This prevents deadlocks in case the destructor re-enters.
(Example: a functor containing a QSharedPointer of a QObject)
This also fixes a leaked slot object in disconnectHelper.
Change-Id: Ia939790e3b54e64067b99540974306b4808a77f2
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Fix connection to pointer to member signal that belongs to the base
class, but whose type is a pointer to a member of the derived class.
The current code only use the QMetaObject of the type coming from the
function type to look up the signal id. But if the signal was casted
to a pointer to member function of a derived type, then we also need to
look in the base classes
Change-Id: Ib98fc38f63942946acb34d9f83c100991d58e4e5
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
When variadic templates and decltype are supported, detect the best
overload of operator() to call.
Currently, the code takes the type of the operator(), which requires that
the functor only has one, and that it has no template parameter.
This feature is required if we want to connect to c++1y generic lambda
(N3418)
Change-Id: Ifa957da6955ea39ab804b58f320da9f98ff47d63
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
While adding a test case for the new behavior, two issues
with the connectSlotsByName implementation came up:
1. for auto-connected slots that don't exactly match a signal,
a 'compatible' one is searched. There might be more than
one of those. The implementation randomly picks any.
2. The "No matching signal for %s" warning gets printed even for
slots that can never be connected via connectSlotsMyName
anyway (e.g. "on_something"). This is inconsistent.
This fixed both: an explicit warning is printed if more than one
'compatible' signal is found and the "No matching signal for %s"
warning is only printed if the slot adheres to the full
"on_child_signal()" naming convention.
In the process I added comments and changed the code slightly to
make it more readable and explicitly hint at non-obvious behavior.
Change-Id: Icc8e3b9936188d2da8dfff9f0373c8e5c776eb14
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
QMetaObject::connectSlotsByName(QObject* o) creates a list of all
children to look for signals that match slots of o. This changeset
simply adds the object o itself to that list.
The motivation is to finally fix the long standing QtCreator bug
QTCREATORBUG-6494. Where executing 'Go to slot...' and choosing
'accepted()' for a simple QDialog named 'MyDialog' will add a
on_MyDialog_accepted() slot to MyDialog. That slot never gets
connected. More details may be found in the linked QTBUG-7595.
Task-number: QTBUG-7595
Task-number: QTCREATORBUG-6494
Change-Id: I35f52761791af697eabb569adb5faee6fae50638
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
I changed the existing test-case code to make it more scalable
in terms of adding more connectSlotsByName related tests. The
old "manual list of ints"-method was really not developer friendly.
This is a preparation for a subsequent modification of
connectSlotsByName behavior.
Change-Id: Ib760e52631ce4b5ae2a3ebdb4854849ff6c93bfe
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
This fixes QMetaType detection of const reference arguments in signals
while connecting using the new syntax and Qt::QueuedConnection
const references should have the same QMetaType as non references.
That means we need to remove the const reference while getting the
QMetaType.
Change-Id: I9b2688da7fb9ae985aec0d8fa62a1165357ffe71
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
Unlike the previous implementation, this implementation is locked:
only one initialisation is ever run at the same time. It is
exception-safe, meaning that a throwing constructor will restart the
process.
Also, start using the thread-safe behaviour that GCC has offered for a
long time and C++11 requires.
Change-Id: I20db44f57d258923df64c0051358fd0d9a5ccd51
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: David Faure (KDE) <faure@kde.org>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
When disconnect()ing through a QMetaObject::Connection, if the
QObjectPrivate::Connection contains a slot object, deref it, so
that it will be destroyed before the next run of cleanConnectionList.
Previously, a copy of the functor passed to connect() was kept until
QObjectPrivate::cleanConnectionLists was called (by adding a new signal,
or the sender was destroyed), even after a successful call to
disconnect(). That is, we were keeping that copy allocated without
any good reason.
Change-Id: Ie6074ea797df1611cb995dec07c5b5a742360833
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
The QObjectPrivate::Connection refcount was not decreased
when disconnect()ing, therefore it was kept alive by the
owning QMetaObject::Connection object.
This removes a leak in case the QMetaObject::Connection
survives the sender object, after a successful disconnect().
Change-Id: Ie2ea59b269a0e589ae23c1457df7533be77c0797
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
More QVERIFY()s were added in the meanwhile to the test function,
so we can drop this one.
Change-Id: If6f137f45ba606b61d6a7004556a667ed316b61f
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Before, the functor slot (or lambda expression) had to have the same amount
of arguments as the signal.
This shown to be a big problem to be able to connect to signals that had
a QPrivateSlot.
This implementation use the type of the operator() of the functor to
know how many arguments we have.
As a bonus, we also can check the arguments in a static assert.
The test comes from https://codereview.qt-project.org/#change,38703
But this patch also works without variadic template
If the compiler does not support decltype, we workaround the lack of it
by using another level of indirection.
