1berry/qtbase
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
qtbase/src/corelib/tools/qcontiguouscache.cpp
Marc Mutz 49db71bb19 QContiguousCache: drag the initialization of atomic ref behind ABI boundary 
All callers of QContiguousCache<T>::allocateData() followed the call
with a storeRelaxed(1) to the ref member. So we can just drag that
into the function itself.

Next, it's UB to storeRelaxed() into a default-constructed std::atomic
(and, therefore, into a QBasicAtomicInt), because until C++17
(inclusive) you're supposed to use std::atomic_init to assign the
first (and only the first) value to a default-constructed
std::atomic. QBasicAtomic doesn't have API for that, so you can never
assign anything to a default-constructed QBasicAtomic.

To fix, use placement new to be able to create a QBasicAtomic directly
with an initial value (replacing QBasicAtomic with QAtomic wouldn't
help here, either, since a malloc doesn't run ctors).

A proper fix has to wait until we can depend on C++20's atomic_ref,
which decouples the underlying type from the atomic operations
performed on it, letting us depend on malloc's zero-initialization
of an int member properly initializing it even for a following atomic
operation on it.

Task-number: QTBUG-137465
Pick-to: 6.10 6.9 6.8 6.5
Change-Id: Ic22d0766bcffb967a86c8ec28b63ee480aebd4a0
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
2025-06-09 20:36:21 +00:00

