https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65890
--- Comment #7 from frankhb1989 at gmail dot com --- (In reply to Jonathan Wakely from comment #6) > (In reply to frankhb1989 from comment #5) > > Mainly for testing of the conformance. > > I don't understand what this means. Testing what? G++? G++ does not exist > for you to test its conformance to a standard. Most users don't care about > slavish conformance to a defective specification, they want a useful > compiler. > Ok, since there is no "strictly conforming program" concept in ISO C++ as it in ISO C, I'd better to clarify, the direct point is portability of the legacy user code. Checking and ensuring the code (carefully kept undefined behavior or "no diagnostics required" away) to be standard-compliant is one aspect of usefulness of the compiler which provides "standard" modes. How can I do if 'g++ -std=c++03 -pedantic-errors' behaves different than other C++03-conforming compilers compiling the ill-formed C++03 code, besides to drop the code away? > > Although it is treated a defect of > > the design and has been changed later, the old rules are still well-defined > > and the published standard itself is consistent. So if I did not get wrong > > about the purpose of '-std=', this should be a bug. Whether it is worth > > being fixed is another problem. > > You are wrong about how -std options work. We incorporate dozens of DRs into > all modes, instead of making them only apply to later standard modes. > The manual says nothing about this. It tells me "to obtain all the diagnostics required by the standard, you should also specify -pedantic (or -pedantic-errors if you want them to be errors rather than warnings)". I thought the "standard" here is one of the published ones. And there seems to be no separate options to control the features in individual DRs. If the compiler frontend does right (by design) as you said, this is a documentation issue, since I am really confused about what g++ are allowed to do. Now it seems I'd better simply not rely on these -std options of g++ for these works. > > On the other hand, I'd debate the resolution of this CWG issue is not pure > > improvement. There could be a trick to distinguish static and non-static > > data members through SFINAE on expressions like 'sizeof(&(C::x))'. It is > > broken now. > > SFINAE in C++03 was not nearly as useful, and doesn't work for private > members. The language is more useful now, there is no reason to hobble it > with a foolish consistency to a defective design. Yes, it is obviously not so useful as C++11/14/1z. But can we just get rid of C++98/03 totally for this reason? For this issue, the (current) result is, 'g++ -std=c++03 -pedantic-errors' actually implements a dialect of C++03 with some subtle "patches" in the standard, which is difficult in many aspects even to experienced users. And even this is a resolved defect, the design of these related features in the language is still arguably more or less defective. The latter is nothing to do with g++ directly, though.