Change-Id: I9850b43e8caf77356a2ec3f4c0b0ed532d96029e
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Pointer to const member function have a different signature, and hence
need their own traits code.
Change-Id: Ie4b2434a412f412444fb07ef1388a37cab105ecd
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Normalise all signal/slot signatures in tests/*/corelib,
except in tst_QObject, where they might be test data.
Change-Id: Id4e101f285b1676bb583b0afae06d235e599e24b
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.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>
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>
This is a simplified port of KDTools' KDAB_SET_OBJECT_NAME.
It simply assigns the variable name as the objectName of
a QObject, uic-style. It uses a small helper function so
that it works on references as well as pointer variables.
QLabel label;
QLabel *pLabel = new QLabel();
Q_SET_OBJECT_NAME(label);
Q_SET_OBJECT_NAME(pLabel);
Change-Id: I25fec0c90f33249a3ea5d2dd622ab708019fd101
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>
Reviewed-by: David Faure <faure@kde.org>
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Make test projects declare TEST_HELPER_INSTALLS rather than calling a
function exported by testcase.prf. load(testcase) may be unsafe, as
testcase.prf should be processed after default_post.prf.
Fixes silent disabling of various autotests.
Change-Id: I56b35ffd653a637ad5ab18d64dd1a1edadfac59f
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@nokia.com>
Reviewed-by: Kurt Korbatits <kurt.korbatits@nokia.com>
Store the signal index in QObjectPrivate::Connection, thereby making
it available in "implicit" disconnect contexts (i.e., receiver
deletion).
This change does not cause the size of QObjectPrivate::Connection
to grow (still 40 bytes on 32-bit Linux, 72 bytes on 64-bit Mac).
Valgrinding the new benchmark indicates that the percentage of the
time spent in the QObject destructor increased from 7.8% to 8.4%
on ia32, for that particular stress test; the increase is the
combined cost of calling metaObject(), QMetaObjectPrivate::signal(),
and disconnectNotify() for one connection. In practice, the measured
wallclock time increased by about 3ms for a 500ms run (which
repeatedly constructs, connects, and destroys an object).
Task-number: QTBUG-4844
Change-Id: I1beb01c753f31542fc0acb62edb4c6d165fcc5b4
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
These tests have passed a parallel stress test on all three of Linux,
Mac, Windows. Mark them with CONFIG+=parallel_test to allow CI to run
them in parallel, saving time.
Change-Id: I19fd333c3c645a67374ca998f6c8530dd236b0f8
Reviewed-by: Toby Tomkins <toby.tomkins@nokia.com>
This completes the transition from connectNotify(const char *) and
disconnectNotify(const char *) to the new QMetaMethod-based
functions.
Removed the old connectNotify autotests and renamed the
connectNotifyMethodXXX autotests to connectNotify, since there is
no longer any ambiguity about which overload is being tested.
Change-Id: Icf108a80177155f21bb73c165fb8ab5d4e997bc2
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
This operation should be a no-op anyway, since at this point in time,
the fromAscii and toAscii functions simply call their fromLatin1 and
toLatin1 counterparts.
Task-number: QTBUG-21872
Change-Id: I38f97ad379deafebef02c75d611343ca15640c8a
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
This is much more performant than calling QObject::receivers(const char*)
Can be used instead of connectNotify in some cases.
Change-Id: I19e0933f678f171f515d9a0f69f0ad4fb7d894b4
Reviewed-by: Kent Hansen <kent.hansen@nokia.com>
qVariantValue and qVariantCanConvert are Compatibility members, while in
Qt4.8 they are marked as Qt 3 Support Members.
qVariantFromValue and qVariantSetValue are Obsolete members.
Change-Id: Ie8505cad1e0950e40c6f6710fde9f6fb2ac670fd
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@nokia.com>
Reimplementations of connectNotify() and disconnectNotify() can
assume that the signal argument is in normalized form, but after the
introduction of the Qt5 meta-object format, it could happen that it's
not.
The problem is that the internal QArgumentType class, which attempts
to resolve a typename to a type id, was calling QMetaType::type().
QMetaType::type() falls back to trying the normalized form of the
typename if the original argument can't be resolved as a type (this
behavior isn't documented, but that's how it works). This means that
e.g. QMetaType::type("const QString &") returns QMetaType::QString.
Since QMetaObjectPrivate::indexOfMethodRelative() (more specifically,
the methodMatch() helper function) prefers to compare type ids
over typenames (since the type ids are stored directly in the meta-
object data for built-in types), the method lookup would *succeed*
for signatures with non-normalized built-in typenames as parameters.
QObject::connect() would then think that it did not have to
normalize the signature (see "// check for normalized signatures").
