On Thu, 2 Apr 2026 at 16:02, Mark Hoemmen <mark.hoemmen_at_[hidden]> wrote:

> Jonathan Wakely wrote:
> > So you can't use contract checks as performance hints
> > in your code without disabling checks at API boundaries between
> > untrusted (or less trusted, or less tested) components.
>
> My understanding is that the preferred idiom is to perform contract
> checks first, then use `assume` or analogous features for performance
> optimization, as in the following example.
>
> template<size_t ByteAlignment>
> void foo(span<float> x)
> pre(std::is_sufficiently_aligned<ByteAlignment>(x.data())
> {
> span x_a(std::assume_aligned<ByteAlignment>(x.data());
> // ... compute with x_a ...
> }
>
> Is that right?
>
>
My view is that no, it's not right in general. For the reasons given in
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2020/p2064r0.pdf

And as found by Adrian in the first post in the thread, turning all
assertions into assumptions can hurt performance as much as it can help.
You should use assumptions more conservatively, in targeted places where
profiling has shown that it matters. See also
https://discourse.llvm.org/t/llvm-assume-blocks-optimization/71609

Not every precondition that you want to check using an assertion is a
suitable optimization hint. There might be some things that you want to
assert and assume (separately), but often you'll want just one or the other.