in most cases, it actually means normalizePath() (because the file name is
used with qt i/o functions afterwards).
this affects QMakeLocalFile::local() as well, so many not immediately
obvious places are affected as well.
there was also one case of fixPathToTargetOS() falling into this category.
this is mostly a no-op, as the qt functions are agnostic to the path
separator.
in some other cases (in particular in the vcproj generator), it actually
means fixPathToTargetOS().
this is mostly a no-op as well, as the two functions are equal except on
msys anyway.
in the <meta file>FileName() functions, the use of a fixPath*() function
is bogus in the first place - fileFixify() already does
fixPathToTargetOS(), and this is correct when the file name is used
verbatim in a make command (which it is). otherwise it's irrelevant.
Change-Id: I26712da8f888c704f8b7f42dbe24c941b6ad031d
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
there is no reason why there should be unexpected leading or trailing
whitespace in an extra compiler's .depends list.
Change-Id: I46be75063180131e135fc6eea0238a482073618a
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
instead of quoting more or less random variable contents early,
consistently quote everything only right before it is needed. this way
we can be sure that everything is correctly quoted, but not over-quoted.
this removed the need for the insanity that unescapeFilePath() and
similar ad-hoc contraptions were.
this had the somewhat counter-intuitive effect that it was possible to
remove escapeFilePath() calls from PBX::writeSettings() calls - these
were actually only unescaping.
[ChangeLog][qmake][Important Behavior Changes] A lot of quoting issues
have been fixed. As a side effect, qmake has become more sensitive to
over-quoted file names in project files.
(*) ok, maybe not. close enough.
Task-number: fatal: out of memory
Change-Id: I8c51cfffb59ccd156b46bd5c56754c480667443a
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
fixing and escaping is now a tri-state option:
- none (this removes the need to unescape the result right afterwards in
some cases)
- local shell (for system())
- target shell (for Makefile)
Change-Id: I5b78d9b70630fe4484dc964eff5f62793da35764
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
Use of FileFixifyAbsolute with non-default in_dir and out_dir
is not defined (and produces bogus results).
Using FileFixifyRelative when handling QMAKE_BUNDLE_DATA as a relative
path is fine.
Change-Id: I49902dc9f5b8029d092a4419c0cff5483e419c30
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Files with same base name cause extra trouble for debuggers.
Similarly to 26cf0f0d5a, it can be avoided here.
Change-Id: Ide47170f417d636aa031b0d84a1951df8bf32316
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Use correct path for executable file.
Change-Id: I50283fc43fe6561cb8eb687f739f21a3d5cbadbd
Reviewed-by: Christian Stromme <christian.stromme@theqtcompany.com>
Same change as was already applied for count() and contains().
Change-Id: Ibd62e4b36e03741993ba33e730c9449ef19bff5f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
If QList data-layout-compatible with QVector and a C array, implement count()
via std::count() and contains() via std::find() and use const_pointer
instead of const_iterator as the iterators. This essentially makes
the QVector and QList implementations identical to each other, at least for
important cases such as QString.
To switch between the different implementations, use tag dispatching instead
of the previously used technique "use 'if' as if it were 'static if'", which
imposes accidental requirements on the element types (something that esp.
QVector is plagued with).
Change-Id: I6caf74442a22059676b5bf115a6089768f3a0952
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Add one tag class for each of QList's three different memory layouts
to QListData, and inherit QList<T>::MemoryLayout from exactly one of
them.
To simplify overloading, added tag classes that express the negation
of the two extreme poles of memory layout (C-compatible and heap),
too. The "missing" one could be added when needed, too.
Change-Id: I45ea603731499fd3fdfb37d60a0a98fb22ac15ec
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The Movable type, surprisingly, was as large as sizeof(void*), so on
32-bit platforms, we were not testing the QList memory layout where
items are placed directly into the void*-slots, but are too small,
leaving gaps.
Fixed by making sure that Movable is smaller than void* and adding
a variant of Movable, Optimal, that is guaranteed to be the same
size as void*, and replacing the int tests with tests of Optimal.