The consequence was that the original, non-normalized form got
passed to connectNotify().
This commit introduces an internal typename-to-type function that
is the same as QMetaType::type(), except it doesn't try to normalize
the name. This way, the only place where normalization can occur in
the signature-to-meta-method processing is through the calls to
QMetaObject::normalizedSignature() in QObject::connect() itself.
The implication is that there are now cases where the method
signature will be decoded and processed twice, where processing it
once was sufficient before. On the other hand, it is consistent with
the pre-Qt5-meta-object behavior, where we predict that the
signature is already normalized, and only perform (comparatively
costly) normalization if the initial lookup fails.
Change-Id: Ie6b60f60b0f9a57ebd378d980329dac62d57bbd9
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
This API will fully replace the const char *-based connectNotify()
and disconnectNotify() in Qt5; the old functions will be REMOVED
before Qt 5.0 final.
The new implementation fixes the long-standing issue of
connectNotify() not being called when using the (internal)
index-based QMetaObject::connect() (e.g., from QML).
As with the old API, there are still two "unintuitive" behaviors
concerning disconnectNotify():
- disconnectNotify() is not called when the signal is disconnected
using the QObject::disconnect(QMetaObject::Connection) overload.
- disconnectNotify() is not called when a receiver is destroyed
(i.e., when a connection is implicitly removed).
The old versions of connectNotify() and disconnectNotify() are kept
for now, and they are still called. They will be removed once known
existing reimplementations (e.g., QtNetwork, QtDBus) have been
ported to the new API.
Change-Id: I8b4f007f3c6d89199c1ba04a3e23c8ca314e0896
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
That was a regression introduced in 1c5db1aff
Example:
signals: int *someSignal();
would produce this code:
int* _t0 = int*();
which does not compile
So have special handling for pointer to change it to '= 0'
Change-Id: Ie695e15e309d15c3cfd5c5a69ac8bf6d61ae9915
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
- Fixed path was failing to find sub program.
Change-Id: I86f1a6941e244c9bc25ad0441cc7a441607560b7
Reviewed-by: Rohan McGovern <rohan.mcgovern@nokia.com>
For Qt5 we no longer want to support the older revisions due to the
dual codepaths that must be maintained, and because the format of the
meta-object data is quite different in revision 7.
The dual codepaths have been replaced by asserts that indicate the
revision in which the feature was introduced, and the older-revision
fallbacks have been removed.
It's not possible to build code generated by moc that has
revision <= 6 with Qt5 because the type of the
QMetaObject::stringdata member changed from const char * to const
QByteArrayData *. For the same reason it's not possible to build a
dynamic meta-object generator targeting revision <= 6 with Qt5.
Hence, too old meta-objects will be caught at compile time, and the
code will have to be ported to generate revision 7 (e.g., by running
Qt5's moc on the original class declaration).
Change-Id: I33f05878a2d3ee3de53fc7009f7a367f55c25e36
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Since the introduction of QMetaType::UnknownType, void is a proper
meta-type, and the normalized form of "void" should be "void", not
an empty string.
Add more tests to ensure that we do remove "void" in the one case
where it actually should be removed (e.g. "foo(void)").
Change-Id: I72dc2d24da67cf52da00c678f50213cff1b92e25
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
This actually involved tiding up QObject sources a little bit
to clearly separate QString / QRegExp overloads of findChildren.
The corresponding qFindChildren overload for MSVC 6 compatibiltiy
was *not* added.
Change-Id: I84826b3df9275a9bda03608a5b66756890eda6f8
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
In Qt5 the meta-data format will be changed to not store the
method signature string explicitly; the signature will be
reconstructed on demand from the method name and parameter type
information.
The QMetaMethod::signature() method returns a const char pointer.
Changing the return type to QByteArray can lead to silent bugs due to
the implicit conversion to char *. Even though it's a source-
incompatible change, it's therefore better to introduce a new
function, methodSignature(), and remove the old signature().
Task-number: QTBUG-24154
Change-Id: Ib3579dedd27a3c7c8914d5f1b231947be2cf4027
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
Up to and including meta-object revision 6, string data have been
stored as 0-terminated C-style strings, that were made directly
accessible as const char pointers through the public API
(QMetaMethod and friends).
This commit changes moc to generate an array of QByteArrayData
instead, and adapts the QObject kernel accordingly.
Generating an array of QByteArrayData (byte array literals)
means that the strings can now be returned from public (or private)
API as QByteArrays, rather than const char *, with zero allocation or
copying. Also, the string length is now computed at compile time
(it's part of the QByteArrayData).
This commit only changes the internal representation, and does
not affect existing public API. The actual (C) string data that the
byte array literals reference still consists of zero-terminated
strings. The benefit of having the QByteArrayData array will only
become apparent in the upcoming meta-object data format change, which
changes the format of property and method descriptors.
