[Lldb-commits] [lldb] [lldb] Speed up SymbolContextList::AppendIfUnique (PR #181952)

[Michael Buch via lldb-commits]

================
@@ -483,6 +484,20 @@ class SymbolContextList {
   // Member variables.
   collection m_symbol_contexts; ///< The list of symbol contexts.
 
+private:
+  struct SymbolContextInfo {
+    static SymbolContext getEmptyKey();
+    static SymbolContext getTombstoneKey();
+    static unsigned getHashValue(const SymbolContext &sc);
+    static bool isEqual(const SymbolContext &lhs, const SymbolContext &rhs);
+  };
+
+  /// Threshold above which we switch from linear scan to hash-set lookup
+  /// for uniqueness checks.
+  static constexpr size_t kSetThreshold = 1024;
+
+  llvm::SmallDenseSet<SymbolContext, 1, SymbolContextInfo> m_lookup_set;
----------------
Michael137 wrote:

I'd be a tiny bit weary of changing the ordering guarantees of this. Especially because this is a regression fix on top of a regression fix, which we likely need to get into a release. So if we can fix the performance issue while maintaining behaviour, that'd be ideal. And anything else can be follow-up

My 2 cents

https://github.com/llvm/llvm-project/pull/181952