Looks like I failed to update this earlier, when the behaviour changed.
Change-Id: Ic020c2a14d4e9153f2bc9d22d943a3a380c0851c
Reviewed-by: Shane Kearns <shane.kearns@accenture.com>
Previously the method attempted to reset only as a last resort.
Now reset() is deprecated and resetting must happen between emitting
modelAboutToBeReset() and modelReset(). Since this suffices in all
cases to notify views that they must reinterrogate the model, it is no
longer necessary to signal explicitly row removals and insertions
within the scope of the reset.
Additionally, fetchMore() is now called within the scope of the reset
so insert signals do not have to be emitted here either.
This improved handling of resetting in QSqlQueryModel also allows the
cache in QSqlTableModel to be cleared directly at select().
This change may actually allow views to operate more efficiently since
they no longer have to react to separate row removal and insert
signals. Views can avoid pointless deallocation and reallocation
by considering row count only after the reset is finished. The cost is
that the columns and horizontal headers must be considered in the view
at each setQuery() call. In any case, it is not clear that trying to
be smart about this in the model justifies additional complexity.
Tests had to be adjusted where they expected explicit row removal
and insert signals.
Change-Id: I4f7eac1419824361d7d9bdcc6a87092b33e80d7a
Task-Id: QTBUG-25419
Reviewed-by: Andy Shaw <andy.shaw@digia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Honglei Zhang <honglei.zhang@nokia.com>
Lots of uses of the annotations and error names, plus a bunch of local
unit test names (including one file that had to be renamed).
The meta object generator is updated to support both the old and new
names. That means some references to com.trolltech *must* remain in the
source code.
Task-number: QTBUG-23274
Change-Id: Icc38ae040232f07c437e7546ee744a4703f41726
Reviewed-by: Jason McDonald <jason.mcdonald@nokia.com>
Reviewed-by: Lorn Potter <lorn.potter@nokia.com>
- change customColor() to return QColor instead of QRgb.
- change setCustomColor() and setStandardColor() to take
a QColor instead of QRgb.
- add missing standardColor() getter method.
Task-number: QTBUG-25087
Change-Id: Ic6adb2031ef47f5e9b15fa3560a5322e6847c0bb
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Algorithmic complexity attacks against hash tables have been known
since 2003 (cf. [1, 2]), and they have been left unpatched for years
until the 2011 attacks [3] against many libraries /
(reference) implementations of programming languages.
This patch adds a qHash overload taking two arguments: the value to
be hashed, and a uint to be used as a seed for the hash function
itself (support the global QHash seed was added in a previous patch).
The seed itself is not used just yet; instead, 0 is passed.
Compatibility with the one-argument qHash(T) implementation is kept
through a catch-all template.
[1] http://www.cs.rice.edu/~scrosby/hash/CrosbyWallach_UsenixSec2003.pdf
[2] http://perldoc.perl.org/perlsec.html#Algorithmic-Complexity-Attacks
[3] http://www.ocert.org/advisories/ocert-2011-003.html
Task-number: QTBUG-23529
Change-Id: I1d0a84899476d134db455418c8043a349a7e5317
Reviewed-by: João Abecasis <joao.abecasis@nokia.com>
Certain QSqlDriver functions were marked to be made virtual in Qt5.
subscribeToNotification, unsubscribeFromNotification,
subscribedToNotifications, isIdentifierEscaped, and stripDelimiters.
This patch makes them virtual and removes the no longer needed
Implementation counterpart functions. It also updates the relevant
drivers. This patch has no regressions on the tests in
tests/auto/sql/kernel/, tested with sqlite and postgres.
Change-Id: Ia2e1c18dfb803531523a456eb4e710031048e594
Reviewed-by: Mark Brand <mabrand@mabrand.nl>
Both classes had such components before, but there were issues with
the NOTIFY signal not being in the same class as the Q_PROPERTY.
This patch solves that problem by using a signal of a different name.
Task-number: QTBUG-15731
Change-Id: Ibc7ce4dba8a6b88c05d62a90e14d0101c5cd3082
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Thorbjørn Lund Martsum <tmartsum@gmail.com>
QMetaMethod::typeName() is documented to return an empty string if
the return type is void. But after the introduction of
QMetaType::UnknownType (where void was made a distinct type),
returning an empty string causes the idiom
QMetaType::type(method.typeName())
to break; the result will be QMetaType::UnknownType rather than
the expected QMetaType::Void for methods that return void.
