctionPointerAST is constructed containing the decl. That decl
is then wrapped in a SILDeclRef which is then mangled.
Previously, the kind of the SILDeclRef was erroneously forced to be
Func. The result was a failure to mangle the TBD symbol for async
constructors correctly.
Here, that argument is omitted so that the kind can be determined
appropriately by SILDeclRef's constructor.
rdar://80485869
We were not making references to async function pointers "weak" when
the function itself was weak, because we were always calculating
linkage as-if we were defining the async function pointer.
Fixes the rest of rdar://79674106.
Gather 'round to hear tell of the saga of autolinking in incremental
mode.
In the beginning, there was Swift code, and there was Objective-C code.
To make one import bind two languages, a twinned Swift module named the same as an
Objective-C module could be imported as an overlay. But all was not
well, for an overlay could be created which had no Swift content, yet
required Swift symbols. And there was much wailing and gnashing of teeth
as loaders everywhere disregarded loading these content-less Swift
libraries.
So, a solution was found - a magical symbol _swift_FORCE_LOAD_$_<MODULE>
that forced the loaders to heed the dependency on a Swift library
regardless of its content. It was a constant with common linkage, and it
was good. But, along came COFF which needed to support autolinking but
had no support for such matters. It did, however, have support for
COMDAT sections into which we placed the symbol. Immediately, a darkness
fell across the land as the windows linker loudly proclaimed it had
discovered a contradiction: "_swift_FORCE_LOAD_$_<MODULE> cannot be
a constant!", it said, gratingly, "for this value requires rebasing."
Undeterred, we switched to a function instead, and the windows linker
happily added a level of indirection to its symbol resolution procedure
and all was right with the world.
But this definition was not all right. In order to support multiple
translation units emitting it, and to prevent the linker from dead
stripping it, Weak ODR linkage was used. Weak ODR linkage has the nasty
side effect of pessimizing load times since the dynamic linker must
assume that loading a later library could produce a more definitive
definition for the symbol.
A compromise was drawn up: To keep load times low, external linkage was
used. To keep the linker from complaining about multiple strong
definitions for the same symbol, the first translation unit in the
module was nominated to recieve the magic symbol. But one final problem
remained:
Incremental builds allow for files to be added or removed during the
build procedure. The placement of the symbol was therefore dependent
entirely upon the order of files passed at the command line. This was no
good, so a decree was set forth that using -autolink-force-load and
-incremental together was a criminal offense.
So we must compromise once more: Return to a symbol with common linkage,
but only on Mach-O targets. Preserve the existing COMDAT-friendly
approach everywhere else.
This concludes our tale.
rdar://77803299
Up to now, there had been no need to define a LinkEntity for a partial
apply forwarder. Now that async partial apply forwarders will each have
their own async function pointer, an entity is needed to pass to the
code that generates the async function pointers.
No demangling or remangling changes are required because that code has
existed for as long as partial apply forwarders to support demangling
their symbols.
- Add `DispatchThunkDerivative` and `MethodDescriptorDerivative` as link entities. The derivative functions of initializers, subscripts, properties, and methods are **all methods**, so we don't need other link entities for this purpose.
- Mangle dispatch thunks and method descriptors. Make `AutoDiffFunction` a context node since it can be nested.
Resolves SR-13866 (rdar://71318828) and SR-13125 (rdar://65240599).
This adds new kinds of link entities corresponding to the three
dispatch thunk link entity kinds:
- DispatchThunkAsyncFunctionPointer
- DispatchThunkInitializerAsyncFunctionPointer
- DispatchThunkAllocatorAsyncFunctionPointer
Previously, the suffix "AD" was used to mangle AsyncFunctionPointers.
That was incorrect because it was already used in the mangling scheme.
Here, that error is fixed by using 'u' under the thunk or specialization
operator 'T' to mangle AsyncFunctionPointers. Additionally, printing
and demangling support is added.
rdar://problem/72336407
Emit a once token when adding canonical prespecialized metadata records
to a nominal type descriptor and add the token itself as a trailing
object to the type descriptor. The new token will, in subsequent
commits, enable the canonical prespecialized metadata records attached
to the type descriptor to be added to the metadata cache exactly once.
When a generic type from a different module is not resilient within the
current module and at least one of its arguments is from the current
module, emit a non-canonical prespecialized record, and access that
metadata via a call to swift_getCanonicalSpecializedMetadata, passing in
the non-canonical record.
rdar://problem/56996727
rdar://problem/56997022
When generic metadata for a class is requested in the same module where
the class is defined, rather than a call to the generic metadata
accessor or to a variant of typeForMangledNode, a call to a new
accessor--a canonical specialized generic metadata accessor--is emitted.
The new function is defined schematically as follows:
MetadataResponse `canonical specialized metadata accessor for C<K>`(MetadataRequest request) {
(void)`canonical specialized metadata accessor for superclass(C<K>)`(::Complete)
(void)`canonical specialized metadata accessor for generic_argument_class(C<K>, 1)`(::Complete)
...
(void)`canonical specialized metadata accessor for generic_argument_class(C<K>, count)`(::Complete)
auto *metadata = objc_opt_self(`canonical specialized metadata for C<K>`);
return {metadata, MetadataState::Complete};
}
where generic_argument_class(C<K>, N) denotes the Nth generic argument
which is both (1) itself a specialized generic type and is also (2) a
class. These calls to the specialized metadata accessors for these
related types ensure that all generic class types are registered with
the Objective-C runtime.
To enable these new canonical specialized generic metadata accessors,
metadata for generic classes is prespecialized as needed. So are the
metaclasses and the corresponding rodata.
