1berry/qtbase
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Only define QT_SUPPORTS_INT128 if type_traits work for them

Browse Source 
It turns out that libstdc++ in -ansi mode defines __SIZEOF_INT128__,
but not the corresponding specializations of <type_traits> and
<limits>. This has caused numerous pain points over time, esp. since
is_signed doesn't work, so qt_saturate e.g. can't be used on qint128
inputs e.g.

After another such issue (cmp_equal()), we decided to require
<type_traits> and <limits> support for extended integer types in the
stdlib, or we will disable Qt's support for it, too.

So check that if QT_SUPPORTS_INT128 is defined, is_signed_v,
is_integral_v and numeric_limits work, too. Disable QT_SUPPORTS_INT128
when compiling with libstdc++ in __STRICT_ANSI__ mode (-ansi or
-std=c++NN instead of (the default) -std=gnu++NN).

[ChangeLog][Potentially Source-Incompatible Changes] Qt's support for
128-bit integers (qint128/quint128) is now conditional on support for
these types from the Standard Library, in particular <type_traits> and
<limits>. Qt no longer tries to work around missing Standard Library
support. As a consequence, e.g. GCC -ansi and GCC -std=c++NN (instead
of -std=gnu++NN, the default) builds will now no longer support these
types.

Task-number: QTBUG-119901
Change-Id: I8529e72a52a2f5da0f469bae543688e18220255f
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
(cherry picked from commit 9bb2ab5978625eee96f703871d5eca8e54b31386)
Reviewed-by: Qt Cherry-pick Bot <cherrypick_bot@qt-project.org>
This commit is contained in:
Marc Mutz 2024-08-05 20:05:24 +02:00 committed by Qt Cherry-pick Bot
parent 073fed11e8
commit 7805b3c32f
2 changed files with 22 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
src/corelib/global/qtypes.h
View File

@ -16,6 +16,7 @@
// P1467 implementation - https://wg21.link/p1467
# include <stdfloat>
# endif // defined(__STDCPP_FLOAT16_T__) && __has_include(<stdfloat>)
# include <type_traits>
#else
# include <assert.h>
#endif
@ -67,6 +68,9 @@ typedef quint64 qulonglong;
# define QT_SUPPORTS_INT128 16
#elif defined(__SIZEOF_INT128__) && !defined(QT_NO_INT128)
# define QT_SUPPORTS_INT128 __SIZEOF_INT128__
# if defined(__GLIBCXX__) && defined(__STRICT_ANSI__) // -ansi/-std=c++NN instead of gnu++NN
# undef QT_SUPPORTS_INT128 // breaks <type_traits> on libstdc++
# endif
#else
# undef QT_SUPPORTS_INT128
#endif
@ -75,6 +79,11 @@ typedef quint64 qulonglong;
__extension__ typedef __int128_t qint128;
__extension__ typedef __uint128_t quint128;
#ifdef __cplusplus
static_assert(std::is_signed_v<qint128>,
"Qt requires <type_traits> and <limits> to work for q(u)int128.");
#endif
// limits:
# ifdef __cplusplus /* need to avoid c-style-casts in C++ mode */
# define QT_C_STYLE_CAST(type, x) static_cast<type>(x)

13
tests/auto/corelib/global/qglobal/tst_qglobal.cpp
View File

@ -16,6 +16,19 @@
#include <limits>
#include <QtCore/qxptype_traits.h>
// Check that <type_traits> works for q(u)int128; if any of these trigger,
// adjust the ifdef'ery in qtypes.h to exclude builds with broken lib support.
#ifdef QT_SUPPORTS_INT128
static_assert(std::is_signed_v<qint128>);
static_assert(std::is_integral_v<qint128>);
static_assert(std::is_integral_v<quint128>);
static_assert(std::numeric_limits<qint128>::is_signed);
static_assert(std::numeric_limits<qint128>::is_specialized);
static_assert(std::numeric_limits<quint128>::is_specialized);
static_assert((std::numeric_limits<qint128>::max)() == Q_INT128_MAX);
static_assert((std::numeric_limits<quint128>::max)() == Q_UINT128_MAX);
#endif // QT_SUPPORTS_INT128
template <typename T>
using AdlSwappableTest = decltype(swap(std::declval<T&>(), std::declval<T&>()));