A default-constructed static QBasicAtomicInt at function scope
will be dynamically initialized. It will still be zero-initialized,
but at least GCC adds guard variables for such objects.
When using aggregate initialization, the guard disappears.
Amends 04d6495bf7.
Change-Id: Id9335cffdd43094cafb231fdde2523d013abe1d6
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The library feature already covers this. As library depends on the
dlopen compile check, we can assume dlopen to be available, also on
vxworks.
Change-Id: Idcdb07ab4688c6158651d9a5ad5e2ba126bd7d9e
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
It is never written after initialization, and accessed concurrently
outside mutex protection, so make sure that it never changes by
marking it as const.
Change-Id: Ib28b2e8b3067d596a95d1699165a6ebfb15fe674
Reviewed-by: David Faure <david.faure@kdab.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Adding 0 each time will obviously not produce a new identifier each
time...
Also use static initialization for QBasicAtomicInt.
A default-constructed static QBasicAtomicInt at function scope
will be dynamically initialized. It will still be zero-initialized,
but at least GCC adds guard variables for such objects.
When using aggregate initialization, the guard disappears.
Amends 265db5ad9b.
Change-Id: Ia71290cf26c486dcbcc74381f12cd0c4712d6019
Reviewed-by: David Faure <david.faure@kdab.com>
Use F16C or ARM FP16 if available at compile time.
Configure check added because older clang compilers have F16C defines
and flags but not all the intrinsics.
Change-Id: I71f358b8fd003e70ab8fcf35097414591e485112
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
By making the destructor (usually the first non-inline, non-pure,
virtual function, and therefore the trigger for most compilers to emit
the vtable and type_info structures for the class in that TU)
out-of-line, vtables and, more importantly, type_info structures for
the class are pinned to a single TU. This prevents false-negative
dynamic_cast and catch evaluation.
Since the classes are already exported, users of these classes are
unaffected by the change, and since it's private API, we don't need to
avoid adding code to the out-of-line destructor until Qt 6.
While at it, de-inline also the empty default implementations of
virtual (non-dtor) functions.
Task-number: QTBUG-45582
Change-Id: I3e6f37eab1dee0db445f6c13638a43ca3bf6ac62
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
We require lambda support in the compiler since Qt 5.7, so drop
the guard macro.
Apart from MSVC 2013, all our supported compiler also support
char16_t:
- VxWorks uses GCC 4.8, which supports char16_t since 4.5.
- ICC supports char16_t since v12.1; we require v14+.
- ONX uses a GCC which supports char16_t in the compiler,
but is equipped with a stdlib that does not sport char16_t
support, which is why we revoke its Q_C_UNICODE_STRINGS in
qcompilerdetection.h. But we don't need stdlib support, we
only need the core language feature.
This is the only platform where this patch actually changes
something. It removes the, rather unfair, pessimization of
the platform which could have supported the static-storage-
duration implementation of QStringLiteral instead of the
fallback, which uses dynamic memory (QString::fromUtf8()).
- GCC and Clang support char16_t since 4.5 and 3.0, resp.,
which is far below our minimum compiler requirements in
effect since Qt 5.7.
- On Windows, MSVC supports char16_t since 2015 only, and we
still support 2013, but on Windows, wchar_t is the same size
as char16_t, so instead of u"str" we can fall back to L"str".
So simplify the implementation of QStringLiteral by assuming
that all these feature are present, adding noexcept to the
lambda to make noexcept() queries return true.
This allows us to guarantee:
[ChangeLog][QtCore] QStringLiteral is now guaranteed to be
evaluated at compile-time. The last platform (QNX) to use the
QString::fromUtf8() fallback has been ported to allocate the
string data statically, too.
Change-Id: I7920d7a77001e5c5550e7c7d57ceb7c51c9eb443
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
A variable of static storage duration that is not zero-initialized
takes up space in the DATA segment of the executable. By making the
counters start at zero and adding the initial value afterwards, we
move them over to the BSS segment, which does not take up space in
the executable.
Wrap atomics used across function boundaries into small functions,
to avoid code duplication and to increase readability.
Change-Id: Ida6ed316ecb8fe20da62a9577161349e14de5aed
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Systematic testing in tst_QStringApiSymmetry revealed a bug in
QStringRef::toLatin1(): a null input did not result in a null output,
but an empty one. This is fixed, for consistency with
QString::toLatin1(), and QString(Ref)::toUtf8(), which behaved
correctly already.
The same bug was found in QString(Ref)::toLocal8Bit(), which is
particularly hideous, as it's documented to fall back to toLatin1(),
which preserves null inputs. Fixed, too.
[ChangeLog][QtCore][QString] toLocal8Bit() now preserves nullness of
the input QString (outputs null QByteArray).