New code should use the new function QMetaMethod::returnType()
instead, but it would be good if existing code still did the right
thing.
The consequence of returning "void" instead of an empty string is
that it breaks existing logic that uses the typeName() length to
determine whether a method returns void. But we judge this as the
lesser of the two evils; it's better to have a typeName() function
that is consistent and keeps the QMetaType::type(method.typeName())
idiom working, than to force the typeName() inconsistency for void
only to keep code that does "strlen(method.typeName()) == 0"
working.
The places in Qt that were relying on a zero-length typeName()
(testlib, dbus, declarative) have already been changed to use
returnType().
Also adapt QMetaObjectBuilder, which is internal API.
Change-Id: I70249174029811c5b5d2a08c24b6db33b3723d19
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@nokia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
QPointer has been un-deprecated and one behavior which slightly different
from Qt4 has been fixed.
see SHA: b8773165d7
and SHA: 497622cafe
Change-Id: I4bae2cce3ebfebd8f59b18b5a6a7a7226b8353b9
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Reviewed-by: Casper van Donderen <casper.vandonderen@nokia.com>
For OnFieldChange and OnRowChange, we don't want more than one row in
the cache with uncommitted changes. This could happen if deletion in
the database fails while other changes are pending.
Chosen solution is to return false if other rows have pending changes.
Also, we only allow 1 row removed at a time.
Updated test, changes and documentation.
Change-Id: I68baf6d221789b4754e891535070011c759a2155
Reviewed-by: Honglei Zhang <honglei.zhang@nokia.com>
The model can never do a good job of knowing when user moves to a
new row in the view. Faking it by detecting when another row
is changed was not a good solution because it cannot detect
when the last edited row is left.
Either the view should automatically submit when the user leaves
a row or the application should provide a way to submit.
This change made it possible to reuse the logic of flags() in
setData().
Change-Id: I2550e5b113bceba1a852fc21203babeca07c5748
Reviewed-by: Honglei Zhang <honglei.zhang@nokia.com>
For OnFieldChange and OnRowChange, inserting rows should not be
allowed if there are pending changes in cache.
Change-Id: Ia794332959a35a1de87e798ba1a74ace3dfae68f
Reviewed-by: Honglei Zhang <honglei.zhang@nokia.com>
For OnFieldChange and OnRowChange, before submitting new changes,
setData() and setRecord() attempt to submit pending changes and
revert them upon failure. However, they fail to consider that
reverting pending insertions removes rows from the model. As a
result, the new change can be applied to a row higher than intended.
One possible solution would be to adjust the targetted index for the
removed rows, so that the intended row is affected by the new change.
But this still causes the strange editing experience as rows jump
up just as they are being edited.
It does not seem right in the first place for the model to initiate
reverting changes. It should be up to the application to decide what
to do when data cannot be committed. In particular, setData() and
setRecord() should not have the side effect of reverting already
pending changes.
The chosen solution is simply to refuse new changes that don't make
sense for the edit strategy. For OnFieldChange, flag() will
indicate read-only when editing is blocked by a pending change.
Since setData() and setRecord() submit data immediately for
OnFieldChange, it no longer makes sense to resubmit changes
automatically before a new change.
For OnRowChange, setData() keeps the behavior of automatically
submitting a pending row before starting on a new row. This is
historical behavior and is probably motivated by the fact that
QTableView does not automatically call submit() when editing leaves a
row. The obvious shortcoming of this is that the last row to be edited
will not be submitted automatically. It also prevents us from flagging
rows other than the pending row as read-only.
For OnRowChange, setRecord(), being row-oriented by nature, should
submit the change immediately rather than waiting for the next call
to setRecord(). This makes setRecord() consistent with insertRecord().
Change-Id: Icb4019d8b7c53a7ee48f8121a7a525e8bc35d523
Reviewed-by: Honglei Zhang <honglei.zhang@nokia.com>
This patch renames the functions in Qt5 according to the
notes. It also renames resizeMode to be consistent.
The old functions are both marked with both QT_DEPRECATED
and '### Qt 6 - remove'
All usage of the function within the qtbase are also
changed to use the new functions.
Change-Id: I9e05fa41d232e9ca43b945fcc949987017f3aedd
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Checks if model has any changes to submit.
