This is consistent with imported error codes, which are always
Hashable. URLError.Code was also Hashable in Swift 3.1 by virtue of
being defined as an enum; the change to a struct broke that.
rdar://problem/32066434
Logs a warning the first time a problematic class is archived or
unarchived. We expect people to actually fix these issues, so the
performance of the warning isn't too important.
Sample output:
[timestamp] Attempting to archive Swift class '_Test.Outer.ArchivedThenUnarchived', which does not have a stable runtime name.
[timestamp] Use the 'objc' attribute to ensure that the runtime name will not change: "@objc(_TtCC5_Test5Outer22ArchivedThenUnarchived)"
[timestamp] If there are no existing archives containing this class, you can choose a unique, prefixed name instead: "@objc(ABCArchivedThenUnarchived)"
Finishes rdar://problem/32414508
To get the full benefit of dyld3 on Darwin platforms, pointer relocations need to be pointer-aligned, which unfortunately requires growing some key path data structures a little bit. This does tidy up some code that had to hack around our lack of unaligned load/store operations on UnsafeRawPointer, at least. While we're here, we can also simplify the identification strategy for reabstracted stored properties; we only need the property index to identify, not the absolute offset. rdar://problem/32318829
* Implements a debugger hook (breakpoint) API and data structure. This structure is passed to the debugger and describes extra information about a fatal error or a non-fatal warning, which should be logged as a runtime issue.
This debugger hook is then used from two places, which currently only log to stderr:
- Runtime exclusivity violations.
- Swift 3 implicit Obj-C entrypoints.
A subsequent LLDB support will be able to catch these callbacks and show the runtime issues in a better way than just logging them to stderr. When the debugger is not attached, this shouldn't have any effect.
* IRGen: EmptyBoxType's representation cannot be nil because of a conflict with extra inhabitant assumption in indirect enums
We map nil to the .None case of Optional. Instead use a singleton object.
SR-5148
rdar://32618580
We need to use the ivar offset variables in this case, since the Swift field offset vector doesn't pick up the adjusted offsets from the ObjC runtime. Fixes SR-5036 | rdar://problem/32488871.
Specifically, objc_setUncaughtExceptionHandler, whose nullability is
now annotated in Swift 4 mode when using the Xcode 9 SDKs. That means
it has a different signature in Swift 3 and Swift 4 mode, which means
that references from SIL will fail when deserialized in the other
mode. (The recovery support added in earlier commits doesn't cover
serialized SIL.)
This resulted in crashes when building with
-build-serialized-stdlib-unittest, which enables -sil-serialize-all /
"magic performance mode" when building the StdlibUnittest module like
it does the standard libarry.
None of this affects developer code, because developer code doesn't
serialize anything (if they're playing by the rules).
SingleValueDecondingContainers in JSON and Plist previously held the
assertion that attempting to decode an array or dictionary from them
was a type mismatch (since those represented unkeyed and keyed
containers, respectively). This assertion is no longer true, though,
since encode<T : Encodable>(_:) and decode<T : Decodable>(_:) allow
you to do just that.
This lifts the assertion and adds unit tests to both implementations to
ensure this works. (Addresses https://bugs.swift.org/browse/SR-5089)
When this initializer is invoked with an integer literal, the expression
is ambiguous, since both UIn32 and Float are
RepresentableByIntegerLiteral.
Fixes: https://bugs.swift.org/browse/SR-5176