Support for the old meta-object string data format was kept; the
codepaths for old revisions (6 and below) will be removed in a
separate commit, once all the other meta-object changes are done and
affected code has been adapted accordingly.
Task-number: QTBUG-24154
Change-Id: I4ec3b363bbc31b8192e5d8915ef091c442c2efad
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
remove "header" and assignmets which are defaults or bogus,
reorder some assignments.
Change-Id: I67403872168c890ca3b696753ceb01c605d19be7
Reviewed-by: Rohan McGovern <rohan.mcgovern@nokia.com>
When configuring with -fast on Windows, a directory which contains two
.pro files, one SUBDIRS and one not, will have the SUBDIRS Makefile
silently clobbered by the non-SUBDIRS Makefile. In practice, this may
cause various subdirectories to be silently excluded from the build.
Rearrange .pro files for this test to avoid triggering this bug.
See also e9015b3bc8.
Task-number: QTBUG-21168
Change-Id: I18fac1ac636fdc6b2aaee1b4cdfee9c4bc2a77ff
Reviewed-by: Rohan McGovern <rohan.mcgovern@nokia.com>
They create noise in the test results.
Change-Id: I40e7239ba7cd41bec577fe8220c86476553a6502
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
- Made test depend on subprogram to make sure it was there when
test ran.
- install signalbug subprogram
Change-Id: Ie0a19e52d131adcd17c97b263389aecffb81520e
Reviewed-by: Rohan McGovern <rohan.mcgovern@nokia.com>
Reviewed-by: Jason McDonald <jason.mcdonald@nokia.com>
Put in qconfig.h whether qt is compiled with reduced relocations.
When using -Bsymbolic-functions (enabled by default on Qt)
but not -fPIE, the comparison of the function pointers fail
because the addresses are different in Qt, and in the executable.
Hence we now enable -fPIE by default on qmake, and force a compilation
error when it is not enabled and built with reduced relocations.
Done-with: Sune Vuorela <sune@vuorela.dk>
Change-Id: Ib3fdba06fab6e8a93b75b4c6cf16cc973ab335db
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The test was failing at:
QCOMPARE(property.userType(), qMetaTypeId<CustomType*>());
The CustomType* metatype was not registered before this part of the
test.
qMetaTypeId<T> will register the metatype for T before returning it if
it is not yet registered, while QMetaProperty::userType() returns 0 if
the metatype is not yet registered. However, the order of evaluation of
these two expressions in the above statement is technically undefined.
Apparently, gcc evaluates the arguments in order from right to left,
allowing the test to pass, while clang evaluates the arguments in order
from left to right, causing the test to fail.
Change-Id: I5059556e860cec29b57c31e4e26f46cf9e6055da
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
There is no point in keeping separate values which should mean the
same.
QVariant::UserType was used also to construct a valid, null QVariant,
containing an instance of unknown custom type. The concept was strange
and useless as there was no operation that could be done on such
QVariant. Therefore it was dropped.
Please note that the patch slightly changes behavior of different
functions accepting a type id as parameter. Before QVariant::UserType
was an invalid type from QMetaType perspective (id 127 was not assigned
to any built-in type), but QMetaType::User points to the first registered
custom type.
Change-Id: I5c7d541a9affdcdacf53a4eda2272bdafaa87b71
Reviewed-by: Kent Hansen <kent.hansen@nokia.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
Reviewed-by: Andrew Stanley-Jones <andrew.stanley-jones@nokia.com>
Reviewed-by: Aaron Kennedy <aaron.kennedy@nokia.com>
As in the past, to avoid rewriting various autotests that contain
line-number information, an extra blank line has been inserted at the
end of the license text to ensure that this commit does not change the
total number of lines in the license header.
Change-Id: I311e001373776812699d6efc045b5f742890c689
Reviewed-by: Rohan McGovern <rohan.mcgovern@nokia.com>
Some objects expect connectNotify to be called in order the signal to be
emitted.
Change-Id: Id0460d9c2aef8f9c3618a2b62b2119a790e06f30
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
QMetaType::QVariant has existed as a proper type for almost two
years, but the qvariant_nameToType function was written in 2006.
Using QMetaType::QVariant means QVariant can be treated just like
any other type. We can get rid of those hacky checks for LastType,
and the remaining checks become more readable.
The fact that QMetaProperty::{type,userType}() returned LastType
(0xffffffff) for QVariants was never documented (LastType itself is
internal). But there are other Qt modules that assume so. I'll fix
the ones I know about (qtdeclarative, qtscript, activeqt).
Change-Id: I799b9079bb8bbb1fe76c132525440b30415cbac5
Reviewed-by: Bradley T. Hughes <bradley.hughes@nokia.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
