The ability to set a container to be unsharable has very little use and
it costs us an extra conditional for every refcount up and possibly
down.
This change is a no-op for current Qt 5. It shuffles a few things around
just so Qt can compile if you define QT_NO_UNSHARABLE_CONTAINERS. That
is done to ease the fixing of the code in Qt 6 and to make my life
easier: I'll keep that defined in my local Qt build so I can catch any
misuses of this deprecated API.
The newly deprecated methods are not marked QT_DEPRECATED because the
bootstrapped tools wouldn't build -- they're built with QT_NO_DEPRECATED
defined, which causes build errors.
[ChangeLog][QtCore] The setSharable() and isSharable() functions in Qt
containers has been deprecated and will be removed in Qt 6. New
applications should not use this feature, while old applications that
may be using this (undocumented) feature should port away from it.
Discussed-on: http://lists.qt-project.org/pipermail/development/2014-February/015724.html
Change-Id: I789771743dcaed6a43eccd99382f8b3ffa61e479
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Unit-test this by making the QList, QVector, QHash and QMap unit tests
be duplicated under strict-iterator mode. There's no test for
QLinkedList.
The tst_Collections test does not compile under strict-iterator
mode. It generated over 15000 errors when I tried.
The strict iterators required a small change: the difference_type
typedef needs to match the operators that get distances
(operator-(iterator)) and move the iterator around (+, -, +=, -=, etc.).
Task-number: QTBUG-29608
Change-Id: I834873934c51d0f139a994cd395818da4ec997e2
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
Reviewed-by: Jason McDonald <macadder1@gmail.com>
The support for QArrayData variadic arguments without C++11 for GCC
has been removed in commit 69478da0f0 . Change the autotest to reflect
that, too.
Change-Id: I40468f5d67cb2db553fd7a7d5b604f46403ac538
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The address behind a string doesn't point to a string.
Change-Id: Ic54f652ae781fea278f60cc49d219c1c610ba29f
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
Change copyrights and license headers from Nokia to Digia
Change-Id: If1cc974286d29fd01ec6c19dd4719a67f4c3f00e
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Sergio Ahumada <sergio.ahumada@digia.com>
QArrayData can point to data it does not own (cf. fromRawData()), which
shouldn't be modified. Not even upon destruction, as this data can live
in Read-Only memory or be otherwise shared outside the QArrayData realm.
Change-Id: I8bdf3050a17802fb003b77d5f543fe31769a7710
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Andreas Hartmetz <ahartmetz@gmail.com>
Doing element-wise insertions for the full range of the test made
testing under valgrind extremely slow. When a reallocation is detected
we now resize() the container close to capacity(), while verifying this
doesn't unnecessarily re-allocate either.
Change-Id: Idf7015cf390e366fe444e7ca14c904a2d54ff48b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Qt 5.0 beta requires changing the default to the 5.0 API, disabling
the deprecated code. However, tests should test (and often do) the
compatibility API too, so turn it back on.
Task-number: QTBUG-25053
Change-Id: I8129c3ef3cb58541c95a32d083850d9e7f768927
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Most fixes are simple and quite obvious. The ones more involved are
the ones to QArrayData, which had probably not been compiled with
strict iterators thus far.
Change-Id: Ic4ff84c34fd9a04fd686fecaa98149b1c47c9346
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
Initialiser lists were not tested before in the QVector rewrite, so
the older malloc call was left behind.
Also, std::initializer_list has const iterators returning const data
and broke the build in a few places where const qualifiers were
missing.
Change-Id: I3c04e58361989aa7438621cda63c7df457d7dad8
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
While QArrayDataPointer offers generic detach() functionality, this is
only useful for operations that may modify data, but don't otherwise
affect the container itself, such as non-const iteration, front() and
back().
For other modifying operations, users of the API typically need to
decide whether a detach is needed based on QArrayData's requirements
(is data mutable? is it currently shared?) and its own (do we have
spare capacity for growth?).
Now that data may be shared, static or otherwise immutable (e.g.,
fromRawData) it no longer suffices to check the ref-count for
isShared().