Includes new test covering isDirty(index) as well the new
overloaded function.
Task-number: QTBUG-3108
Change-Id: I0ccbda45d5d9f06434cf1e1c037a9efb76d0cc37
Reviewed-by: Honglei Zhang <honglei.zhang@nokia.com>
this makes the "sysrootable" properties more magic, with the raw
versions being omitted from the qmake -query output and automatically
falling back to the "cooked" variant if there is no sysroot set.
this makes the "normal" qmake -query less noisy. this will become even
more obvious when i add more "overloads" of the properties.
Change-Id: I08000986427264ec6238c8fe0a77f5cecdbf1201
Reviewed-by: Joerg Bornemann <joerg.bornemann@nokia.com>
Mention in changes and document Qt version (merci à dfaure).
Follow-up to 291e2c7d54.
Change-Id: Ie5626e9cd268812c1173ca494ccd8d6bd9be2687
Reviewed-by: Honglei Zhang <honglei.zhang@nokia.com>
Calling select refreshes the query data but disrupts view
navigation.
For OnFieldChange and OnRecordChange it makes sense to only
select the row in question. This does not disturb view navigation.
Assume disruption of view navigation is not a problem
for OnManualSubmit because the user or application decides
when submitAll is called.
Task-number: QTBUG-2875
Change-Id: I1e5f68668fb9102f6296d67d543e80daa403f1c4
Reviewed-by: Yunqiao Yin <charles.yin@nokia.com>
Added QRegularExpression, QRegularExpressionMatch and
QRegularExpressionMatchIterator as PCRE-enabled, regexp classes.
Documentation is included, as well as a first round of autotests.
Task-number: QTBUG-23489
Change-Id: Id47031b80602c913ccd2fd740070e3024ea06abc
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Similarly to change id I2f429fa7ef93bd75bb93a7f64c56db15b7283388, the capability
to arbitrarily alter the encoding of literals is very destructive, especially in
a world with libraries and plugins.
Change-Id: If0d4cd8dcf89792e39c1984cbde6b036cebfc02f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
load()/include() with a target namespace would inherit the current
context. however, if you source a project with all bells and whistles,
this makes completely no sense and may be actually counterproductive.
infile()/$$fromfile() would have interited only the functions from the
current context. that was only a hack to support abusing them.
Change-Id: I2e992b923d9e5b0e5056001ca49b35de573abc63
Reviewed-by: Joerg Bornemann <joerg.bornemann@nokia.com>
don't inject the build pass specific variables into the project even
before evaluating the .spec file and the .qmake.cache. they are not
supposed to base configuration on that - feature files should do that
later.
the immediate advantage of this is that base_vars is never manipulated
upfront any more, which allows for cleaner setup paths. also, we can do
more caching of the spec+cache contents.
Change-Id: I19d7f8bec1fb7c3b54121e26794340b287055ebf
Reviewed-by: Joerg Bornemann <joerg.bornemann@nokia.com>
QMetaType::Void was ambiguous, it was pointing to a valid type (void)
and in the same time it was signaling errors in QMetaType. There was
no clean way to check if returned type was valid void or some
unregistered type.
This feature will be used by new QMetaObject revision which will
store type ids instead of type names. So it will be easy to
distinguish between:
void mySlot();
MyUnregisteredType mySlot();
Change-Id: I73ff097f75585a95e12df74d50c6f3141153e771
Reviewed-by: Kent Hansen <kent.hansen@nokia.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
instead of being a variable added to the makespec (via qconfig.pri),
QT_SYSROOT is now a property.
the QT_INSTALL_... properties are now automatically prefixed with the
sysroot; the raw values are available as QT_RAW_INSTALL_... - this is
expected to cause the least migration effort for existing projects.
-hostprefix and the new -hostbindir & -hostdatadir now feed the new
QT_HOST_... properties.
adapted the qmake feature files and the qtbase build system accordingly.
Change-Id: Iaa9b65bc10d9fe9c4988d620c70a8ce72177f8d4
Reviewed-by: Marius Storm-Olsen <marius.storm-olsen@nokia.com>
There's not a single in-tree user of this function, and
the concept is a broken one in MT programs: By the time
qIsDetached() returns, the result can already be
different due to another thread taking a copy, or a
copy in another thread being destroyed (note that this
doesn't require mutex use by the user, since we promise
(implicitly, if not explicitly) that you can copy from
const objects without holding a lock).
