[llvm] Fix compress/decompress in LLVM Offloading API (PR #150064)

[Joseph Huber via llvm-commits]

================
@@ -32,42 +32,56 @@ namespace llvm {
 
 namespace object {
 
+// CompressedOffloadBundle represents the format for the compressed offload
+// bundles.
+//
+// The format is as follows:
+// - Magic Number (4 bytes) - A constant "CCOB".
+// - Version (2 bytes)
+// - Compression Method (2 bytes) - Uses the values from
+// llvm::compression::Format.
+// - Total file size (4 bytes in V2, 8 bytes in V3).
+// - Uncompressed Size (4 bytes in V1/V2, 8 bytes in V3).
+// - Truncated MD5 Hash (8 bytes).
+// - Compressed Data (variable length).
 class CompressedOffloadBundle {
 private:
-  static inline const size_t MagicSize = 4;
-  static inline const size_t VersionFieldSize = sizeof(uint16_t);
-  static inline const size_t MethodFieldSize = sizeof(uint16_t);
-  static inline const size_t FileSizeFieldSize = sizeof(uint32_t);
-  static inline const size_t UncompressedSizeFieldSize = sizeof(uint32_t);
-  static inline const size_t HashFieldSize = sizeof(uint64_t);
-  static inline const size_t V1HeaderSize =
-      MagicSize + VersionFieldSize + MethodFieldSize +
-      UncompressedSizeFieldSize + HashFieldSize;
-  static inline const size_t V2HeaderSize =
-      MagicSize + VersionFieldSize + FileSizeFieldSize + MethodFieldSize +
-      UncompressedSizeFieldSize + HashFieldSize;
   static inline const llvm::StringRef MagicNumber = "CCOB";
-  static inline const uint16_t Version = 2;
 
 public:
-  LLVM_ABI static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
+  struct CompressedBundleHeader {
+    unsigned Version;
+    llvm::compression::Format CompressionFormat;
+    std::optional<size_t> FileSize;
+    size_t UncompressedFileSize;
+    uint64_t Hash;
+
+    static llvm::Expected<CompressedBundleHeader> tryParse(llvm::StringRef);
+  };
+
+  static inline const uint16_t DefaultVersion = 2;
+
+  static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
   compress(llvm::compression::Params P, const llvm::MemoryBuffer &Input,
-           bool Verbose = false);
-  LLVM_ABI static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
-  decompress(llvm::MemoryBufferRef &Input, bool Verbose = false);
+           uint16_t Version, bool Verbose = false);
+  static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
+  decompress(const llvm::MemoryBuffer &Input, bool Verbose = false);
 };
 
 /// Bundle entry in binary clang-offload-bundler format.
 struct OffloadBundleEntry {
   uint64_t Offset = 0u;
   uint64_t Size = 0u;
   uint64_t IDLength = 0u;
-  StringRef ID;
-  OffloadBundleEntry(uint64_t O, uint64_t S, uint64_t I, StringRef T)
-      : Offset(O), Size(S), IDLength(I), ID(T) {}
+  std::string ID;
+  OffloadBundleEntry(uint64_t O, uint64_t S, uint64_t I, std::string T)
+      : Offset(O), Size(S), IDLength(I) {
+    ID.reserve(T.size());
+    ID = T;
----------------
jhuber6 wrote:

This is confusing, wouldn't the copy constructor already allocate this? Was the issue `StringRef` going out of scope? That's more easily solved with `T.str()` in the initializer list.

https://github.com/llvm/llvm-project/pull/150064