There are scenarios where different compilers are distributed with
compatible serialization format versions and the same tag. Distinguish
swiftmodules in such a case by assigning them to different distribution
channels. A compiler expecting a specific channel will only read
swiftmodules from the same channel. The channels should be defined by
downstream code as it is by definition vendor specific.
For development, a no-channel compiler loads or defining the env var
SWIFT_IGNORE_SWIFTMODULE_REVISION skips this new check.
rdar://123731777
When scanning finds a dependency in the same package, do not load
public/private swiftinterface since they do not have the package level
decl to compile the current module. Always prefer package module (if
enabled), or use binary module, unless it is building a public/private
swiftinterface file in which case the interface file is preferred.
This also does some clean up to sync up the code path between implicit
and explicit module finding path.
rdar://122356964
Recently, unexpected binary module sharing between compilers that had
inconsistent views of the `PlatformKind` enumeration caused me to spend several
days debugging undefined behavior in the compiler. We should validate that
enumeration values decoded during deserialization are valid, and fail fast when
they are not to make debugging this kind of issue less time consuming. This
change just validates the `PlatformKind` value decoded for an `AvailableAttr`,
but more of deserialization could be updated to do similar validation.
Resolves rdar://123770273
For years, optimizer engineers have been hitting a common bug caused by passes
assuming all SILValues have a parent function only to be surprised by SILUndef.
Generally we see SILUndef not that often so we see this come up later in
testing. This patch eliminates that problem by making SILUndef uniqued at the
function level instead of the module level. This ensures that it makes sense for
SILUndef to have a parent function, eliminating this possibility since we can
define an API to get its parent function.
rdar://123484595
Fix a use-after-free bug in package-name extraction code, where the
BumpPtrAllcoator is destroyed before a StringRef that uses the allocated
name is used.
We want to preserve the borrow scope during switch dispatch so that move-only
checking doesn't try to analyze destructures or consumes out of it. SILGen
should mark anywhere that's a potential possibility with its own marker so that
it gets borrow checked independently.
While working on #71425, I hit an assert indicating the `@_section` attribute was not
handled in `DeclDeserializer::deserializeDeclCommon`. This change adds the missing
deserialization for the `@_section` attribute.
The `ABI` headers had accidentally grown an `#include` into compiler headers,
allowing the enum constant values of the `ValueOwnership` enum to leak into
the runtime ABI. Sever this inappropriate relationship by declaring a separate
`ParameterOwnership` enum with ABI-stable values in the ABI headers, and
explicitly converting between the AST and ABI representation where needed.
Fixes rdar://122435628.
When `-enable-lazy-typecheck` is specified, serialization may be expected to
run on an AST containing invalid declarations since type checking may happen
on-demand, during serialization, in this mode. If the declarations that are
invalid are not skipped, then the compiler is likely to crash when attempting
to serialize them. Now, invalid declarations are skipped and an error is
emitted at the end of serialization to note that serialization failed.
Additionally, a new `-Rmodule-serialization` flag can be specified to request
more detailed information about module serialization failures. This would be
useful in a situation where lazy typechecking does not produce any diagnostic
for some reason, but module serialization fails and more information is
therefore required to debug.
Resolves rdar://123260476
Always prefer binary module when using @testable imports because the
swiftmodule rebuilt from interface cannot be imported as testable.
rdar://123120159
When a NoncopyableGenericsMismatch happens between the compiler and
stdlib, allow the compiler to rebuild the stdlib from its interface
instead of exiting with an error.
By populating the memory cache before loading the module, we can avoid a cycle
where a module is imported that is an overlay, which then triggers
ClangImporter, which then (redundantly) triggers the import of the overlay
module, which would reimport the module again, since it's import is still
underway and it hasn't been entered into the cache yet.
rdar://118846313
[transferring] Implement transferring result and clean up transferring param support by making transferring a bit on param instead of a ParamSpecifier.
Instead it is a bit on ParamDecl and SILParameterInfo. I preserve the consuming
behavior by making it so that the type checker changes the ParamSpecifier to
ImplicitlyCopyableConsuming if we have a default param specifier and
transferring is set. NOTE: The user can never write ImplicitlyCopyableConsuming.
NOTE: I had to expand the amount of flags that can be stored in ParamDecl so I
stole bits from TypeRepr and added some logic for packing option bits into
TyRepr and DefaultValue.
rdar://121324715
The reason why I am doing this is that I am going to be changing transferring to
not be a true ParamSpecifier. Instead, it is going to be a bit on Param that
changes the default ParamSpecifier used. That being said, I cannot use consuming
for this purpose since consuming today implies no implicit copy semantics, which
we do not want unless the user specifically asks for it by writing consuming.
Relax some existing pattern matches and add some unhandled instructions to the
walkers so that borrowing switches over address-only enums are properly analyzed
for incorrect consumption. Add a `[strict]` flag to `mark_unresolved_move_only_value`
to indicate a borrow access that should remain a borrow access even if the subject
is later stack-promoted from a box.
Ad-hoc requirements are now obsolete by making `remoteCall`,
`record{Argument, ReturnType}`, `decodeNextArgument` protocols
requirements and injecting witness tables for `SerializationRequirement`
conformances during IRGen.
The SDK build version is a decent heuristic for expected changes in the
SDK. Any change in SDK, to clang headers in particular, can break
references from cached swiftmodules.
Track the SDK build version as part of the swiftmodule cache hash. This
will ensure we rebuild from swiftinterfaces on SDK updates.
rdar://122655978