437 lines
14 KiB
C++
Raw Blame History

// Copyright (C) 2016 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only
#include "qcontiguouscache.h"
#ifdef QT_QCONTIGUOUSCACHE_DEBUG
#include <QDebug>
#endif
#include <QtCore/qmalloc.h>
QT_BEGIN_NAMESPACE
#ifdef QT_QCONTIGUOUSCACHE_DEBUG
void QContiguousCacheData::dump() const
{
qDebug() << "capacity:" << alloc;
qDebug() << "count:" << count;
qDebug() << "start:" << start;
qDebug() << "offset:" << offset;
}
#endif
QContiguousCacheData *QContiguousCacheData::allocateData(qsizetype size, qsizetype alignment)
{
void *mem = qMallocAligned(size_t(size), size_t(alignment));
return new (mem) QContiguousCacheData{/*ref=*/1, 0, 0, 0, 0};
}
void QContiguousCacheData::freeData(QContiguousCacheData *data)
{
qFreeAligned(data);
}
/*! \class QContiguousCache
\inmodule QtCore
\brief The QContiguousCache class is a template class that provides a contiguous cache.
\ingroup tools
\ingroup shared
\reentrant
\since 4.6
The QContiguousCache class provides an efficient way of caching items for
display in a user interface view. Unlike QCache, it adds a restriction
that elements within the cache are contiguous. This has the advantage
of matching how user interface views most commonly request data, as
a set of rows localized around the current scrolled position. This
restriction allows the cache to consume less memory and processor
cycles than QCache.
QContiguousCache operates on a fixed capacity, set with setCapacity() or
passed as a parameter to the constructor. This capacity is the upper bound
on memory usage by the cache itself, not including the memory allocated by
the elements themselves. Note that a cache with a capacity of zero (the
default) means no items will be stored: the insert(), append() and
prepend() operations will effectively be no-ops. Therefore, it's important
to set the capacity to a reasonable value before adding items to the cache.
The simplest way of using a contiguous cache is to use the append()
and prepend().
\snippet code/src_corelib_tools_qcontiguouscache.cpp 0
If the cache is full then the item at the opposite end of the cache from
where the new item is appended or prepended will be removed.
This usage can be further optimized by using the insert() function
in the case where the requested row is a long way from the currently cached
items. If there is a gap between where the new item is inserted and the currently
cached items then the existing cached items are first removed to retain
the contiguous nature of the cache. Hence it is important to take some care then
when using insert() in order to avoid unwanted clearing of the cache.
The range of valid indexes for the QContiguousCache class are from
0 to INT_MAX. Calling prepend() such that the first index would become less
than 0 or append() such that the last index would become greater
than INT_MAX can result in the indexes of the cache being invalid.
When the cache indexes are invalid it is important to call
normalizeIndexes() before calling any of containsIndex(), firstIndex(),
lastIndex(), at() or \l{QContiguousCache::operator[]()}{operator[]()}.
Calling these functions when the cache has invalid indexes will result in
undefined behavior. The indexes can be checked by using areIndexesValid()
In most cases the indexes will not exceed 0 to INT_MAX, and
normalizeIndexes() will not need to be used.
See the \l{Contiguous Cache Example}{Contiguous Cache} example.
*/
/*! \fn template<typename T> QContiguousCache<T>::QContiguousCache(qsizetype capacity)
Constructs a cache with the given \a capacity.
\sa setCapacity()
*/
/*! \fn template<typename T> QContiguousCache<T>::QContiguousCache(const QContiguousCache<T> &other)
Constructs a copy of \a other.
This operation takes \l{constant time}, because QContiguousCache is
\l{implicitly shared}. This makes returning a QContiguousCache from a
function very fast. If a shared instance is modified, it will be
copied (copy-on-write), and that takes \l{linear time}.
\sa operator=()
*/
/*! \fn template<typename T> QContiguousCache<T>::~QContiguousCache()
Destroys the cache.
*/
/*! \fn template<typename T> void QContiguousCache<T>::detach()
\internal
*/
/*! \fn template<typename T> bool QContiguousCache<T>::isDetached() const
\internal
*/
/*! \fn template<typename T> void QContiguousCache<T>::setSharable(bool sharable)
\internal
*/
/*! \typedef QContiguousCache::value_type
\internal
*/
/*! \typedef QContiguousCache::pointer
\internal
*/
/*! \typedef QContiguousCache::const_pointer
\internal
*/
/*! \typedef QContiguousCache::reference
\internal
*/
/*! \typedef QContiguousCache::const_reference
\internal
*/
/*! \typedef QContiguousCache::difference_type
\internal
*/
/*! \typedef QContiguousCache::size_type
\internal
*/
/*! \fn template<typename T> QContiguousCache<T> &QContiguousCache<T>::operator=(const QContiguousCache<T> &other)
Assigns \a other to this cache and returns a reference to this cache.
*/
/*!
\fn template<typename T> QContiguousCache<T> &QContiguousCache<T>::operator=(QContiguousCache<T> &&other)
Move-assigns \a other to this QContiguousCache instance.
\since 5.2
*/
/*! \fn template<typename T> void QContiguousCache<T>::swap(QContiguousCache<T> &other)
\since 4.8
\memberswap{cache}
*/
/*! \fn template<typename T> bool QContiguousCache<T>::operator==(const QContiguousCache<T> &other) const
Returns \c true if \a other is equal to this cache; otherwise returns \c false.
Two caches are considered equal if they contain the same values at the same
indexes. This function requires the value type to implement the \c operator==().
\sa operator!=()
*/
/*! \fn template<typename T> bool QContiguousCache<T>::operator!=(const QContiguousCache<T> &other) const
Returns \c true if \a other is not equal to this cache; otherwise
returns \c false.
Two caches are considered equal if they contain the same values at the same
indexes. This function requires the value type to implement the \c operator==().
\sa operator==()
*/
/*! \fn template<typename T> qsizetype QContiguousCache<T>::capacity() const
Returns the number of items the cache can store before it is full.
When a cache contains a number of items equal to its capacity, adding new
items will cause items farthest from the added item to be removed.
\sa setCapacity(), size()
*/
/*! \fn template<typename T> qsizetype QContiguousCache<T>::count() const
Same as size().
*/
/*! \fn template<typename T> qsizetype QContiguousCache<T>::size() const
Returns the number of items contained within the cache.
\sa capacity()
*/
/*! \fn template<typename T> bool QContiguousCache<T>::isEmpty() const
Returns \c true if no items are stored within the cache.
\sa size(), capacity()
*/
