On 11/09/2021 4:19 am, Patrice Roy wrote:
> Noted; I'll find a way to put in in the document, v2 that I hope to send
> next week, so please verify that I communicated your thoughts properly
> when you get it. Thanks Matt!

Just to follow up, did you get anywhere with this Patrice?





On 12/09/2021 12:34 am, Patrice Roy wrote:
> Small personal note : I don't think a switch to turn constexpr off alone
> can work, as constexpr kicks in with constexpr contexts :
>
> unsigned long long factorial(int n) {
> return n == 0? 1ULL : n * factorial(n-1);
> }
> int main() {
> float arr[factorial(5)]{}; // array of 120 float all initialized
to 0.0f
> return factorial(5); // runtime function call
> }
>
> We can debug with the runtime call, but if we turned constexpr off the
> whole program would stop compiling due to arr.
>
> If you think we need a mechanism to selectively disable constexpr, I can
> add it to the document but we'd need to define the intent more clearly.


I think just something which ignores the constexpr on functions and
always runs them at runtime, is all that's likely to be needed, if I'm
right about it becoming a performance problem (and I could be wrong if
the compilers get smart about when they're resolving at compile time vs
runtime).
But obviously stuff like "constexpr if" cannot be switched off as the
non-called branch may create invalid code depending on the arguments of
the if statement.