================
@@ -4672,39 +4672,69 @@ static bool
buildCapturedStmtCaptureList(Sema &S, CapturedRegionScopeInfo *RSI,
SmallVectorImpl<CapturedStmt::Capture> &Captures,
SmallVectorImpl<Expr *> &CaptureInits) {
+ llvm::SmallPtrSet<VarDecl *, 4> CapturedDecomposed;
for (const sema::Capture &Cap : RSI->Captures) {
if (Cap.isInvalid())
continue;
+ ValueDecl *CapVar = nullptr;
+ if (Cap.isVariableCapture()) {
+ CapVar = Cap.getVariable();
+ if (auto *BD = dyn_cast<BindingDecl>(CapVar)) {
+ VarDecl *DD = cast<VarDecl>(BD->getDecomposedDecl());
+ if (!CapturedDecomposed.insert(DD).second) {
+ continue; // Skip duplicate.
+ }
----------------
alexey-bataev wrote:
Skipping duplicate captures means the resulting `CapturedStmt` has fewer
captures than `RSI->Captures` originally tracked. AST consumers (template
instantiation, AST dumpers, source-to-source rewriters, libTooling clients)
that walk `S.captures()` in lockstep with the original tracked bindings will be
off by one or more. Verify this is safe for existing template instantiation
paths (e.g. a function template that captures a binding pair in OpenMP,
instantiated multiple times) and add a regression test.
https://github.com/llvm/llvm-project/pull/190832
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