This commit adds needsDetach() which, from the point-of-view of
QArrayData(Pointer), answers the question: 'Can contained data and
associated metadata be changed?'.
This fixes QArrayDataPointer::setSharable for static data (e.g.,
Q_ARRAY_LITERAL), previously it only catered to shared_null.
SimpleVector is also fixed since it wasn't checking Mutability and it
needs to because it supports fromRawData().
Change-Id: I3c7f9c85c83dfd02333762852fa456208e96d5ad
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This enables a truncating resize() to be implemented. It is similar to
destroyAll(), but updates the size() as it goes, so it is safe to use
outside a container's destructor (and doesn't necessarily destroy all
elements).
The appendInitialize test was repurposed and now doubles as an
additional test for QArrayDataOps as well as exercising SimpleVector's
resize().
Change-Id: Iee94a685c9ea436c6af5b1b77486734a38c49ca1
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This follows QArrayData::detachFlags's lead. Given the (known) size for
a detached container, the function helps determine capacity, ensuring
the capacityReserved flag is respected.
This further helps aggregating behaviour on detach in QArrayData itself.
SimpleVector was previously using qMax(capacity(), newSize), but there's
no reason to pin the previous capacity value if reserve() wasn't
requested. It now uses detachCapacity().
Change-Id: Ide2d99ea7ecd2cd98ae4c1aa397b4475d09c8485
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Adds given number of default-initialized elements at end of array. For
POD types, initialization is reduced to a single memset call. Other
types get default constructed in place.
As part of adding a test for the new functionality the arrayOps test was
extended to verify objects are being constructed and assigned as
desired.
Change-Id: I9fb2afe0d92667e76993313fcd370fe129d72b90
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Commit 3fe1eed0 changed the QVERIFY in line 1354 to QCOMPARE. This was
done to work around a (not yet understood) compiler issue. That however
was wrong, as char pointers in QCOMPARE are assumed to point to
'\0'-terminated strings and will get dereferenced.
In this case the intent was to compare the actual pointer values, as the
pointers point past the end of the array and should not be dereferenced.
Explicitly casting to (void *) and using QCOMPARE will not only keep the
intent, it will hopefully also provide meaningful output on failures. As
such the fix was applied throughout the test.
Change-Id: Ib0968df492ccc11d7c391bb69037cd7241e55493
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
Seen with gcc 4.6:
tst_qarraydata.cpp: In member function 'void tst_QArrayData::grow()':
tst_qarraydata.cpp:1445:29: error: narrowing conversion of 'i' from
'size_t {aka long unsigned int}' to 'int' inside { } [-fpermissive]
Change-Id: Iad55659554b64ee34655640d606153f058a8cd05
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Besides rvalue-references, this test depends on the compiler to generate
implicit move operators on a derived class, based on the ones available
on its base class.
At least Visual Studio 2010 and some variations of clang 3.0 are known
not to generate implicit move constructors and assignment operators. Gcc
4.6 and up seem to support the feature.
Change-Id: Ied464ef678f517321b19f8a7bacddb6cd6665585
Reviewed-by: Kent Hansen <kent.hansen@nokia.com>
This is meant to reduce the number of allocations on growing containers.
It serves the same purpose as the existing qAllocMore which is currently
used by container classes.
While only a container knows when it is growing, it doesn't need to care
how that information is used. qAllocMore is currently treated as a
black-box and its result is (basically) forwarded blindly to an allocate
function. In that respect, container code using qAllocMore acts as an
intermediary.
By merging that functionality in the allocate function itself we offer
the same benefits without the intermediaries, allowing for simpler code
and centralized decisions on memory allocation.
Once all users of qAllocMore get ported to QArrayData and
QArrayData::allocate, qAllocMore can be moved or more closely integrated
into qarraydata.cpp and qtools_p.h can be dropped.
Change-Id: I4c09bf7df274b45c399082fc7113a18e4641c5f0
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
I can't figure this one out, but it seems to be a clang compiler bug
that is triggered in association with -DQT_NO_DEBUG. Changing the
test from QVERIFY to QCOMPARE keeps the intent and the check, but makes
the failure go away.
