On Wed, Mar 24, 2021 at 06:52:40PM +0100, Thomas Monjalon wrote: > 24/03/2021 18:28, Tyler Retzlaff: > > > > so to qualify. you mean maybe it is breaking compilation of c++ in a > > compiler that explicitly violates c++ standard when compiling c++? that > > would mean it is not a c++ compiler. > > The asm keyword is part of all C++ standards? > It seems asm is non-standard in C, > that's why we use __asm__.
the keyword is standard the meaning of the keyword is implementation defined for both C and C++. though the C11 standard describes a "common implementation is via statement of the form ..." [J.5.10] > > in general i don't think it is a good practice to have dpdk introduce > > names into the application namespace unqualified, but the point you make > > is valid it can break c++ compilation if something was using this macro > > as a convenience to the compiler specific extension __asm__. there will > > be further issues with varying syntaxes that __asm__-style extensions > > take from compiler to compiler as well. > > Yes we need to make sure there is no specific extension involved. > Is C++ asm the same as the C __asm__? i don't think I can answer that as it depends on the compiler. there may be no implementation at all or different implementations for asm or __asm__. so basically neither form is portable (for an arch). though for __asm__ probably there is de-facto standardization around what gcc does and clang mimics. > > would you prefer that i change the preprocessor protection to include only > > windows? since i'm certain that this will break for any c++ compiler on > > windows the moment any stl header is included. > > No, C++ is probably the right scope. > > I don't know yet. I would like to understand the global picture, > and have it properly documented in this commit log. yep, no problem. i suspect we are probably the only ones using c++ and dpdk (though others can speak up if they do too) which may be why this has gone unnoticed until now.
