Previously when a new QOpenGLFramebufferObject was bound, the
QOpenGLContextPrivate::qgl_current_fbo member was also updated to point
to this new object.
But if a user called QOpenGLFramebufferObject::bindDefault(),
qgl_current_fbo was not unset, meaning that if the FBO object would be
deleted at some point, qgl_current_fbo would be a dangling pointer.
This patch makes sure to clear the value of qgl_current_fbo when
bindDefault() is called. It is cleared, and not set to point to another
object because the default platform OpenGL FBO is not backed by a
QOpenGLFramebufferObject.
Task-number: QTBUG-56296
Change-Id: I68b53d8b446660accdf5841df3d168ee2f133a90
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
If a global shared context is created, and afterwards a new default
QSurfaceFormat with a different version or profile is set, it can
lead to issues with context sharing.
The documentation for QSurfaceFormat::setDefaultFormat mentions
this. It would be helpful to actually show a warning in case it
happens.
Change-Id: I71f7ce95496e1ecbfc6a0c7d7bed146ef8dc351e
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Clip the transformed and rounded sourceClip to the source rectangle,
so we don't try to rotate pixels outside the source.
Task-number: QTBUG-56252
Change-Id: Ib9cb80f9856724118867aea37ead0b02a6c71495
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
It was introduced in Qt 4 by the commit
13a31fe82845f8b1f4d86919080d3b2a87c4d061 and was unused
even there.
Change-Id: I5f3861918ea1f443f7e13439fafa067f2b28a91a
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
QWindows that aren't created, and therefore don't have a platform window
attached, should not be treated as "normal" windows. The expectation,
in our code, is that these windows won't get their geometry updated by
the platform plugin, and that the geometry is only changed by the owner.
In QQuickWidget's case this was causing the scene to be rendered
incorrectly.
Task-number: QTBUG-50973
Change-Id: Iea62dfb7fa90cbe450a662c5792c7c4f49c991fd
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
A gradient table may be deallocated while in use because we don't keep
track of references. To fix it we now reference count the cache entries.
Task-number: QTBUG-14614
Change-Id: I772ebf565ccf41d476811ca9a51b721f10de8aeb
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
... and avoid detach()ing potentially large data for just
preserving the QPlatformPixmap::pixelType().
A QBitmap differs from a QPixmap (its base class, urgh)
by always having a data != nullptr and a Bitmap pixel
type, yet load() was unconditionally setting 'data' to
nullptr on failure, turning a QBitmap into a non-QBitmap.
Fix by move-assigning a null QBitmap instead of resetting
'data'.
Add some tests.
Change-Id: Ida58b3b24d96472a5f9d0f18f81cc763edcf3c16
Reviewed-by: Anton Kudryavtsev <a.kudryavtsev@netris.ru>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
After e109b8a0f3, it is possible
that the cache will be flushed as a result of inserting a new
font rather than just when the timer event triggers. When doing
superscript and subscript text layouts, we would first get
a regular font engine, then a scaled one, and then reference
the regular font engine *after* getting the scaled one. If the
regular font engine was deleted as a result of inserting the scaled
one, we would get a dangling pointer and crash.
The situation was improved by 49926bb9ef.
You would now to switch between 256 different fonts in the layout
in order to trigger it. The test in the commit will trigger the
crash even with this change.
[ChangeLog][Qt Gui][Text] Fixed a crash that could happen if you
were doing many different text layouts with different fonts
and superscript or subscript alignment.
Task-number: QTBUG-53911
Change-Id: Ia33108252e030eff25924ef1b7c10b9d59b5bc8c
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The Windows Intel drivers reject shader sources without
a version directive in 3.2+ compatibility profiles. This
is odd but can be worked around by adding #version 110
(which should be the default...)
Change-Id: I1ccac41b80121e6423d4f8964d03dda52a433296
Task-number: QTBUG-55733
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Writing to solid overwrites part of the gradient input messing up later
lines. In particular repeating gradients tended to be broken.
Tested by lancelot (once the baseline is correct).
