[clang] Thread Safety Analysis: Support reentrant capabilities (PR #137133)

[Marco Elver via cfe-commits]

================
@@ -235,6 +266,20 @@ class FactSet {
     return false;
   }
 
+  std::optional<FactID> replaceLock(FactManager &FM, iterator It,
+                                    std::unique_ptr<FactEntry> Entry) {
+    if (It == end())
+      return std::nullopt;
+    FactID F = FM.newFact(std::move(Entry));
+    *It = F;
+    return F;
+  }
+
+  std::optional<FactID> replaceLock(FactManager &FM, const CapabilityExpr &CapE,
+                                    std::unique_ptr<FactEntry> Entry) {
+    return replaceLock(FM, findLockIter(FM, CapE), std::move(Entry));
+  }
+
----------------
melver wrote:

replaceLock without caller-iterator still saves a copy+pop_back+push_back. Not much, but not nothing either.

This PR should include the bare minimum use of replaceLock - I have one patch queued that is not part of this PR to do an unrelated substitution with replaceLock.

If anything I'd move replaceLock introduction before this PR, but I think we're just splitting hairs at that point.

https://github.com/llvm/llvm-project/pull/137133