libstdc++: questions about P3663

Luc Grosheintz luc.grosheintz@gmail.com
Fri Nov 7 13:10:28 GMT 2025


Dear libstdc++ devs,

While adding the checking of mandates/prerequisites I
noticed that:

     7. N5014 requires that for a strided_slice either:
     slice.extent == 0 or slice.stride > 0.

     However, in P3663, the condition is split into:
       + constants (8.4.5.3.2): if slice.extent and slice.stride
          are constant_wrappers, then slice.stride > 0.
       + runtime (8.4.10.3): either slice.extent == 0 or
         slice.stride > 0.

Is this intentional? Essentially, AFAICT,
   strided_slice{0, cw<0>, cw<0>}; // invalid
   strided_slice{0, 0, 0}; // valid

Kind regards,
Luc

On 10/31/25 9:06 AM, Luc Grosheintz wrote:
> Dear libstdc++ devs,
> 
> While working on P3663, the following questions arose:
> 
>    1. What are the requirements for user provided integer-like objects
>       (when used as slices)? Do they need to have copy and move ctors?
>       There's a few locations where non-canonicalized slices are copied,
>       e.g. when passed to `submdspan_canonical_slices`.
> 
>    2. What happens if in the future the standard wants to add a new kind
>       of *canonical* selection? For example, block-wise selection: pick
>       N, skip M, pick N, skip M, etc. This can't be expressed by any of
>       the current canonical slices (currently there's also no layout to
>       express this idea, but presumably it would be added at the same
>       time).
> 
>    3. Is there a way to `static_assert` the mandate that something is a
>       "submdspan slice type for IndexType"?
> 
>    4. The notion of "unit-stride slice" is false for pairs and tuples,
>       even though they intuitively have unit strides. By renaming the
>       concept "canonical unit-stride slice", the name would match my
>       intuition again.
> 
>    5. There's a notion of "lower bound of the submdspan slice range"
>       which is also used in contexts where the slice could be empty, e.g.
>       `strided_slice(n, 0, 1)`. This represents the same empty interval
>       regardless of the value of `n`, the question is what is the "lower
>       bound"? Concretely, let `n == exts[i]`. Then, is `pair{n, n}` a
>       valid slice for dimension `i`? It sounds like yes, because the
>       condition:
>         > the kth interval of e contains the submdspan slice range of s
>           for the kth extent of e;
>      is satisfied due to the interval being empty. By the same logic,
>      `pair{n+1, n+1}` should also be valid, but `pair{cw<n+1>, cw<n+1>}`
>      isn't, because it's not a slice type for the kth extent.
> 
>    6. I'd like to suggest adding "canonical" to "valid submdspan slice
>       for the kth extent of e", because:
>       - a "submdspan slice type for IndexType" is not required to be
>         canonical,
>       - there's a couple of related concepts that explicitly mention the
>         word canonical,
>       - it makes it harder to know when `SliceSpecifiers` refer to
>         canonical slices, since that property isn't mentioned and only
>         implied by "valid submdspan slice for the kth extent of e".
> 
> Typos:
>    * In 8.15.2, `src` should be `src.extents()`.
> 
> Kind regards,
> Luc



More information about the Libstdc++ mailing list