In https://github.com/apple/swift/pull/66560 , a bug in the lowering of
`global_addr` was fixed. Part of that fix was to postpone mapping the
type of the global into context until getting the address of the global
and projecting the buffer for the out-of-line value; at that point, the
type is mapped into context and the address is cast.
It introduced an issue for fixed-size globals, however: the type of such
globals was not mapped into context; the result was that the lowered
value set for the corresponding SIL value would have the wrong type.
Fix that by extracting the code which mapped the type into context and
cast the address to the appropriate lowered type into a lambda and call
that lambda both in both the fixed-size (newly) and non-fixed-size (as
before) cases.
rdar://114013709
Deallocate dynamic allocas done for metadata/wtable packs. These
stackrestore calls are inserted on the dominance frontier and then stack
nesting is fixed up. That was achieved as follows:
Added a new IRGen pass PackMetadataMarkerInserter; it
- determines if there are any instructions which might allocate on-stack
pack metadata
- if there aren't, no changes are made
- if there are, alloc_pack_metadata just before instructions that could
allocate pack metadata on the stack and dealloc_pack_metadata on the
dominance frontier of those instructions
- fixup stack nesting
During IRGen, the allocations done for metadata/wtable packs are
recorded and IRGenSILFunction associates them with the instruction that
lowered. It must be the instruction after some alloc_pack_metadata
instruction. Then, when visiting the dealloc_pack_metadata instructions
corresponding to that alloc_pack_metadata, deallocate those packs.
Fix the type of the `alloca` created by `GenPack`'s for type metadata
and witness tables by fixing its callee, `emitDynamicAlloca` to always
return a `StackAddress` whose `Address`' type is the one specified by
the caller.
rdar://109540863
As an optimization, there is a fast-cast to non-final classes that just
compares isa pointers. If the source of the cast is an Optional<class>,
though, it's not allowed to "directly compare the isa-pointer"; because
the source might be Optional.none, i.e. null.
Add a check for nil when emitting the fast class cast.
rdar://108614878
We started using clang to emit the _OBJC_PROTOCOL_ definition.
But we would use a different name for the proto_list definition than clang.
"OBJC_LABEL_PROTOCOL$" (objc)
|--> OBJC_PROTOCOL
"\01l_OBJC_LABEL_PROTOCOL$_" (swift)
|--> OBJC_PROTOCOL
If an Objective C object also emitted the same protocol definition you could
end up in a situation where both clang's and swift's proto_list definitions
point to the same protocol definition.
Older linkers don't like that.
rdar://108505376
* [IRGen] Pass component generic sig when emitting key path component for external property
rdar://101179225
When no generic environment was present, we passed nullptr, which is not correct when the property uses an external associated type, causing crashes in IRGen. In those cases, we have to pass the component generic sig instead.
* Fix test
There are certain protocol method decls types that swift does not import
today.
```
@protocol Incomplete
- (id)getObjectFromVarArgs:(id)first, ...;
@end
```
Furthermore, the old method also emitted duplicate entries for protocols
methods when Swift synthesized methods for diagnosics.
We won't import this method into Swift. So if we emit protocol metadata
from swift delcs we would generate incomplete records.
rdar://60888524
Extended existential type shapes can trigger this by introducing
more entities (and thus causing GlobalVars to be rehashed) during
the lazy-emission callback.
Fixes rdar://98995607
Previously, switches over extremely large integers (e.g. i832) were
emitted when emitting switches with many spare bits. Such switches
result in unfortunate downstream codegen.
Here, for large enums (currently, more than two words) the preexisting
EnumPayload::emitCompare function is used to compare each of the enum
cases in turn with the payload's value. The result is a series of
cond_br where the conditional is made by anding together word-size
chunks of the payload value with word-size chunks of each enum case's
tag, subject to masking.
rdar://83158525
When analyzing a struct's layout to determine whether it contains a
single non-empty field, bail upon encountering a field that is not ABI
accessible.
Previously, rather than bailing (though that was the intent), the field
was ignored. The result was that struct's with a single non-empty field
and any number of ABI inaccessible fields would be treated as if they
only had a single non-empty field. Trouble ensued.
rdar://79513293
This allows them to be used in generic arguments for NSArray et al.
We already do this for the ones that wrap bridged values (like
NSString/String), but failed to do it for objects that /weren't/
bridged to Swift values (class instances and protocol compositions),
or for Error-which-is-special.
In addition to this being a sensible thing to do, /not/ doing this led
to IRGen getting very confused (i.e. crashing) when we imported a
Objective-C protocol that actually used an NS_TYPED_ENUM in this way.
(We actually shouldn't be using Swift's IRGen logic to emit protocol
descriptors for imported protocols at all, because it's possible we
weren't able to import all the requirements. But that's a separate
issue.)
https://bugs.swift.org/browse/SR-6844
