[Lldb-commits] [lldb] [lldb][TypeSystemClang] Add support for floating point template argument constants (PR #127206)

[Michael Buch via lldb-commits]

================
@@ -1973,6 +1973,27 @@ class DWARFASTParserClang::DelayedAddObjCClassProperty {
   ClangASTMetadata m_metadata;
 };
 
+static clang::APValue MakeAPValue(CompilerType clang_type, uint64_t bit_width,
+                                  uint64_t value) {
+  bool is_signed = false;
+  const bool is_integral = clang_type.IsIntegerOrEnumerationType(is_signed);
+
+  llvm::APSInt apint(bit_width, !is_signed);
+  apint = value;
+
+  if (is_integral)
+    return clang::APValue(apint);
+
+  uint32_t count;
+  bool is_complex;
+  assert(clang_type.IsFloatingPointType(count, is_complex));
+
+  if (bit_width == 32)
+    return clang::APValue(llvm::APFloat(apint.bitsToFloat()));
+
+  return clang::APValue(llvm::APFloat(apint.bitsToDouble()));
----------------
Michael137 wrote:

Hmmm good question

The only differentiator between the different float types we have in DWARF is the `DW_AT_name` of the type and the `DW_AT_byte_size`. So we could set the semantics based on the floating point type, so we probably won't be able to support all the different available `fltSemantics`. We already create Clang types from the `DW_AT_name`, so we could, e.g., differentiate between `_Float16` and `__bf16` semantics by looking at the typename. Then 32-bits and 64-bits we would assume IEEE. And if none of those work then we could return a `std::nullopt` here? Wdyt?

https://github.com/llvm/llvm-project/pull/127206