[ChangeLog][QtCore][QStringRef] toLocal8Bit() and toLatin1() now
preserve nullness of the input QStringRef (output null QByteArrays).
Change-Id: I7026211922c287e03d07e89edbad2987aa646e51
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
3e9db01a2e4be5034d10d6eeafd65d0a8609615b entered dev after the 5.8
branch.
Change-Id: I4139d5f93dcb4b429ae9fffd14a349e85e62a298
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
It was unnecessary, since we only cached the static types, which are all
generated by the macro anyway. The way it was implemented, this produced
data races that are strictly-speaking UB, even if all the threads were
writing the same values to the same data locations.
This commit changes a little the code to simplify, since we're changing
those lines anyway.
Task-number: QTBUG-58851
Change-Id: Idc5061f7145940f987dffffd14a30047846e3113
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
GHS compiler is not fully compliant with iec559.
Therefore we need to replace is_iec559 assertion
with separate checks to build quint16.
Change-Id: I88c57e394b8d4e7899ee7d4a13cbfbac9436b2fc
Reviewed-by: Rolland Dudemaine <rolland@ghs.com>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
If the QSemaphore::tryAcquire() call times out, we mustn't
touch *res, because there was no happens-before relation
established between *res = result in the lambda and our
returning *res;
Fix by returning a default-constructed hash in that case.
Add a strategic std::move().
The same problem exists in runOnAndroidThreadSync(), but
I have no idea how to solve it, because there the shared
object is the runnable itself.
Change-Id: I9a2c431144c169fbd545763555d96153143a11bf
Reviewed-by: BogDan Vatra <bogdan@kdab.com>
The code obtained an iterator into a QHash under mutex protection,
then dropped the lock, dereferenced the iterator several times and
only retook the lock to erase the element from the QHash.
This is very smelly. QHash provides no official iterator validity
guarantees, and the container isn't const, either (which would imply
thread-safety). In particular, the dereference into the container
outside the critical section is cause for concerns.
Simplify the code, removing any doubts about its race-freedom, by
taking the payload item out of the hash before dropping the lock, and
using only the local strong reference in the remainder of the
function.
The only other references to g_pendingPermissionRequests are
insertions with unique-by-construction keys in QtAndroidPrivate's
requestPermissions(), so there was no reason to keep the item in the
hash for the whole duration of the sendRequestPermissionsResult()
call.
Change-Id: I39fe0803b13b3046d1f0fd9c8e96c531406d57da
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Reviewed-by: BogDan Vatra <bogdan@kdab.com>
Solves a data race found by TSan.
Since thread and threadId are QAtomicPointer, I've removed the explicit
initialization in the QThreadData constructor
Task-number: QTBUG-58855
Change-Id: I4139d5f93dcb4b429ae9fffd14a34082f2683f76
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
1. Do not use Q_GLOBAL_STATIC to hold QAtomicInt or QMutex, use
file-static QBasicAtomicInt and QBasicMutex instead. They are
zero-initialized PODs.
2. Use only QMutexLocker to lock mutexes.
Also wrap the atomic counter into a next...() function, as done
elsewhere.
Change-Id: I4b14ac0de9d4cb6780b1f1372c2b5fc88e918e4c
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
$ git grep -we futexFlagSupport
src/corelib/thread/qmutex_linux.cpp:static QBasicAtomicInt futexFlagSupport = Q_BASIC_ATOMIC_INITIALIZER(-1);
The last user of this variable was removed in 9ef59b5.
Change-Id: I818a13a481ad25baa5ff7d389a737b8801adcfcc
Reviewed-by: David Faure <david.faure@kdab.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
All implementations of QCollator::sortKey() can benefit from
moving their CollatorKeyType into the QCollatorSortKeyPrivate
on construction.
So make the QCollatorSortKeyPrivate ctor a perfect forwarder
for its m_key member, and add std::move() calls where they
were missing (in all but one case, lvalues were passed).
Make the ctor explicit, as it should have been from the
beginning.
Change-Id: I2a1cdda5fd23990ace019b963df895c621a1fa85
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
QThreadPool maintains three containers of QThreadPoolThread*:
- allThreads, a QSet
- waitingThreads, a QQueue
- expiredThreads, also a QQueue
None of the operations on allThreads make use of QSets fast lookup.
The only functions called on it are isEmpty(), count(), insert(),
and swap().
Since therefore QSet adds nothing but overhead, causes indeterminism
(e.g. when deleting threads in Private::reset()) and code bloat, use
the same container for allThreads that underlies QQueue: QList.
Port insert() to append(). Add an assert to verify that we're not
running into an ABA problem here (but this should never fire, since
we're never deleting threads except in Private::reset(), where we
do remove them from allThreads), just in case.
