tahonermann added a comment. > Note we might be confused, the parens there aren't completely clear as to > what your intent is.
Well, I know that I'm confused and not clear what my intent is :) I asked the question because there appears to be an asymmetry between (temp.friend)p9 sentence 1 <http://eel.is/c++draft/temp.friend#9.sentence-1> and (temp.friend)p9 sentence 2 <http://eel.is/c++draft/temp.friend#9.sentence-2>. Sentence 1 applies to all constrained non-template friend declarations regardless of any template argument dependence while sentence 2 applies to just a subset of constrained friend function templates; those that have some amount of template dependence. The difference impacts when mangling differences are required. I spent some time analyzing how gcc, clang, and MSVC handle these different cases. See https://godbolt.org/z/85E5eMh3x. Search for FIXME? to find cases where I think one or more of the compilers is misbehaving or where it is unclear to me whether or how [temp.friend]p9 applies. Some highlights: - Some compilers diagnose some ill-formed cases when parsing the class template, others don't until the class template is instantiated. Not surprising. - All of the compilers reject non-template friend function definitions with non-dependent constraints due to duplicate definitions, presumably in violation of [temp.friend]p9; see the `ff2()` example. - All of the compilers reject non-template friend function definitions with dependent constraints due to duplicate definitions, presumably in violation of [temp.friend]p9; see the `ff6()` example. - Name mangling is currently insufficient to differentiate (otherwise) non-dependent friend function templates with dependent constraints; see the `fft5()` and `fft6()` examples. - None of the compilers reject some cases of non-definitions that should be rejected by [temp.friend]p9; see the `fft5()` and `fft7()` examples. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D126818/new/ https://reviews.llvm.org/D126818 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