Had to demote the State variable to uchar, since MSVC will apparently
not make a collection of bit-fields smaller than the largest type used
in any of the consituent bitfields, so State s : 8 wouldn't work.
Change-Id: I4f0e24bd6928f076b4ce60e8d977d5d98a724161
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Instead of using a QString::arg() cascade, which creates tons of
temporaries, use good 'ol sprintf().
Effects on Linux GCC 4.9 stripped release builds:
text -216B
data +-0B
relocs +-0
Change-Id: I6ff551cb9f42e0c05a64f03a8e177fb527915481
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Instead of using a QString::arg() cascade, which creates tons of
temporaries, use good 'ol sprintf(). As a consequence, this
function is now inlined into all four callers and the total
executable size _still_ goes down:
Effects on Linux GCC 4.9 stripped release builds:
text -420B
data +-0B
relocs +-0
Change-Id: I10d6abd94b489db7c2f01dc5424f30a798602522
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
...by using var op= ... instead of var = var op ...
No change in executable code size.
Change-Id: I1c29ff6700f0f21be07768af8d002f0823c89fbd
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
For any 1-bit flag:
bool(var & flag) <=> (var & flag) == flag
but gcc didn't seem to get it:
(Surprising) effects on Linux GCC 4.9 stripped release builds:
text -4936B (!!)
data +-0B
relocs +-0
It seems this enables some dead code detection, but
I must confess I don't quite understand how such a
small change can have such a dramatic effect on the
executable size, even after diffing the assembler
output.
Change-Id: Ia307fde0de16160ea51bbb3ed6c1ff203d4f9091
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The compiler-generated one is just as fine, more maintainable,
and doesn't inhibit moves (which probably doesn't matter here).
No change in executable code size.
Change-Id: Ideee493a5911808430d3e09e6eb07e91d7a19b12
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Compilers *really* don't like out parameters.
(Impressive) effects on Linux GCC 4.9 stripped release builds:
text -2512B
data +-0B
relocs +-0
Change-Id: I0fe370a438f7b82aaa9cc04ddd56e45a5969e7a9
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Get the captured texts once and use indexing into the QStringList
instead of repeatedly calling QRegExp::cap(n).
(Impressive) effects on Linux GCC 4.9 stripped release builds:
text -2876B
data +-0B
relocs +-0
Change-Id: I3a02eab1a691f31c30654cd89a0c030414b40de0
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Compilers don't like out parameters.
Effects on Linux GCC 4.9 stripped release builds:
text -528B
data +-0B
relocs +-0
Change-Id: I32ee1a6c4388900bacfc6eb20feb4b81d71cb1f2
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The /tmp directory doesn't exist on Android, and the test needs
a directory that it can cd into which is one level above root.
So we just use /system which should be available on all devices.
Change-Id: I8e6a15f278429491fd871f87af497e5d7184ddf8
Reviewed-by: BogDan Vatra <bogdan@kde.org>
When the default font has the pixel size set instead of point
size, then pointSize() will return -1 and the output from the
HTML will include the pixel size instead. This happens e.g.
on Android when we extract font information from the system.
Change-Id: I26dc49fff215c9887cf0b78dcfff88df0f74450d
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
QSet, as a node-based container, requires one memory allocation per element
inserted. QList, as a contiguous-memory container (at least in the case of
a QByteArray payload), requires one memory allocation per container.
The higher lookup speed might still speak for using QSet, but there are only
two uses of the sets:
1. Checking for existence (or lack thereof) of timezone names.
For this, first generating a container full of data just to check for
existence of one item of data is extremely wasteful. The QTZPrivate
API should be extended to allow said lookup to be performed on the
native data store instead.
That leaves
2. Returning a sorted(!) list(!) from the public QTimeZone API.
There is no reason why, during the construction of those sorted
lists, the data should be held in a set. Instead, the well-known
technique of first cramming everything into a result container,
which is subsequently sorted and has its duplicates removed,
can be used here.
Saves more than 8K of text size on AMD64 stripped release builds.
