The code was skipping 4/8 bytes to jump overn embedded reference, but it actually needs to skip 5/9 bytes in order to skip over the leading control character as well.
Also change the abort() calls to return nullptr so that we can fail more gracefully if this code is ever presented with bad data, since we want inspection tools to be robust in the face of garbage.
rdar://problem/56460096
TypeRefBuilder and MetadataReader had nearly identical symbolic reference resolvers,
but diverged because TypeRefBuilder had its own local/remote address management mechanism,
and because TypeRefBuilder tries to resolve opaque types to their underlying types, whereas
other MetadataReader clients want to preserve them as written in source. The first problem
has been addressed by making TypeRefBuilder use `RemoteRef` everywhere, and the second
can be handled with a flag (and might be able to be handled more elegantly with some more
refactoring of general opaque type handling in MetadataReader).
Instead of passing around raw local pointers and references, and spreading
tricky offset arithmetic around with the Local/RemoteAddress fields in
ReflectionInfo, have the TypeRefBuilder code use RemoteRefs everywhere,
which keep the remote/local mapping together in one unit and provide
centralized API for this logic.
This doesn't yet change how code uses the RemoteRef address data to
follow pointers across objects, for things like reading type refs, but
that should be much easier to do after this lands.
These are now always zero, because memory readers handle virtual address mapping.
The `swift_reflection_info_t` structure used by the C RemoteMirror API keeps
its offset fields because it's supposed to be a stable API, but we now assert that
the values are always zero.
This makes for a cleaner and less implicit-context-heavy API, and makes it easier for symbolic
reference resolvers to do context-dependent things (like map the in-memory base address back to a
remote address in MetadataReader).
There are situations where you want to build against a libc that is out
of tree or that is not the system libc (Or for cross build scenarios).
This is a change for passing the -sdk and include paths for things like
this.
The reason why I am doing this is so that I can create an adaptor class
(templated on Runtime) for reading protocol info from ReflectionInfo without
having to make ReflectionInfo itself generic. If ReflectionInfo becomes generic
on Runtime, it will cause a cascading need to mark classes in Reflection as
generic as well.
This dramatically reduces the number of needed malloc calls.
Unfortunately I had to add the implementation of SmallVectorBase::grow_pod to the runtime, as we don't link LLVM. This is a bad hack, but better than re-inventing a new SmallVector implementation.
SR-10028
rdar://problem/48575729
When compiling for a 32 bit machine, uintptr_t from ReflectionInfo will
be the integer sized to hold a 32 bit pointer, so a 64 bit pointer might
not fit.
This commit removes the solution in
0f20c486e0 and does a runtime check that
the calculated offset will fit into the target machine uintptr_t, which
might not be true for 32 bits machines trying to read 64 bits images,
which should not be that common (and those images have to have offsets
bigger than what a 32 bits number can hold).
Seems that the change in the two variables was spilling into the other
target of the file, but returning it back to the original values seems
to avoid that issue.
This should unbreak the Android CI build. In it, the Linux static
library was changing to the host compiler, and that compiler was being
used for the Android runtime library, which would have never compile
that way (since the host compiler in CI is an old-ish Clang without the
necessary argument).
If we nest a type inside a local context inside a generic type,
we have to look through the local context(s) to find the outer
generic type when stripping off generic arguments.
We don't support nominal types inside generic local context
right now, but this can happen with type aliases.
This is essentially a long-belated follow-up to Arnold's #12606.
The key observation here is that the enum-tag-single-payload witnesses
are strictly more powerful than the XI witnesses: you can simulate
the XI witnesses by using an extra case count that's <= the XI count.
Of course the result is less efficient than the XI witnesses, but
that's less important than overall code size, and we can work on
fast-paths for that.
The extra inhabitant count is stored in a 32-bit field (always present)
following the ValueWitnessFlags, which now occupy a fixed 32 bits.
This inflates non-XI VWTs on 32-bit targets by a word, but the net effect
on XI VWTs is to shrink them by two words, which is likely to be the
more important change. Also, being able to access the XI count directly
should be a nice win.
Previously, they would forward their unused spare bits to be used by other multi-payload enums, but
did not implement anything for single-payload extra inhabitants.
This restores the ability to build the standard library and the tools in two
separate build invocations. This is required to cross-compile the standard
library on various targets without building complete toolchains.
Bitwise takability is now part of the layout of a type, because
non-bitwise takable types are never stored inline in an
existential or resilient global's buffer, even if they would
fit.
The basic rule is that weak references, unknown-refcounted
unowned references, and aggregates that contain them, are not
bitwise takable, whereas everything else is bitwise takable.
Also, since the bitwise takable for an unowned reference
depends on the reference counting style, we have to record the
superclass of a protocol, if any, to correctly determine the
reference counting style of the protocol existential.
