[clang] [APINotes] Add SWIFT_RETURNS_(UN)RETAINED support (PR #118938)

via cfe-commits cfe-commits at lists.llvm.org
Fri Dec 6 16:35:50 PST 2024


================
@@ -373,6 +373,13 @@ void ReadFunctionInfo(const uint8_t *&Data, FunctionInfo &Info) {
       endian::readNext<uint16_t, llvm::endianness::little>(Data);
   Info.ResultType = std::string(Data, Data + ResultTypeLen);
   Data += ResultTypeLen;
+
+  unsigned SwiftReturnOwnershipLength =
+      endian::readNext<uint16_t, llvm::endianness::little>(Data);
+  Info.SwiftReturnOwnership = std::string(reinterpret_cast<const char *>(Data),
+                                          reinterpret_cast<const char *>(Data) +
+                                              SwiftReturnOwnershipLength);
----------------
fahadnayyar wrote:

@compnerd my approach approach for `SwiftReturnOwnership` was to do something similar to `ResultType`.  Another reason for choosing `std:string` instead of `enum` was that `SwiftImportAs` is also implemented as `std:string`. `SWIFT_RETURNS_(UN)_RETAINED` annotation is closely related to `SWIFT_SHARED_REFERENCE` annotation. So I wanted to keep APINotes implementation of both of these similar. 

But I do agree that `enum` is a better approach for compression and diagnostics. But I also feel we'd not se much benefit unless we change other things like `SwiftImportAs` also to use `enums` for implementation. Do you think we should do such refactoring for all the annotations together in a separate follow-up patch?

https://github.com/llvm/llvm-project/pull/118938


More information about the cfe-commits mailing list