Change-Id: I71c2298e94e02d55b0c9fb6f7ebeaed79a1fe2db
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
We where allowing conversion from jobject to QJNIObject without taking
ownership of the jobject. Since we are managing the JNI environment we
should not allow conversions without having the option of taking
ownership of the local ref. This is now done by making the conversions
explicit, i.e., local refs are converted through
QJNIObjectPrivate::fromLocalRef() and global refs through the
QJNIObjectPrivate's jobject constructor.
This change breaks SC, but the API is private and no usage have been
found outside QtBase.
Change-Id: I3175f171699ec3f8e65144aaebc6246bc6e5bb4d
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@theqtcompany.com>
wait() was always returning immediately, regardless of the timeout
value, due to a timespec comparison from different clock types.
On Android 5.0, qt_gettime() uses the monotonic clock but the wait
condition was using the real time clock.
__pthread_cond_timedwait_relative() is not exported anymore in
Android 5.0, we therefore fall back to pthread_cond_timedwait().
Since the monotonic clock is now available, qt_gettime returns
a time based on it. The wait condition consequently needs
to use the monotonic clock.
Change-Id: Ie7cf909b81107edd7207c3c039b3ec1f5422303f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@theqtcompany.com>
Most of the time the widget passed in to finish() will not yet have a
window handle so it would just call close() straight away. This
way it ensures has a window handle.
Task-number: QTBUG-35757
Change-Id: I2e212a2ea496cc4e4bba57a8c2d0d127afc43e48
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Reviewed-by: Thorbjørn Lund Martsum <tmartsum@gmail.com>
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
If we call them just in the constructor all the signals they sent
out can't be connected and will be lost, particularly this means
the QNetworkConfigurationManager doesn't see my ethernet connection
and thus thinks i'm not online
Change-Id: I1480f76338d6ae4fbed676f9fa40ada18ea431ad
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
The problem is that the operator<< was taking a non-const reference to the QDebug
object. This causes a problem as all other operator<< return a temporary.
Since every other roperator<< takes the QDebug by value, we should also take it by value
in this case.
Move the operator<< in qdebug.h because i don't want to #include qdebug.h from qobject.h
And move the qt_QMetaEnum_debugOperator to be in the corresponding .cpp
Task-number: QTBUG-44462
Change-Id: Ia01629224c58930c2997e767efc43de90d6309e2
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
The EGL integration should not call Xlib functions in this case.
Change-Id: I46656b12d603ca1b4b0b41f34c3c0e8410c91db8
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
Reviewed-by: Jørgen Lind <jorgen.lind@theqtcompany.com>
This makes it easier to debug QJsonDocument binary bugs.
Note that the last character is never printed as printable, but by way
of construction it's always binary data anyway (the offsets table is at
the end after parsing JSON sources).
Change-Id: I8a7a116f51864cecb52fffff13bc24ad01ad8a49
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
The variable c is unsigned, so the second operand is unsigned,
constraing the -1 to be unsigned too and causing a change of sign.
Instead, cast the middle operations to int, as that's the return value
anyway.
error #68: integer conversion resulted in a change of sign
Change-Id: I1a800c709d3543699131ffff13c2fd79f14f8b43
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
This is the right way:
void (*getProcAddress(const QByteArray &procName)) () Q_DECL_OVERRIDE;
However, ICC and GCC do not get it. They expect it in the form:
void (*getProcAddress(const QByteArray &procName) Q_DECL_OVERRIDE) ();
ICC complains with an error, so this needs a workaround. GCC ignores the
incorrectly-placed "override", so it just looks to GCC like no "override"
was added.
See also: https://groups.google.com/a/isocpp.org/forum/#!topic/std-discussion/IjVB8CWiS5A
Change-Id: I1a800c709d3543699131ffff13c2ffc9cee46245
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
The mime type stuff generates one of the biggest translation units in QtCore
due to the compressed 1.7MB freedesktop.org.xml resource.