/*! \fn template<typename T> bool QContiguousCache<T>::isFull() const
Returns \c true if the number of items stored within the cache is equal
to the capacity of the cache.
\sa size(), capacity()
*/
/*! \fn template<typename T> qsizetype QContiguousCache<T>::available() const
Returns the number of items that can be added to the cache before it becomes full.
\sa size(), capacity(), isFull()
*/
/*! \fn template<typename T> void QContiguousCache<T>::clear()
Removes all items from the cache. The capacity is unchanged.
*/
/*! \fn template<typename T> void QContiguousCache<T>::setCapacity(qsizetype size)
Sets the capacity of the cache to the given \a size. A cache can hold a
number of items equal to its capacity. When inserting, appending or prepending
items to the cache, if the cache is already full then the item farthest from
the added item will be removed.
If the given \a size is smaller than the current count of items in the cache
then only the last \a size items from the cache will remain.
\sa capacity(), isFull()
*/
/*! \fn template<typename T> const T &QContiguousCache<T>::at(qsizetype i) const
Returns the item at index position \a i in the cache. \a i must
be a valid index position in the cache (i.e, firstIndex() <= \a i <= lastIndex()).
The indexes in the cache refer to the number of positions the item is from the
first item appended into the cache. That is to say a cache with a capacity of
100, that has had 150 items appended will have a valid index range of
50 to 149. This allows inserting and retrieving items into the cache based
on a theoretical infinite list
\sa firstIndex(), lastIndex(), insert(), operator[]()
*/
/*! \fn template<typename T> T &QContiguousCache<T>::operator[](qsizetype i)
Returns the item at index position \a i as a modifiable reference. If
the cache does not contain an item at the given index position \a i
then it will first insert an empty item at that position.
In most cases it is better to use either at() or insert().
\note This non-const overload of operator[] requires QContiguousCache
to make a deep copy. Use at() for read-only access to a non-const
QContiguousCache.
\sa insert(), at()
*/
/*! \fn template<typename T> const T &QContiguousCache<T>::operator[](qsizetype i) const
\overload
Same as at(\a i).
*/
/*! \fn template<typename T> void QContiguousCache<T>::append(const T &value)
Inserts \a value at the end of the cache. If the cache is already full
the item at the start of the cache will be removed.
\sa prepend(), insert(), isFull()
*/
/*! \fn template<typename T> void QContiguousCache<T>::prepend(const T &value)
Inserts \a value at the start of the cache. If the cache is already full
the item at the end of the cache will be removed.
\sa append(), insert(), isFull()
*/
/*! \fn template<typename T> void QContiguousCache<T>::insert(qsizetype i, const T &value)
Inserts the \a value at the index position \a i. If the cache already contains
an item at \a i then that value is replaced. If \a i is either one more than
lastIndex() or one less than firstIndex() it is the equivalent to an append()
or a prepend().
If the given index \a i is not within the current range of the cache nor adjacent
to the bounds of the cache's index range, the cache is first cleared before
inserting the item. At this point the cache will have a size of 1. It is
worthwhile taking effort to insert items in an order that starts adjacent
to the current index range for the cache.
The range of valid indexes for the QContiguousCache class are from
0 to INT_MAX. Inserting outside of this range has undefined behavior.
\sa prepend(), append(), isFull(), firstIndex(), lastIndex()
*/
/*! \fn template<typename T> bool QContiguousCache<T>::containsIndex(qsizetype i) const
Returns \c true if the cache's index range includes the given index \a i.
\sa firstIndex(), lastIndex()
*/
/*! \fn template<typename T> qsizetype QContiguousCache<T>::firstIndex() const
Returns the first valid index in the cache. The index will be invalid if the
cache is empty.
\sa capacity(), size(), lastIndex()
*/
/*! \fn template<typename T> qsizetype QContiguousCache<T>::lastIndex() const
Returns the last valid index in the cache. The index will be invalid if the cache is empty.
\sa capacity(), size(), firstIndex()
*/
/*! \fn template<typename T> T &QContiguousCache<T>::first()
Returns a reference to the first item in the cache. This function
assumes that the cache isn't empty.
\sa last(), isEmpty()
*/
/*! \fn template<typename T> T &QContiguousCache<T>::last()
Returns a reference to the last item in the cache. This function
assumes that the cache isn't empty.
\sa first(), isEmpty()
*/
/*! \fn template<typename T> const T& QContiguousCache<T>::first() const
\overload
*/
/*! \fn template<typename T> const T& QContiguousCache<T>::last() const
\overload
*/
/*! \fn template<typename T> void QContiguousCache<T>::removeFirst()
Removes the first item from the cache. This function assumes that
the cache isn't empty.
\sa removeLast()
*/
/*! \fn template<typename T> void QContiguousCache<T>::removeLast()
Removes the last item from the cache. This function assumes that
the cache isn't empty.
\sa removeFirst()
*/
/*! \fn template<typename T> T QContiguousCache<T>::takeFirst()
Removes the first item in the cache and returns it. This function
assumes that the cache isn't empty.
If you don't use the return value, removeFirst() is more efficient.
\sa takeLast(), removeFirst()
*/
/*! \fn template<typename T> T QContiguousCache<T>::takeLast()
Removes the last item in the cache and returns it. This function
assumes that the cache isn't empty.
If you don't use the return value, removeLast() is more efficient.
\sa takeFirst(), removeLast()
*/
/*! \fn template<typename T> void QContiguousCache<T>::normalizeIndexes()
Moves the first index and last index of the cache
such that they point to valid indexes. The function does not modify
the contents of the cache or the ordering of elements within the cache.
It is provided so that index overflows can be corrected when using the
cache as a circular buffer.
\snippet code/src_corelib_tools_qcontiguouscache.cpp 1
\sa areIndexesValid(), append(), prepend()
*/
/*! \fn template<typename T> bool QContiguousCache<T>::areIndexesValid() const
Returns whether the indexes for items stored in the cache are valid.
Indexes can become invalid if items are appended after the index position
INT_MAX or prepended before the index position 0. This is only expected
to occur in very long lived circular buffer style usage of the
contiguous cache. Indexes can be made valid again by calling
normalizeIndexes().
\sa normalizeIndexes(), append(), prepend()
*/
QT_END_NAMESPACE
Reference in New Issue View Git Blame Copy Permalink