<ratio>: Use constexpr functions instead of TMPs
#2450
Conversation
|
Regarding the |
|
Closing this PR because the use of TMPs was to implement LWG-2094, which makes overflow a SFINAE error instead of emitting diagnostics. There is no way to achieve the same thing with |
|
|
Since it is not required to be used for SFINAE, it can be made into a function, and a non-constexpr function can be called when overflow errors, triggering compiler error.
|
|
||
| template <intmax_t _Ax, intmax_t _Bx, bool _Good, bool _Also_good> | ||
| struct _Safe_addX : integral_constant<intmax_t, _Ax + _Bx> {}; // computes _Ax + _Bx without overflow | ||
| inline void _Safe_add_integer_arithmetic_overflow_error() noexcept {} |
There was a problem hiding this comment.
Maybe comment as intentionally empty and not constexpr?
There was a problem hiding this comment.
Maybe call abort() to make is safe for runtime too?
There was a problem hiding this comment.
Actually never mind, as it is ready to merge.
There was a problem hiding this comment.
Same pattern as:
Line 175 in 59a87cc
I'd be fine with commenting this pattern with // intentionally not constexpr in the future, although I agree that it isn't worth resetting testing here. However, calling abort() seems excessive, and then I'd worry about the compiler actually emitting codegen for functions that aren't going to be called.
|
I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed. |
|
Thanks for using modern |
Fixes #2291
Blocked on https://developercommunity.visualstudio.com/t/1202945Why not make
_Gcda copy ofstd::gcd?std::gcdincludes large headers such as<xutility>and<limits>. I did not dare try to make a new header (<xratio>?) because I was not 100% sure if I can do it correctly.The callers of
_Gcdonly work withintmax_t, so I figured it is easier to just update_Gcdand its callers rather than worrying about making a new header.