[lldb] [llvm] Annotate disassembly with register‐resident variable locations (PR #147460)

[Abdullah Mohammad Amin via llvm-commits]

================
@@ -53,6 +54,37 @@ bool DWARFExpressionList::ContainsAddress(lldb::addr_t func_load_addr,
   return GetExpressionAtAddress(func_load_addr, addr) != nullptr;
 }
 
+std::optional<DWARFExpressionList::DWARFExpressionEntry>
+DWARFExpressionList::GetExpressionEntryAtAddress(lldb::addr_t func_load_addr,
+                                                lldb::addr_t load_addr) const {
+  if (const DWARFExpression *always = GetAlwaysValidExpr()) {
+    return DWARFExpressionEntry{std::nullopt, always};
+  }
+
+  if (func_load_addr == LLDB_INVALID_ADDRESS)
+    func_load_addr = m_func_file_addr;
+
+  // Guard against underflow when translating a load address back into file space.
+  if (load_addr < func_load_addr)
+      return std::nullopt;
+
+  // Guard against overflow.
+  lldb::addr_t delta = load_addr - func_load_addr;
+  if (delta > std::numeric_limits<lldb::addr_t>::max() - m_func_file_addr)
----------------
UltimateForce21 wrote:

>From when I dug into MathExtras last time, I found that its `AddOverflow` and `SubOverflow` helpers are only defined for **signed types**.

I they rely on two’s-complement signed overflow semantics and return the truncated signed result. Since `lldb::addr_t` is an **unsigned address,** I think using those helpers would either fail to compile or silently convert to signed semantics. 

That's why I tried to implement the signed overflow check into the function.

https://github.com/llvm/llvm-project/pull/147460