On 12/17/25 22:11, Jonathan Wakely wrote:
On Wed, 17 Dec 2025 at 21:10, Jonathan Wakely <[email protected]> wrote:

On Wed, 17 Dec 2025 at 20:52, Luc Grosheintz <[email protected]> wrote:

On Solaris, same_as<int8_t, char> is true. Therefore, int8_t isn't a
valid IndexType, because char is neither a signed nor an unsigned
integer type.

This commit fixes the tests by avoiding int8_t (and uint8_t) by using
'signed char' (and 'unsigned char').

         PR libstdc++/123176

libstdc++-v3/ChangeLog:

         * 
testsuite/23_containers/mdspan/submdspan/submdspan_canonicalize_slices_neg.cc: 
Avoid
         int8_t with signed char.

Signed-off-by: Luc Grosheintz <[email protected]>
---

This fix is purely based on a theoretical understanding of the compiler
messages reported in the bug report. Unfortunately, I don't have access
to a Solaris machine; and therefore am not able to test on Solaris. I
have tested on linux-x86_64.

I've CC'ed Rainer Orth who reported the issue, maybe Rainer's willing
and has time to test it on Solaris.

  .../submdspan_canonicalize_slices_neg.cc      | 32 +++++++++++--------
  1 file changed, 18 insertions(+), 14 deletions(-)

diff --git 
a/libstdc++-v3/testsuite/23_containers/mdspan/submdspan/submdspan_canonicalize_slices_neg.cc
 
b/libstdc++-v3/testsuite/23_containers/mdspan/submdspan/submdspan_canonicalize_slices_neg.cc
index 94bca183aa3..0098547d750 100644
--- 
a/libstdc++-v3/testsuite/23_containers/mdspan/submdspan/submdspan_canonicalize_slices_neg.cc
+++ 
b/libstdc++-v3/testsuite/23_containers/mdspan/submdspan/submdspan_canonicalize_slices_neg.cc
@@ -8,9 +8,12 @@ constexpr size_t dyn = std::dynamic_extent;
  constexpr auto dyn_empty = std::extents<int32_t, dyn>{0};
  constexpr auto sta_empty = std::extents<uint32_t, 0>{};

-constexpr auto dyn_uexts = std::extents<uint8_t, dyn>{5};
+// On Solaris same_as<int8_t, char> is true. Therefore, int8_t is not a signed
+// or unsigned integer type and hence not a valid IndexType. We'll use signed
+// char and unsigned char for int8_t throughout.
+constexpr auto dyn_uexts = std::extents<unsigned char, dyn>{5};
  constexpr auto sta_uexts = std::extents<uint16_t, 5>{5};
-constexpr auto dyn_sexts = std::extents<int8_t, dyn>{5};
+constexpr auto dyn_sexts = std::extents<signed char, dyn>{5};
  constexpr auto sta_sexts = std::extents<int16_t, 5>{5};

  constexpr bool
@@ -69,7 +72,7 @@ template<typename Offset, typename Extent, typename Stride, 
typename Extents>
      return true;
    }

-constexpr auto i8_1 = int8_t{1};
+constexpr auto i8_1 = (signed char){1};

  static_assert(test_under2(-i8_1, 0, 1, dyn_uexts));   // { dg-error "expansion 
of" }
  static_assert(test_under2(0, -i8_1, 1, dyn_uexts));   // { dg-error "expansion 
of" }
@@ -84,7 +87,7 @@ static_assert(test_under2(-i8_1, 0, 1, sta_sexts));   // { 
dg-error "expansion o
  static_assert(test_under2(0, -i8_1, 1, sta_sexts));   // { dg-error "expansion 
of" }
  static_assert(test_under2(0, 1, -i8_1, sta_sexts));   // { dg-error "expansion 
of" }

-constexpr auto c_i8_m1 = std::cw<int8_t{-1}>;
+constexpr auto c_i8_m1 = std::cw<(signed char){-1}>;
  constexpr auto c_i16_m1 = std::cw<int16_t{-1}>;
  constexpr auto c_i64_m1 = std::cw<int64_t{-1}>;

@@ -109,8 +112,8 @@ template<typename Offset, typename Extent, typename Stride, 
typename Extents>
      return true;
    }

-constexpr auto i8_6 = int8_t{6};
-constexpr auto c_i8_6 = std::cw<int8_t{6}>;
+constexpr auto i8_6 = (signed char){6};
+constexpr auto c_i8_6 = std::cw<(signed char){6}>;

Actually all these braces should be removed, so just:

(signed char){6} not (signed char){6}, otherwise it's a C99 compound
literal, which is not valid in C++ and will warn with -pedantic.

