[committed 1/3] libstdc++: Fix std::span constraint for sizeof(size_t) < sizeof(int) [PR108221]

[committed 1/3] libstdc++: Fix std::span constraint for sizeof(size_t) < sizeof(int) [PR108221]

[Jonathan Wakely]

Tested x86_64-linux. Pushed to trunk.

-- >8 --

The default constructor has a constraint that is always false if
arithmetic on size_t values promotes to int. Rewrite the constraint
exactly as written in the standard, which works correctly.

libstdc++-v3/ChangeLog:

	PR libstdc++/108221
	* include/std/span (span::span()): Un-simplify constraint to
	work for size_t of lesser rank than int.
---
 libstdc++-v3/include/std/span | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/libstdc++-v3/include/std/span b/libstdc++-v3/include/std/span
index 251fed91abf..b336332b190 100644
--- a/libstdc++-v3/include/std/span
+++ b/libstdc++-v3/include/std/span
@@ -145,7 +145,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 
       constexpr
       span() noexcept
-      requires ((_Extent + 1u) <= 1u)
+      requires (_Extent == dynamic_extent || _Extent == 0)
       : _M_ptr(nullptr), _M_extent(0)
       { }
 
-- 
2.39.0

Re: [committed 1/3] libstdc++: Fix std::span constraint for sizeof(size_t) < sizeof(int) [PR108221]

[Jakub Jelinek]

On Tue, Jan 10, 2023 at 11:46:55AM +0000, Jonathan Wakely via Gcc-patches wrote:
> Tested x86_64-linux. Pushed to trunk.
> 
> -- >8 --
> 
> The default constructor has a constraint that is always false if
> arithmetic on size_t values promotes to int. Rewrite the constraint
> exactly as written in the standard, which works correctly.
> 
> libstdc++-v3/ChangeLog:
> 
> 	PR libstdc++/108221
> 	* include/std/span (span::span()): Un-simplify constraint to
> 	work for size_t of lesser rank than int.
> ---
>  libstdc++-v3/include/std/span | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/libstdc++-v3/include/std/span b/libstdc++-v3/include/std/span
> index 251fed91abf..b336332b190 100644
> --- a/libstdc++-v3/include/std/span
> +++ b/libstdc++-v3/include/std/span
> @@ -145,7 +145,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
>  
>        constexpr
>        span() noexcept
> -      requires ((_Extent + 1u) <= 1u)
> +      requires (_Extent == dynamic_extent || _Extent == 0)
>        : _M_ptr(nullptr), _M_extent(0)
>        { }

If it would be C++23 only, you could use ((_Extent + 1uz) <= 1uz).
As this is evaluated at compile time only, it is unfortunate it is
3 operations compared to former 2, but not a big deal.  If this was
in code that would be emitted at runtime, GCC already optimizes
(x == -1uz || x == 0)
or
(x == 0 || x == -1uz)
to
((x + 1uz) <= 1uz)

	Jakub

Re: [committed 1/3] libstdc++: Fix std::span constraint for sizeof(size_t) < sizeof(int) [PR108221]

[Jonathan Wakely]

[-- Attachment #1: Type: text/plain, Size: 1807 bytes --]

On Tue, 10 Jan 2023, 12:26 Jakub Jelinek via Libstdc++, <
libstdc++@gcc.gnu.org> wrote:

> On Tue, Jan 10, 2023 at 11:46:55AM +0000, Jonathan Wakely via Gcc-patches
> wrote:
> > Tested x86_64-linux. Pushed to trunk.
> >
> > -- >8 --
> >
> > The default constructor has a constraint that is always false if
> > arithmetic on size_t values promotes to int. Rewrite the constraint
> > exactly as written in the standard, which works correctly.
> >
> > libstdc++-v3/ChangeLog:
> >
> >       PR libstdc++/108221
> >       * include/std/span (span::span()): Un-simplify constraint to
> >       work for size_t of lesser rank than int.
> > ---
> >  libstdc++-v3/include/std/span | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/libstdc++-v3/include/std/span
> b/libstdc++-v3/include/std/span
> > index 251fed91abf..b336332b190 100644
> > --- a/libstdc++-v3/include/std/span
> > +++ b/libstdc++-v3/include/std/span
> > @@ -145,7 +145,7 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
> >
> >        constexpr
> >        span() noexcept
> > -      requires ((_Extent + 1u) <= 1u)
> > +      requires (_Extent == dynamic_extent || _Extent == 0)
> >        : _M_ptr(nullptr), _M_extent(0)
> >        { }
>
> If it would be C++23 only, you could use ((_Extent + 1uz) <= 1uz).
>

That would still promote _Extent and 1uz to int before the addition, and so
would not wrap to zero when _Extent == -1uz

size_t(_Extent + 1) <= 1 would work, but I'm just going to KISS and do what
the standard says.


As this is evaluated at compile time only, it is unfortunate it is
> 3 operations compared to former 2, but not a big deal.  If this was
> in code that would be emitted at runtime, GCC already optimizes
> (x == -1uz || x == 0)
> or
> (x == 0 || x == -1uz)
> to
> ((x + 1uz) <= 1uz)
>
>         Jakub
>
>