Should `sycl::exception::what()` be noexcept?

[KornevNikita]

Specification Version

SYCL 2020 (Revision 11)

Section Number(s)

4.13.2

Bug Description

std::exception declares what() as:

virtual const char* what() const noexcept;

sycl::exception inherits std::exception, therefore sycl::exception::what() should also be noexcept, otherwise it causes weaker exception specification.

[KornevNikita]

Please ping me if needed, I'll create a PR

[tahonermann]

Assuming sycl::exception::what() doesn't add any preconditions (I would be quite surprised if it did), I agree this can, and probably should, be declared noexcept.

[TApplencourt]

> const char* what() const

Returns an implementation-defined non-null constant C-style string that describes the error that triggered the exception.

So no precondition, so I trust you for the noexcep because I guess if not we can recurse infinitely ? And it's indeed no-except in c++ :)

[tahonermann]

Recursion isn’t the concern. Calling a function such that a precondition is violated is UB. UB can do anything; including throwing exceptions. Not declaring functions with preconditions noexcept allows implementations to detect detectable precondition violations and to respond by throwing an exception, which is sometimes useful (e.g. in debugging modes).

[gmlueck]

Makes sense to me.