It can't hurt...
Change-Id: Ib34e5e850e5b731d729e417430dec55e372805ac
Reviewed-by: Chris Adams <christopher.adams@nokia.com>
I love how this magically makes SimpleVector move-aware :-)
Change-Id: I5cb75954d70cf256863c33e684ebc4551ac94f87
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
- Updated copyright year, per 1fdfc2abfe
- Updated contact information, 629d6eda5c
- Drop "All rights reserved", 5635823e17
(Empty line added to maintain license header line count)
Change-Id: Ie401e2b6e40a4b79f4191377dd50dc60be801e1f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
There isn't a good reason to impose the additional library call and
using the shared null here matches existing behavior in QByteArray,
QString and QVector.
Change-Id: Idd0bb9c7411db52630402534a11d87cbf2b1e7ba
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
The extra bool arguments weren't needed in the first place as they
specify the default, but were left behind when allocate parameters were
changed from bools to AllocationOptions.
Clang saves the day by pointing out the weird conversion going through
void ** (!?)
Change-Id: Ia0dafce06bf0ee62bd825a2db819c890343b6342
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
This provides the same functionality as the specialized QStringLiteral
and QByteArrayLiteral, but on top of QArrayData.
The macro has two variations, variadic and simple. The variadic version
depends on compiler support for (C99) variadic macros and enables
static initialization of arrays of any POD data. Use of this macro is
not recommended on code or applications that need to work in
configurations where variadic macros are not supported.
The simple version is more portable and is enough to support the use
cases of QStringLiteral and QByteArrayLiteral, also providing a fallback
that allocates and copies data when static initialization is not
available.
Change-Id: I7154a24dcae4bbbd7d5978653f620138467830c5
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
, as those are going away.
This cleans use of those operations in the QArrayData stack.
Change-Id: I67705fe0a2f8d99ea13739b675021356a5736f83
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
Reviewed-by: hjk <qthjk@ovi.com>
By default, QTypedArrayData::fromRawData provides the same semantics as
already exist in QByteArray and QString (immutable, sharable data), but
more combinations are possible. In particular, immutable-unsharable
leaves the data owner in control of its lifetime by forcing deep copies.
As part of this, a new isMutable property is introduced in QArrayData.
This could be taken to be implicit in statics that are initialized with
a proper size but with alloc set to 0. QStringLiteral and QByteLiteral
already did this, forcing re-allocations on resize even before the
(static, thus shared) ref-count is considered.
The isMutable property detaches data mutability and shared status, which
are orthogonal concepts (at least in the unshared state). For the time
being, there is no API to explicitly (re)set mutability, but statics and
RawData mark data immutable.
Change-Id: I33a995a35e1c3d7a12391b1d7c36095aa28e221a
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
(This is only for a test case, but still...)
Change-Id: Ied205860e5469000249e15a5478c10db53f1fdaa
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
This approach is better for future ABI evolution than using individual
bool parameters. QArrayData now also offers to calculate allocate
options for typical detach and clone operations: the CapacityReserved
flag is preserved, while cloning resets the Unsharable state.
Change-Id: I256e135adcf27a52a5c7d6130069c35c8b946bc3
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
They still exist and help avoid allocation of "empty" array headers, but
they're no longer part of the public API, thus reducing relocatable
symbols and relocations in inline code.
This means an extra non-inline call on QArrayDataPointer::clear and
setSharable operations, which are (expensive) detaching operations,
anyway.
Change-Id: Iea804e5ddc8af55ebc0951ca17a7a4e8401abc55
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Making use of the same feature added in RefCount.
To keep with the intention of avoiding the allocation of "empty" array
headers, this introduces an unsharable_empty, which allows users to
maintain the "unsharable bit" on empty containers, without imposing any
actual allocations.
(Before anyone asks, there is no point to a zero-sized capacity-reserved
container so no other combinations are needed for now.)
Change-Id: Icaa40ac3100ad954fdc20dee0c991861136a5b19
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
This removes const qualification on data members of QConst*Data, which
was subjecting QString's and QByteArray's shared_null to the "order of
static initialization fiasco", with up-to-date VS 2010.
