Am Dienstag, dem 09.07.2024 um 17:28 +0000 schrieb Joseph Myers: > On Sat, 29 Jun 2024, Martin Uecker wrote: > > > + bool typeless = flag_isoc2y > > + && ((char_type_p (type) > > + && !(type_quals & TYPE_QUAL_ATOMIC)) > > + || (AGGREGATE_TYPE_P (type) > > + && TYPE_TYPELESS_STORAGE (type))); > > I'm not convinced this should be limited to C2Y mode, rather than applied > also in older modes as an extension. How likely is it that users of older > modes would thereby see a performance regression in code that already > worked reliably (as opposed to having code start to work more reliably > because they were assuming things about aliasing that weren't actually > guaranteed)?
I agree that it is better to activate it in earlier modes too. I believe both C++ and clang and for both C and C++ have this, so the risk of making false assumptions is very high. I do not have any benchmarks though (except my own projects where it does not matter). Should we simply try to activate it and see what happens? Martin >
