Re: [PATCH v2 0/3] rust: const_eval: add a mechanism to do const trait calls

["Eliot Courtney" <ecourtney@nvidia.com>]

On Wed Sep 9, 2026 at 11:32 PM JST, Gary Guo wrote:
> On Mon Sep 7, 2026 at 10:21 AM BST, Eliot Courtney wrote:
>> On Fri Sep 4, 2026 at 12:46 AM JST, Gary Guo wrote:
>>> On Thu Sep 3, 2026 at 4:34 PM BST, Miguel Ojeda wrote:
>>>> On Thu, Sep 3, 2026 at 5:21 PM Gary Guo <gary@garyguo.net> wrote:
>>>>>
>>>>> Add a mechanism to perform const trait method calls, with the same type
>>>>> inference capabilities. This would only work for concrete types (not inside
>>>>> generic functions).
>>>>>
>>>>> As an example, convert `as_char_ptr_in_const_context`, so instead of
>>>>>
>>>>>     kernel::str::as_char_ptr_in_const_context(c_str)
>>>>>
>>>>> one may write
>>>>>
>>>>>     const_call!(c_str.as_char_ptr())
>>>>>
>>>>> instead.
>>>>>
>>>>> Signed-off-by: Gary Guo <gary@garyguo.net>
>>>>
>>>> Thanks for resending this!
>>>>
>>>> So I definitely like having `#[const_eval_only]`, because we do want
>>>> to limit those calls sometimes and it is nice to have C++'s equivalent
>>>> to `consteval` (perhaps this could be called just `#[consteval]` to
>>>> match?); i.e. what we talked about in the call.
>
> I'm okay with either. I put in `_only` because I think it makes it very clear
> that this is not just const-evalutable (that's provided by `const`) and is
> *always* const evaled.
>
>>>>
>>>> On `Const` and `const_call!`, i.e. the other two patches: I am not
>>>> sure if we care that much to introduce a macro for it, but it doesn't
>>>> look bad either. I guess perhaps it helps to have a consistent way to
>>>> do it and to easily remove it later when the actual feature lands in
>>>> Rust. I would like to hear some thoughts on it from others.
>
>>>
>>> I think it's good to have such macro compared to our current approach; it can be
>>> made into a identity macro once we have const trait impl.
>>>
>>> FWIW, I even considered an extra macro that allow you to write
>>>
>>>     #[const_trait_impl] // #[const_trait_impl(foreign)] for core types
>>>     impl MyTrait for MyType {}
>>>
>>> and generate the `impl Const` part too (and also add a line to the documentation
>>> that the method is const-callable with `const_call!()`).
>>>
>>> Best,
>>> Gary
>>
>> I agree with the consteval macro, that sounds very nice.
>>
>> On the const_call! I have similar thoughts to Miguel. I reckon if we
>> have a bunch of prospective users of this right now, and some thoughts
>> on what would use this in the future then it's good. If it's just one
>> use case right now, then not so sure.
>
> Some other candidates include `Alignable` (we currently have a "const_align_up")
> and maybe `IntoSafeCast`.
>
> Best,
> Gary

I'm definitely not opposed to a const trait polyfill, it's pretty cool
actually IMO. I also do think there is a bit of survivorship bias on the
argument (which I made) that there are not many users. And it can be
removed once const traits become available in MSRV. I am keen to hear
from others on this.