Thanks Marshall! I agree about the unfortunate weaving of GCC specific logic in type_traits and my first attempt locally went down the route of per-feature macro abstraction of __has_feature(). It became unwieldy rather quickly. Also the existing code had multiple patterns for feature checking. :(
With this patch, things should be normalized to (a) Use _GNU_VER and __cplusplus where possible (b) use a feature macro *if* the gcc and clang built-in are syntactically different (eg __is_literal() vs __is_literal_type()). I personally think that directly using the gcc version check to gate which feature is used in type_traits is an "okay" tradeoff. An alternative is to simulate __has_feature() for gcc in __config. If you can tell me which pattern you'd like, I'm happy to go through and beat things into that shape. lmk. -Albert http://reviews.llvm.org/D4340 _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