Saves ~0.5KiB in text size on optimized Linux AMD64 GCC 7.0 builds.
Change-Id: I53a4d5ef2c204420f7c8852f1e72ab3d6ea43d08
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Add proper support for 'so' and 'bundle' suffixes. Qt wrongly assumes
.so libraries are not versioned on Apple platforms, which is
wrong. Also, the shared library .bundle which is what Apple recommends
instead of .so, are also versioned (not to be confound with the
different Core Foundation bundles, which are directory hierarchy).
For more info, see http://docstore.mik.ua/orelly/unix3/mac/ch05_03.htm.
Especially the part that reads:
"Loadable modules, called bundles in Mac OS X, have the file type
MH_BUNDLE. Most Unix-based software ports usually produce bundles with a
.so extension, for the sake of consistency across platforms. Although
Apple recommends giving bundles a .bundle extension, it isn't
mandatory."
Task-number: QTBUG-50446
Change-Id: Iacd5136397a12d65d83821434f332eb602550b4b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
And if __AES__ or __PCLMUL__ is defined, then we expect wmmintrin.h to
exist.
Change-Id: I445bb15619f6401494e8fffd149db77dc513e071
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
We have observed that MSVC 2013 gets lost with the overloads that this
header adds, causing compilation bugs. This is believed to be a compiler
bug, but it's not something we can work around.
Task-number: QTBUG-58555
Change-Id: I536c32a88bff44dab37afffd14a1bad1d31dc16d
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
Reviewed-by: Glen Mabey <Glen.Mabey@swri.org>
Since 5.7, QIODevice::peek() implementation is based on transaction
mechanism. While technically it's correct, seeking backward on a
buffered random-access device clears the internal buffer that affects
the performance of reading.
To solve the problem, this patch implements peek mode directly inside
the reading procedure.
Task-number: QTBUG-56032
Change-Id: Ic5269f76e44c491a0309e13aba87fa7cf7b9259f
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Move device-dependent part of the code into the private function for
further reusing by peek() procedure.
Task-number: QTBUG-56032
Change-Id: Iedceafe4b0bab109ca5c64ad274d779efe87c27b
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
We're already using it, for example in qpaintengine_pic.cpp:502:
d->s << p << ti.text() << fnt << ti.renderFlags() << [...]
^
[ChangeLog][QtCore][QFlags] Added operator<< and operator>> for
streaming QFlags into and out of QDataStreams.
Change-Id: I33dc971f005a4848bb8ffffd1478e79d6102d1bc
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
the file contains no code.
this avoids complaints from ar/ranlib in static/bootstrapped builds.
Change-Id: Iee22ffc61a5f9ea8c25f5455b7e8e017ac521624
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Since they are all used in a typical application, this reduces the
number of memory allocations (thus, the overhead) as well as the
state-keeping in the libc atexit() functions.
Change-Id: Ifaee7464122d402991b6fffd14a0e59457ad9cb7
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
If these lists weren't created in the first place, then they are empty.
We don't need to create it in order to conclude that. Unlike most
Q_GLOBAL_STATICS, these are almost never used and yet they were
always created due to where they were checked.
Since we're calling exists() before, there are two consequences: first,
since the list already exists, we're not allocating memory so it cannot
throw std::bad_alloc when being accessed. Second, since we've just
checked it exists, we can use QGlobalStatic's operator*(), which is
slightly faster than operator()(). The weird &(*list) syntax is only to
avoid changing the rest of the code that used a pointer
Change-Id: Ifaee7464122d402991b6fffd14a0e44f533dc3d9
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Seems to be a write-only variable and QThread::currentThreadId has no
side-effects.
Change-Id: Ifaee7464122d402991b6fffd14a0c8666968dfe4
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Give QCollator access to QLocalePrivate::bcp47Name(), to avoid
both the latin-1 -> UTF-16 conversion in QLocale::bcp47Name(),
as well all as
- the replace('-', '_').toLatin1() call in ICU
- the toLocal8Bit() call in macOS
- the toUtf8() call in Windows
implementations of QCollatorPrivate::init().
This is safe, since, according to https://tools.ietf.org/html/bcp47,
a BCP47 name only contains US-ASCII (ALPHA used, which is defined by
https://tools.ietf.org/html/rfc5234 to be [a-zA-Z] only).
Change-Id: Id56befb1b5a7983494d848cdabf7ebeda377cf9f
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
When a non-QString-backed mode (via the (QChar*, int) ctor) was added
for Qt 4.5, the author forgot to adjust the setCaseSensitivity()
function. It still uses q_pattern instead of (p.uc, p.len) as the
pattern for which to create the skip-table. Since there is no
setPattern() overload for this mode, the correctness of the matcher is
not harmed by this, but its performance degrades to that of a linear
scan: the skip-table, being filled from an empty pattern, will be
all-zeros, sending bm_find() into the 'possible match' case at every
character.