There is a build of the reflection library for the tools which we want to build
if we are building the tools. The target library should only be built if we are
building the standard library.
Right now we expect that every class and protocol has a field
descriptor that tells us if the entity is @objc or not.
For imported types, the descriptor will not exist if we did not
directly emit a field whose concrete type contains the imported
type. For example, in lldb, we might have a generic type whose
runtime substituted type includes an imported type.
In this case, TypeLowering would fail to produce a layout because
it did not find a field descriptor for the imported type.
A better approach is to have the TypeDecoder call a different
factory method for imported types, and handle them specially in
TypeLowering, bypassing the field type metadata altogether.
Remove this special case handling for building a host library as a target
library. This is the last piece needed to support cross-compiling lldb. As a
bonus, it cleans up some of the logic in our special build system.
The key thing here is that all of the underlying code is exactly the same. I
purposely did not debride anything. This is to ensure that I am not touching too
much and increasing the probability of weird errors from occurring. Thus the
exact same code should be executed... just the routing changed.
Like we did for structs, make it so that tuple types can also get extra inhabitants from whichever element with the most, not only the first. This lets us move all of the extra inhabitant handling functionality between structs and tuples in IRGen up to the common RecordTypeInfo CRTP base.
We can use the extra inhabitants of the type metadata field as extra inhabitants of the entire
existential container, allowing `Any?` and similar types to be the same size as non-optional
existentials.
Remote mirrors was hitting an assertion failure due to a generic parameter not being concrete. This check catches that case early and returns a clean failure from getSubstMap, which callers can then handle appropriately.
It also hit a casting failure in visitDependentMemberTypeRef, which assumed that SubstBase was either a NominalTypeRef or a BoundGenericTypeRef. This does a dynamic cast with a graceful failure.
In general there is a tension in this code between its use in the runtime, where we usually want to treat bad data as a horrible bug and fail loudly, and its use in remote mirrors, where we need to assume that the data we're examining might be horribly broken and we just want to do the best we can. Longer term we might want to make this code configurable so that we can have an "assert and die" mode for the runtime, and a "fail gracefully" mode for remote mirrors.
rdar://problem/40136609
This DEBUG(...) emulates LLVM's DEBUG(...), but it's controlled by a
different mechanism. LLVM's DEBUG(...) is getting renamed to
LLVM_DEBUG(...), so to keep this from looking like a use of the old
name I'm adjusting it to DEBUG_LOG.
When building with assertions enabled, link the demangle tree dumper into
the runtime and remote mirrors libraries. This makes debugging demangling-related issues a whole lot easier.
And use them in the reflection library (TypeRef). These were
private to `TypeRef.cpp` but can be moved to the demangler as
they can be of general use, and we can use them from lldb (which
has homemade versions of the functions as well). Bonus point,
it probably makes sense for these helpers to live in the demangler
anyway.
<rdar://problem/37710513>
This makes resolving mangled names to nominal types in the same module more efficient, and for eventual secrecy improvements, also allows types in the same module to be referenced from mangled typerefs without encoding any source-level name information about them.
Update IRGen to trigger generation of type metadata for foreign
struct types found in fields. And fix TypeRefBuilder to handle
the case where struct has fields but at the same time has opaque
metadata.
Update IRGen to trigger generation of type metadata for foreign
struct types found in fields. And fix TypeRefBuilder to handle
the case where struct has fields but at the same time has opaque
metadata.
Various TypeDecoder clients will depend on having the "bare" nominal
type declaration demangled node for looking up nominal type descriptors,
so move the generic argument-stripping code into TypeDecoder.
This makes them consistent no matter what shenanigans are pulled by
the importer, particularly NS_ENUM vs. NS_OPTIONS and NS_SWIFT_NAME.
The 'NSErrorDomain' API note /nearly/ works with this, but the
synthesized error struct is still mangled as a Swift declaration,
which means it's not rename-stable. See follow-up commits.
The main place where this still falls down is NS_STRING_ENUM: when
this is applied, a typedef is imported as a unique struct, but without
it it's just a typealias for the underlying type. There's also still a
problem with synthesized conformances, which have a module mangled
into the witness table symbol even though that symbol is linkonce_odr.
rdar://problem/31616162
TypeDecoder's interface with its builders already treated protocols as
a type (due to their being mangled as "protocol composition containing
one type"), and intermixed protocols with superclasses when forming
compositions. This makes for some awkwardness when working with
protocol descriptors, which are very much a distinct entity from a
type.
Separate out the notion of a "protocol declaration" (now represented
by the builder-provided BuiltProtocolDecl type) from "a protocol
composition containing a single type", similarly to the way we handle
nominal type declarations. Teach remote mirrors and remote AST to
handle the new contract.
