ilya-biryukov wrote: > This patch resolves it by visiting the body twice (of course, it's > suboptimal), however patch #99882 still crashes because that relies on the > mechanism above, I believe.
I have also observed it crashes, but I still think that in principle we don't need to traverse the body twice. If `DeclStmt` calls `DiagnoseUnexpandedPacks`, it will set the correct flag on the inner `LambdaScopeInfo`, which should later be available in the `LambdaExpr` itself, and can either be "consumed" by `CXXFoldExpr` (as in the example test case) or propagated into the outer `LamdaScopeInfo` from another call to `DiagnoseUnexpandedPacks`. I believe that's what's already happening during parsing, but we seem to somehow miss setting this flag during substitution. Are we crashing because we are missing calls to `DiagnoseUnexpandedPacks` during substitution or because we don't set up the `LambdaScopeInfo` during substitution correctly? https://github.com/llvm/llvm-project/pull/86265 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