Since matching is still correct, but slow, it's not possible to write
a test for this. I did, however, leave my attempts in the auto-test,
for when we add QStringView overloads of setPattern() which will then
be able to expose the bug.
Change-Id: I7b803e8624b0352a0a974900affbbfc0c260d93b
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
The intention when using an overload that takes (const QChar*, int)
instead of a QString is probably to avoid creating a QString in the
first place. If the implementation stabbs the user in the back by
internally creating a QString just so it can call QString::compare()
on it, then that is rather unacceptable.
QCollator is already a friend of QString, so, absent QStringView,
resolve to calling compare_helper().
Change-Id: Ia7de1a095e3e310bd1b9957ed67291cc9cc95b32
Reviewed-by: David Faure <david.faure@kdab.com>
The key is immutable except for assignment and swap,
so don't run the risk of auto-detaching and use the
explicit-detach version of QSharedDataPointer.
Change-Id: Ib2cfe5981e6dfe375d6208289ff58247ef9d4870
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
... by not creating three temporary QVectors just to concatenate them.
There's no QVectorBuilder, so what works well with QStrings doesn't
work well at all with QVectors. The chaining of op+ causes three
temporary QVectors to be created and thrown away.
Instead, use clear() (which preserves the vector's capacity these days),
followed by four op+=.
Change-Id: I300bd35544ea41037d28db0f48f210c33c826b85
Reviewed-by: Milian Wolff <milian.wolff@kdab.com>
Fix warnings:
src/network/access/qhstspolicy.cpp:158: warning: Undocumented parameter 'expiry' in QHstsPolicy::setExpiry()
src/network/access/qhstspolicy.cpp:178: warning: Undocumented parameter 'include' in QHstsPolicy::setIncludesSubDomains()
src/network/access/qhstspolicy.cpp:178: warning: Can't link to 'includeSubdomains()'
src/widgets/kernel/qsizepolicy.cpp:402: warning: Cannot find 'transposed(...)' in '\fn' QSizePolicy QSizePolicy::transposed()
src/widgets/kernel/qsizepolicy.h:165: warning: No documentation for 'QSizePolicy::transposed()'
src/widgets/util/qundostack.cpp:164: warning: Undocumented parameter 'obsolete' in QUndoCommand::setObsolete()
Fix the signature of the QStaticByteArrayMatcher member functions. The warnings
src/corelib/tools/qbytearraymatcher.cpp:372: warning: Cannot find 'indexIn(...)' in '\fn' QStaticByteArrayMatcher::indexIn(const char *haystack, int hlen, int from)
src/corelib/tools/qbytearraymatcher.cpp:382: warning: Cannot find 'indexIn(...)' in '\fn' QStaticByteArrayMatcher::indexIn(const QByteArray &haystack, int from)
remain, though since apparently qdoc is thrown off by the constructor.
Change-Id: I157359a881ff3fbc80d2eeb52dd5c27249d009c4
Reviewed-by: Jesus Fernandez <Jesus.Fernandez@qt.io>
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
This is private API, and used nowhere else except in qcollator*,
so don't export these classes.
Change-Id: I217fde97c60ab0a3e19774ed5a6eed8b156fff1d
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
When a QFile object is reused, the atEnd() method may return incorrect
values. The reason for this is that QFileDevicePrivate::cachedSize is
not cleared. Setting cachedSize = 0 in the close() method fixes this issue.
Task-number: QTBUG-57698
Change-Id: I828a2cf844e98d581098f2c781fa47d2cd3275ce
Reviewed-by: Alex Trotsenko <alex1973tr@gmail.com>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
QDateTimeEdit (QtWidgets) inherits its Private class from
QDateTimeParser.
By making the destructor (usually the first non-inline, non-pure,
virtual function, and therefore the trigger for most compilers to
emit the vtable and type_info structures for the class in that TU)
out-of-line, vtables and, more importantly, type_info structures for
the class are pinned to a single TU. This prevents false negative
dynamic_cast and catch evaluation.
Since the class is already exported, users of these classes are
unaffected by the change, and since it's private API, we don't need
to avoid adding code to the out-of-line destructor until Qt 6.
Task-number: QTBUG-45582
Change-Id: Id57289a7eff13564dddeca0af1c45a6180c36c48
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Liang Qi
Marc Mutz
Ulf Hermann
Allan Sandfeld Jensen
Thiago Macieira
Frederik Schwarzer
Samuel Gaist
Kimmo Ollila
Heikki Haveri
Friedemann Kleint
Oswald Buddenhagen
Filipe Azevedo
Alex Trotsenko
Aleksey Lysenko