https://github.com/AaronBallman commented:
We have been taking a harder stance on diagnostics which are ignored by default because there's plenty of evidence that they don't get enabled often enough to warrant adding them. This diagnostic seems to suffer from that: https://sourcegraph.com/search?q=context:global+lang:Makefile+-Wlarge-by-value-copy&patternType=keyword&sm=0 https://sourcegraph.com/search?q=context:global+lang:CMake+-Wlarge-by-value-copy&patternType=keyword&sm=0 Those numbers are low enough I wonder whether the diagnostic carries enough weight to retain... Oh look, it has bugs: https://godbolt.org/z/WE6Kb58Td (I think these bugs apply to other warnings where we do special work for them in the driver). An alternative is to give `NumLargeByValueCopy` a default value of 64 like we do in the driver: https://github.com/llvm/llvm-project/blob/ddd7b231806c8c515a6c033e7f799e413cfb982a/clang/lib/Driver/ToolChains/Clang.cpp#L6839 but given how little this seems to be used in practice, it's not clear how disruptive such a change would be. It wouldn't fix the usability issues with the pragma either. https://github.com/llvm/llvm-project/pull/214696 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
