[Lldb-commits] [lldb] [lldb][NFCI] Rewrite UnwindAssemblyInstEmulation in terms of a CFG visit (PR #169630)

[David Spickett via lldb-commits]

================
@@ -150,29 +152,38 @@ bool UnwindAssemblyInstEmulation::GetNonCallSiteUnwindPlanFromAssembly(
   EmulateInstruction::InstructionCondition last_condition =
       EmulateInstruction::UnconditionalCondition;
 
-  for (const InstructionSP &inst : inst_list.Instructions()) {
-    if (!inst)
-      continue;
-    DumpInstToLog(log, *inst, inst_list);
+  std::deque<std::size_t> to_visit = {0};
+  llvm::SmallSet<std::size_t, 0> enqueued = {0};
+
+  // Instructions reachable through jumps are inserted on the front.
+  // The next instruction in inserted on the back.
+  // Pop from the back to ensure non-branching instructions are visited
+  // sequentially.
+  while (!to_visit.empty()) {
+    std::size_t current_index = to_visit.back();
----------------
DavidSpickett wrote:

No I just forgot that C++ be weird sometimes.

I'd put the pop_back() immediately after the back() to make it clear there isn't a state in this code where you rely on having a copy of the index but that index still being the back of the queue.

But it's only one line away so do whatever you prefer.

Surprising to me there isn't a pop_back that returns a value but if there were, you'd be "paying" for the copy and hoping the compiler was smart enough to discard it. And you know how C++ is with paying for things you don't use.

https://github.com/llvm/llvm-project/pull/169630