Furthermore, the const qualification in the places where it was removed
had little meaning and no value. It was unnecessary. As such, "Const"
was removed from the struct's names and "Static" used in its place, to
imply their usefulness in supporting statically-initialized fixed-size
(string and byte) containers.
A test case was added to QArrayData as that is meant to replace both
QStringData and QByteArrayData in the near future.
VS issue reported at:
https://connect.microsoft.com/VisualStudio/feedback/details/716461
Change-Id: I3d86f2a387a68f359bb3d8f4d10cf3da51c6ecf7
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
A reference count of 0 (zero) would never change. RefCount::deref to
zero would return false (resource should be freed), subsequent calls on
the same state would return true and not change state. While safe from
RefCount's side, calling deref on a reference count of zero potentially
indicated a dangling reference.
With this change, a reference count of 0 is now abused to imply a
non-sharable instance (cf. QVector::setSharable). This instance is to be
deleted upon deref(), as the data is not shared and has a single owner.
In practice, this means an (intentional) change in behaviour in that
deref'ing zero still won't change state, but will return false, turning
previous access to dangling references into double free errors.
Users of RefCount wanting to support non-sharable instances are required
to check the return of RefCount::ref() and use RefCount::isShared() to
determine whether to detach (instead of directly checking count == 1).
New functions are introduced to determine whether RefCount indicates a
"Static" (permanent, typically read-only) or "Sharable" instance and
whether the instance is currently "Shared" and requires detaching prior
to accepting modifications..
This change formalizes -1 as the value used to flag persistent,
read-only instances, no longer reserving the full negative domain. The
concrete value is part of the ABI, but not of the API. (isStatic and
Q_REFCOUNT_INITIALIZE_STATIC are part of the API, instead)
Change-Id: I9a63c844155319bef0411e02b47f9d92476afefe
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
This was only being used to initialize static read-only RefCount
instances, where the value is hard-wired to -1. Instead of allowing
initialization with arbitrary values (which for a reference count can be
error prone) the intent of the macro is made explicit with its
replacement Q_REFCOUNT_INITIALIZE_STATIC.
Change-Id: I5b0f3f1eb58c3d010e49e9259ff4d06cbab2fd35
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
This class provides RAII functionality for handling QArrayData pointers.
Together with QArrayDataHeader and QArrayDataOps, this offers common
boilerplate code for implementing a container which, itself, defines its
own interface.
Change-Id: If38eba22fbe8f69038a06fff4acb50af434d229e
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
Inserting elements anywhere in the array requires moving the elements
that follow out of the way and writing in the new ones. Trivial for PODs
and almost as much for movable types.
For "complex" types, we start by extending the array with placement new
and copy constructing elements. Then, copy assignment resets the
elements that were previously part of the array.
QPodArrayOps uses non-throwing operations. QMovableArrayOps provides
full rollback in the face of exceptions (strong guarantee).
QGenericArrayOps enforces that no data is leaked (all destructors
called) and invariants are maintained on exceptions -- the basic
guarantee.
With 3 different implementations, 2 of which are non-trivial, this
operation is a good showcase for QArrayOpsSelector and the different
implementations. As such, it warrants its own commit.
Change-Id: I21d9b4cb8e810db82623bcd1d78f583ebf3b6cb7
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
This class, the selector and underlying implementations provide
specialized operations on QArrayData, while allowing for optimized
implementations that benefit from type-specific information.
Currently, offering a generic implementation and specializations for
PODs (trivial ctor, dtor and move operations) and movable types (can be
trivially moved in memory).
Change-Id: I2c5829b66c2aea79f12f21debe5c01f7104c7ea3
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
A bug has been reported against GCC 4.4.3 (present in other version as
well), where the use of an array of size 1 to implement dynamic arrays
(such as QVector) leads to incorrect results in optimized builds as the
compiler assumes the index to be 0.
This test tries to ensure QArrayDataHeader is not affected by this bug,
as QVector currently is.
Change-Id: Id701496bae4d74170de43399c1062da40eb078e7
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