Previously, the lazy objc naming hook was registered during process
execution when the first generic class metadata was instantiated. Since
that instantiation may occur "before process launch" (i.e. if the
generic metadata is prespecialized), the lazy naming hook is now
installed at process launch.
MSVC does not realize that the switch is exhaustive and requires that
the path is explicitly marked as unreachable. This silences the C4715
warning ("not all control paths return a value").
It looks like the only thing that fails is the linkage computation
for the dynamic replacement key of class methods. Even though
methods have hidden linkage to prevent them from being directly
referenced from outside a resilient module, we need to ensure
the dynamic replacement key is visible.
Fixes <rdar://problem/58457716>.
This slightly regresses the standard library build (intentionally) while
generally improving the build of dispatch, foundation, xctest.
Rather than continuing to rely on the short-term hack of special casing
the standard library, identify the module where the decl originates
from. If the module is the current module, then assume that the symbol
need not be imported (static linking does not currently work on Windows
anyways). This allows for properly identifying the module where the
symbol will be homed.
Because we no longer special case the standard library here, a few known
metadata types will be incorrectly marked as being imported rather than
local.
Since linked entities which have `Shared` SILLinkage should be module
local, special case them to always be local. Without this the metadata
access function is still marked incorrectly.
With this computation we now get nearly all the cases correct. Dispatch
no longer has to rely on the linker relaxing the import to a local
binding. XCTest is also clean. Foundation misses the following case:
- `$sSS10FoundationE19_bridgeToObjectiveCAA8NSStringCyF`
The regressed cases in swiftCore are:
- `$sBi64_WV`
- `$sBi8_WV`
- `$sBi16_WV`
- `$sBi32_WV`
- `$sBpWV`
- `$syycWV`
- `$sBoWV`
- `$sBOWV`
- `$sBbWV`
- `$sytWV`
This seems unused currently, as attempting to query the DeclContext of
these types results in an immediate assertion that the kind of entity is
invalid. These are non-root protocol conformances, use
`getProtocolConformance` instead of the copy-paste value of
`getRootProtocolConformance`.
Prespecialized metadata may be requested in multiple modules and those
definitions must be deduplicated when statically linking those modules
together.
Here, the SILLinkage for prespecialized metadata is specified to be
Shared. Consequently, the IRLinkage's Linkage for this record is made
to be LinkOnceODRLinkage.
rdar://problem/56997414
SIL differentiability witnesses are a new top-level SIL construct mapping
an "original" SIL function and derivative configuration to derivative SIL
functions.
This patch adds `SILDifferentiabilityWitness` IRGen.
`SILDifferentiabilityWitness` has a fixed `{ i8*, i8* }` layout:
JVP and VJP derivative function pointers.
Resolves TF-1146.
The weak imported flag is now only set if the attribute is unconditionally
weak linked, which is the case when it or one of its parent contexts has a
@_weakLinked attribute.
To correctly handle weak linking based availability with serialized SIL
functions, we need to serialize the actual version tuple when the SIL function
was introduced. This is because the deployment target of the client app can
be older than the deployment target that the original module was built with.
Fixes <rdar://problem/52783668>.
First, remove the AvailabilityContext parameter; it was confusing because
we actually always want to use the deployment target here.
Then, split this method up into three methods:
- isAlwaysWeakImported(): simply checks for a @_weakLinked attribute, either
on the declaration itself or one of its parent contexts.
- getAvailabilityForLinkage(): returns the OS version availability when
this declaration was introduced, or if the declaration does not have
explicit availability, check it's storage (if its an accessor), or its
parent contexts.
- isWeakImported(ModuleDecl *fromModule): combines these two checks to
determine if the declaration should be weak linked when referenced from
the given module, or if it might be weak referenced from some module
(if the module parameter is null).
When we generate code that asks for complete metadata for a fully concrete specific type that
doesn't have trivial metadata access, like `(Int, String)` or `[String: [Any]]`,
generate a cache variable that points to a mangled name, and use a common accessor function
that turns that cache variable into a pointer to the instantiated metadata. This saves a bunch
of code size, and should have minimal runtime impact, since the demangling of any string only
has to happen once.
This mostly just works, though it exposed a couple of issues:
- Mangling a type ref including objc protocols didn't cause the objc protocol record to get
instantiated. Fixed as part of this patch.
- The runtime type demangler doesn't correctly handle retroactive conformances. If there are
multiple retroactive conformances in a process at runtime, then even though the mangled string
refers to a specific conformance, the runtime still just picks one without listening to the
mangler. This is left to fix later, rdar://problem/53828345.
There is some more follow-up work that we can do to further improve the gains:
- We could improve the runtime-provided entry points, adding versions that don't require size
to be cached, and which can handle arbitrary metadata requests. This would allow for mangled
names to also be used for incomplete metadata accesses and improve code size of some generic
type accessors. However, we'd only be able to take advantage of the new entry points in
OSes that ship a new runtime.
- We could choose to always symbolic reference all type references, which would generally reduce
the size of mangled strings, as well as make runtime demangling more efficient, since it wouldn't
need to hit the runtime caches. This would however require that we be able to handle symbolic
references across files in the MetadataReader in order to avoid regressing remote mirror
functionality.
Since the return value of getAccessor() depends on mutable state, it
does not make sense in the request evaluator world. Let's begin by
removing some utility methods derived from getAccessor(), replacing
calls to them with calls to getAccessor().
This was done even for non-public inits because subclasses would always
override the base class's designated initializers, even if they were
inaccessible.
This is an ABI break, however in practice the only affected class
initializer was ManagedBuffer.init(_doNotCallMe:()), and we can just
make it @usableFromInline.
