[Lldb-commits] [lldb] [lldb] Step over non-lldb breakpoints (PR #174348)

Duncan McBain via lldb-commits lldb-commits at lists.llvm.org
Mon Jan 5 10:38:45 PST 2026


================
@@ -138,6 +138,17 @@ class Architecture : public PluginInterface {
       std::shared_ptr<const UnwindPlan> current_unwindplan) {
     return lldb::UnwindPlanSP();
   }
+
+  /// Returns whether a given byte sequence is a valid breakpoint for the
+  /// architecture. Some architectures have breakpoint instructions that
+  /// have immediates that can take on any value, resulting in a family
+  /// of valid byte sequences. Bases the size comparison on the reference.
+  virtual bool IsValidBreakpointInstruction(llvm::ArrayRef<uint8_t> reference,
+                                            llvm::ArrayRef<uint8_t> observed) const {
+    if (reference.size() > observed.size())
----------------
DuncanMcBain wrote:

The intent here is to make sure that one way or another, I don't run off the end of either array. Logically in my head I expect that if we have matched the correct instruction, the size check doesn't matter, and if we are unable to read as many bytes as are in the reference sequence then they don't match.

I think in general the handling around this is a little bit unsatisfying - the code as it exists seems to expect that if the small encoding is available for a platform, it *will* be used. It is possible that, where different encodings are possible for an architecture, we should just check all of them in turn.

https://github.com/llvm/llvm-project/pull/174348


More information about the lldb-commits mailing list