With QT_NO_MIMETYPE, libQt5Core.so is almost 400Kb smaller
(4.8MB->4.4MB gcc 4.9 stripped release build)
Change-Id: I5339090994034355724ff4deddb64720e81baeaf
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Reviewed-by: David Faure <david.faure@kdab.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Including math.h can pollute the default namespace, and break
some compilers if cmath versions of the method are declared as using.
Switching to C++ math functions also greatly simplifies handling of
float qreal as C++ automatically chooses the right method.
[ChangeLog][QtCore][QtMath] qmath.h no longer includes math.h, so any
sources depending on that indirect inclusion may fail to build.
Change-Id: I4d0e331dafba354ec05dc5052e61ef4ff8d387fe
Reviewed-by: Rafael Roquetto <rafael.roquetto@kdab.com>
Qt Multimedia can be compiled with either GStreamer 0.10 or 1.0.
0.10 takes precedence over 1.0 if both are available (1.0 will be used
by default in Qt 5.6).
Auto-detection can be overridden with -gstreamer <version>.
Change-Id: I74d58d2c146c842902375b4d1e5e6e96d32addac
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Do not return early on Windows if bytes are still in the buffer.
This is not in line with the behavior on other platforms,
and also breaks apps where the caller knows that the
bytes available are insufficient.
[ChangeLog][Network][QLocalSocket] On Windows, waitForReadyRead
now always waits for more data, even if bytes are still in the
buffer.
Task-number: QTBUG-16688
Change-Id: I1425a5780c7707295374934a6b5446ff8e148cc8
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
We must do something when requestUpdate() is called on a QWidgetWindow.
The semantics of UpdateRequest for QWindow and QWidget are unfortunately
different: for widgets an UpdateRequest means "sync the backing store".
For QWindow it also involves marking as dirty.
Change-Id: Idf40b3fc0873652dc081edeb12c96b3007a126ef
Reviewed-by: Morten Johan Sørvig <morten.sorvig@theqtcompany.com>
Reviewed-by: Jørgen Lind <jorgen.lind@theqtcompany.com>
the qt headers live in the source dir in this configuration.
instead of hard-coding the path in the project file, use the correct
module variable. this requires harfbuzz-ng to be built after corelib.
Change-Id: If1b64b59a0939d3b5190331fcf70da566d4eeedd
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Whenever the selection changes it will call panelSelectionDidChange
even if the panel is not visible which causes it to overwrite our
chosen initial file with "untitled". Therefore we only set the name
when the panel is actually visible.
Change-Id: I09f7ebabce5c2139a7f74b953391d39dfad65db2
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
We don't receive ButtonRelease event after closing a popup
by clicking outside of the client area. Thus the internal
state of mouse buttons in the xcb plugin becomes outdated
until we receive ButtonRelease event.
This commit updates the internal state of mouse buttons
from MotionNotify events. So when a user will move a mouse
on the client area, the xcb plugin will send a mouse event
with updated buttons to Qt Gui and QGuiApplication will
detect the following mouse events correctly.
Task-number: QTBUG-32609
Task-number: QTBUG-35065
Task-number: QTBUG-43776
Task-number: QTBUG-44166
Task-number: QTBUG-44231
Change-Id: Ica334dfbf04f7ef81db86b25262328fe5da11808
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
Reviewed-by: Gatis Paeglis <gatis.paeglis@digia.com>
This change is a step closer to working with openssl 1.2 which makes
this struct opaque.
Change-Id: I3897142657edc0fa4053142b6ef743c2b00c013e
Reviewed-by: Peter Hartmann <peter-qt@hartmann.tk>
QDoc keeps a list of files it generates (or copies) to the output
directory, for writing it to a .qhp file later on.
In single-exec mode, QDoc processes several qdocconf files
without deleting the generator in between. Therefore, we need to
clear the list of files whenever the generator is initialized,
to avoid duplicating the filenames across multiple .qhp files.
Change-Id: Ibc2a6b171466aa1db6cfe3da9a820d5ba2845004
Reviewed-by: Martin Smith <martin.smith@digia.com>
