[embedded] Initial support for generic classes in embedded Swift #68461
Conversation
- VTableSpecializer, a new pass that synthesizes a new vtable per each observed concrete type used - Don't use full type metadata refs in embedded Swift - Lazily emit specialized class metadata (LazySpecializedClassMetadata) in IRGen - Don't emit regular class metadata for a class decl if it's generic (only emit the specialized metadata)
…alized vtables have intentionally different ABIs from the non-generic one
… on non-final generic functions
kubamracek
requested review from
hborla,
slavapestov and
xedin
as code owners
|
@swift-ci please test |
|
@swift-ci please test |
|
@swift-ci please test |
| @@ -3322,7 +3329,8 @@ IRGenFunction::emitTypeMetadataRef(CanType type, | |||
|
|
|||
| if (type->hasArchetype() || | |||
| !shouldTypeMetadataAccessUseAccessor(IGM, type) || | |||
| isa<PackType>(type)) { | |||
| isa<PackType>(type) || | |||
| type->getASTContext().LangOpts.hasFeature(Feature::Embedded)) { | |||
There was a problem hiding this comment.
Optional nit: might be good to assert that type is a class type so that it's easier to diagnose if this path is taken by a non-class-type accidentally (so we'd assert sooner).
There was a problem hiding this comment.
I believe this is already covered by the assert at the top of the function.
| /*ConvertIndirectToDirect=*/true, | ||
| /*dropMetatypeArgs=*/false); | ||
|
|
||
| if (!ReInfo.canBeSpecialized()) { |
There was a problem hiding this comment.
I guess this would just be if it's not serialized, right?
| llvm::errs() << "unsupported use of class method " | ||
| << newCM->getMember().getDecl()->getName() << " in function " | ||
| << newCM->getFunction()->getName() << '\n'; | ||
| llvm::report_fatal_error("unsupported class method"); |
There was a problem hiding this comment.
What's the case where this could happen? If there are uses of a class method that aren't an apply? For example escaping the function pointer maybe? Seems like some of these fatal errors should be turned into diagnostics (that can be a follow up though).
There was a problem hiding this comment.
We discussed this with Erik a bit, and I think we concluded that this is something that you shouldn't be able to trigger this assert from any Swift source code, but it's technically possible to write (custom) SIL that fails this (e.g. by doing what you're saying, but SILGen or SILOptimizer shouldn't really ever produce that code).
as?) of class instances