QTBUG-10813 talks about a use in QCache::trim(), but
677cf76340 removed it, so
there's no reason to keep it anymore.
Change-Id: I20380c12bdf00ac764b89d84392f0f34727b1971
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>
we already initialize it before parsing a project. if a project is daft
enough to clear TARGET, it does not deserve differently than breaking.
Change-Id: I6c727bc27d72a00e84b676ae3c169024bdb2d929
Reviewed-by: Joerg Bornemann <joerg.bornemann@nokia.com>
This setting is extremely harmful, as code cannot know whether or not to expect
it. It also made the behaviour of QString::fromAscii and ::toAscii unintuitive,
and caused a lot of people to make mistakes with it.
Change-Id: I2f429fa7ef93bd75bb93a7f64c56db15b7283388
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
This abstraction imposed serious performance penalties and is being
dropped from the public API.
In particular, by allowing file names to be arbitrarily hijacked by
different file engines, and requiring engines to be instantiated in
order to decide, it imposed unnecessary overhead on all file operations.
Another flaw in the design with direct impact on performance is how
engines have no way to provide (or retain) additional information
obtained when querying the filesystem. In many places this has meant
repeated operations on the file system, where useful information is
immediately discarded to be queried again subsequently.
For Qt 4.8 a major refactoring of the code base took place to allow
bypassing the file-engine abstraction in select places, with
considerable performance gains observed. In Qt 5 it is expected we'll be
able to take this further, reaping even more benefits, but the
abstraction has to go.
[Dropping this now does not preclude that virtual file systems make an
appearance in Qt at a later point in Qt 5's lifecycle. Hopefully with a
new and improved abstraction.]
Forward declarations for QFileExtension(Result) were dropped, as the
classes were never used or defined.
Tests using "internalized" classes will only fully run on developer
builds. QFSFileEngine was removed altogether from exception safety test,
as it isn't its intent to test internal API.
Change-Id: Ie910e6c2628be202ea9e05366b091d6d529b246b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Add QSysInfo::macVersion() instead, to match the windowsVersion()
function.
Change-Id: I783e59583ca21653d25586156cbb0cb1f301868b
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Qt 5 seems like an excellent opportunity to simplify logic and separate
concerns by making indexInQuery() virtual. Note that this wasn't my
idea, but was mentioned in a helpful comment.
Change-Id: Ie29ead110def45297c32de3ce6d07a8eefb08d8c
Reviewed-by: Yunqiao Yin <charles.yin@nokia.com>
Constructors taking explicit sizes got a default -1 size argument that
triggers length calculation from nul-terminated strings.
This imposes a slight change in behavior: negative size arguments would
previously be ignored and generate an empty string whereas with this
patch we expect to see a nul-terminated string.
On the other hand, keeping the previous behavior could effectively hide
errors in user code and I can't find a good reason to support it.
Documentation for the constructors was updated and made more consistent
between the classes.
Change-Id: I738ac3298cffe3221c8a56e85ba2102623e7b67d
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
For data-driven tests, testlib previously counted one fail or skip for
each data row that failed or skipped, while it counted only one pass
for a test function where all rows passed and counted no passes for a
test function where some rows passed and some rows failed. A similar
problem also existed for benchmark tests, which could run multiple
iterations of the same test, with each fail and skip being counted but
only a single pass being counted for the entire series of iterations.
This commit makes testlib count one pass, fail or skip for each data
row. Test functions that are not data-driven count one result for the
test function, as before. Benchmark tests count one pass, fail or skip
per iteration.
A side-effect of this change is that the test output in plain text, xml
and light xml formats now shows a result for every data row and
benchmark iteration executed, allowing post-processors to correctly
calculate the total number of tests executed. Previously, individual
rows were not shown in the test output if they passed, making such
calculations impossible.
The only change to the xunitxml output format is to correct a bug where
no test result was recorded for a test function if the last data row
was skipped and all other rows passed -- in which case the overall
result should be a pass. Note that there is also a pre-existing bug
in the xunit logger, where no result is reported if all rows are
skipped; that bug is unaffected by this commit.
Task-number: QTBUG-21848
Task-number: QTBUG-22124
Change-Id: I7e17177e10d6e89e55b9684c159bd506f21d002b
Reviewed-by: Ed Baak <ed.baak@nokia.com>
