This code would previously read off the end of the allocated metadata to fetch these values. This was usually harmless, as the value was never used in that case. However, on rare occasions the metadata would be right before unmapped memory, and this read would crash trying to access that unmapped memory.
rdar://problem/39866044
Dynamic subclasses aren't /really/ valid Swift type metadata, but
they can still be used as values of type AnyClass. Make sure we
don't assert when that happens.
No intended functionality change.
If we only emit an opaque reflection record for a struct or class, then we can't reflect its fields. We failed both to clear the "is reflectable" bit in the context descriptor for non-reflectable structs, and to check for the bit before trying to present a struct's fields as children in the runtime. rdar://problem/41274260
Clang-importer-synthesized declarations get an extra tag character included in their mangling, which was not being preserved in type context descriptors. This caused runtime lookup for these synthesized types to fail. Fix this by adding the tag information to type context descriptors and teaching the runtime to match it up when fetching metadata by mangled name. Fixes rdar://problem/40878715.
This provides a slight amount of defense against attackers constructing mangled names with offsets crafted to JOP the runtime into attacker-controlled code. (Someone could still find some random code or constant data artifact in a binary that *looks* like a mangled string with symbolic references and theoretically attack that way, but they at least wouldn't be able to construct their own string entirely.)
Protocol name mangling didn’t always go through a path that allowed the use
of standard substitutions. Enable standard substitutions for protocol name
manglings where they make sense.
Removes ~277k from the standard library binary size.
Since the mangling scheme and set of standard library types is effectively
fixed now, introduce known mangling substitutions for a number of new
standard library types, filling out the S[A-Za-z] space.
Reduces standard library binary size by ~195k.
LLVM r334283 changed StringRef::split(char) to be implemented using
StringRef::split(StringRef), which is not defined inline. Because Swift
uses StringRef without linking LLVM's libSupport.a, we can only use
functions that are defined inline in the headers. Swift currently only
builds LLVM for the host, so we cannot link libSupport.a without building
it for every target, which would be a big change. Instead, this changes
a few places in Swift to avoid using those split and rsplit functions.
rdar://problem/41029268
libstdc++ included with GCC 4.8 does not define `std::max_align_t` as
required by the C++11 specification. As a workaround, explicitly create
the definition locally. This was fixed in GCC 4.9 and later.
The C compiler on some platforms (such as s390x) assumes that the
data pointed to by symbols meets certain alignment requirements.
The swift sections do not necessarily meet these alignment
requirements so this change adds alignment attributes to them to
force the compiler to emit the instruction sequences and relocations
required to address unaligned data.
This fixes a 'R_390_PC32DBL target misaligned' warning issued by
gold on s390x.