Gah, pasted the wrong thing. I mean:
(signed char)6 not (signed char){6}


I had to remove `-pedantic` due to:

libstdc++-v3/testsuite/23_containers/mdspan/extents/ctor_ints.cc:20: warning: 
ISO C++ does not support '__int128' for 'type name' [-Wpedantic]

--------------------

Regarding your comment about fixing `int8_t` on Solaris. If we only fix the
testsuite, then code that uses

  std::extents<int8_t>

will fail to compile on Solaris. On the one hand: that's correct, because
int8_t on Solaris isn't a signed or unsigned integer type. On the other
hand it's wrong because the standard states that int8_t is.

What we could consider is implementing the requirement that something
is a valid IndexType in mdspan as:

    __is_signed_or_unsigned_integer<_IndexType>::value
      || same_as<_IndexType, int8_t>
      || same_as<_IndexType, uint8_t>;

we'd need to be more careful, because we can't blindly assume that
int8_t exists. The idea is that I can't remember anything in the
standard that requires us to fail if index_type == char and I can't
remember anything that would make 'char' problematic. Hence, we could
be lenient and allow char or int8_t.

I don't know if this is a good idea.




  constexpr auto c2 = std::cw<2>;
  constexpr auto c4 = std::cw<4>;

@@ -155,21 +158,22 @@ static_assert(test_over2(c2, c4, 1, sta_sexts));     // { 
dg-error "expansion of
  constexpr bool
  test_overflow1(auto o, auto e)
  {
-  auto exts = std::extents<uint8_t, dyn>{255};
+  auto exts = std::extents<unsigned char, dyn>{std::numeric_limits<unsigned 
char>::max()};
    auto slice = std::strided_slice{o, e, 1};
    std::submdspan_canonicalize_slices(exts, slice);
    return true;
  }

-static_assert(test_overflow1(128, 128));                    // { dg-error 
"expansion of" }
-static_assert(test_overflow1(std::cw<128>, 128));           // { dg-error 
"expansion of" }
-static_assert(test_overflow1(128, std::cw<128>));           // { dg-error 
"expansion of" }
-static_assert(test_overflow1(std::cw<128>, std::cw<128>));  // { dg-error 
"expansion of" }
+constexpr int half_max = std::numeric_limits<unsigned char>::max() / 2 + 1;
+static_assert(test_overflow1(half_max, half_max));                    // { dg-error 
"expansion of" }
+static_assert(test_overflow1(std::cw<half_max>, half_max));           // { dg-error 
"expansion of" }
+static_assert(test_overflow1(half_max, std::cw<half_max>));           // { dg-error 
"expansion of" }
+static_assert(test_overflow1(std::cw<half_max>, std::cw<half_max>));  // { dg-error 
"expansion of" }

  constexpr bool
  test_overflow2(auto b, auto e)
  {
-  auto exts = std::extents<uint8_t, dyn>{255};
+  auto exts = std::extents<unsigned char, dyn>{std::numeric_limits<unsigned 
char>::max()};
    auto slice = std::pair{b, e};
    std::submdspan_canonicalize_slices(exts, slice);
    return true;
@@ -180,8 +184,8 @@ static_assert(test_overflow2(std::cw<5>, 4));           // { 
dg-error "expansion
  static_assert(test_overflow2(5, std::cw<4>));           // { dg-error "expansion 
of" }
  static_assert(test_overflow2(std::cw<5>, std::cw<4>));  // { dg-error "expansion 
of" }

-constexpr auto u8_4 = uint8_t{4};
-constexpr auto u8_5 = uint8_t{5};
+constexpr auto u8_4 = (unsigned char){4};
+constexpr auto u8_5 = (unsigned char){5};
  static_assert(test_overflow2(u8_5, u8_4));                    // { dg-error 
"expansion of" }
  static_assert(test_overflow2(std::cw<u8_5>, u8_4));           // { dg-error 
"expansion of" }
  static_assert(test_overflow2(u8_5, std::cw<u8_4>));           // { dg-error 
"expansion of" }
--
2.52.0


Reply via email to