Change-Id: I64222048ba67b0424b44822f09ddc947973145a6
Reviewed-by: Eirik Aavitsland <eirik.aavitsland@qt.io>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Windows of type Qt::ToolTip should not prevent closing the
application when checking for the last window.
Task-number: QTBUG-55523
Change-Id: I47c51abe79bf6c857aba229960de95e1a23efb10
Reviewed-by: J-P Nurmi <jpnurmi@qt.io>
Text is copied from corresponding QFontMetrics methods.
Change-Id: Ife79e0d1b06ca3f691f2fd8bd796b41aeaa76954
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
As the d-pointer, QBrush uses a QScopedPointer with a
custom deleter that checks for QBrushData::style and
casts the QBrushData pointer down to corresponding
subclasses before calling delete on them.
In QBrush::detach(), however, any of the three brush
data classes were held in a QScopedPointer _without_
the custom deleter, invoking UB when that scoped
pointer would ever get to be the one that deleted
the payload instead of handing it over to the objects
d-pointer.
Found by making dtors protected following a Coverity
report wrongly marked as 'dismissed' (these static
checks are not included in this patch, since they
are binary-incompatible), to find out where Coverity
could possibly see a problem.
Also replace the d.reset(x.take()) that allowed this
mismatch to compile with d.swap(x), which nicely
ensures that x and d are of the same type.
Coverity-Id: 11772
Change-Id: I85e2c205df9291bd7508b6c90f7b03fbe8c3bcd2
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The calculated end position of script lines, consisting of nothing but
trailing spaces, is lower than the start position. This results in not
shaping those script lines.
Task-number: QTBUG-55255
Change-Id: Ib70c1a800d2f70b7f61e3d05c10618e275f45f12
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
An entry was missing in one of the tables. This patch adds it back in
and adds the enum comments used to keep track of whether they all are
there and correctly set.
Change-Id: Ic6a55a8f81f9c42a3174a2a75c80c3a354f173b7
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Otherwise we will return an empty list of fallbacks if no QFontDatabase
has been created yet.
Task-number: QTBUG-55222
Change-Id: I50508162fad3206e0acf3cc6eb39aefac5c3e197
Reviewed-by: Peter Varga <pvarga@inf.u-szeged.hu>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
When the pixmap is initially from a higher device pixel ratio and it is
being used on a window that does not have a device pixel ratio other
than 1.0 set then the pixmap should also have it set to 1.0. This ensures
that the size of the pixmap is preserved and it is not scaled down as a
result on the normal display.
Change-Id: Ie5d96b3e1508867b723000bea182c8157640af02
Reviewed-by: Morten Johan Sørvig <morten.sorvig@qt.io>
[ChangeLog][QtGui][QTextDocument] Fixed a bug that would return a wrong
position when searching backward from the end of the document.
Task-number: QTBUG-48182
Change-Id: I6e88f808a50cb840f61e7bc579e2a28c5300089d
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
Passing a string literal to QWindow::setObjectName() is rude:
it wants a QString.
Change-Id: Ic4c1079889002f0d5f1028c443456a8be0b7b0fe
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
qt_div_255 does a fast rounded division, the form of rounding is however
only valid for positive integers.
Correct one case where it was used incorrectly and add it one place
it would be valid, and adds a comment to the function.
We were using the optimization (x*a + y*(255-a))/255 ==
(x-y)*a/255 + y but that makes the division by 255 potentially
negative.
Change-Id: Ie53aa82b66ef801f5d43f8d2ec48880cb4972f69
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Remove QWindow instance from QGuiApplicationPrivate::modalWindowList
in the destructor.
Task-number: QTBUG-54566
Change-Id: I1f07fb46a371f69f04907b20657f3b05571445bd
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Gabriel de Dietrich <gabriel.dedietrich@qt.io>
This improves code generation, since the compiler statically knows
the number of loop iterations. But it also fixes a compile-error
on Clang with a following change of the name field from char* to
char[].
Change-Id: I7ef18adf3cb9b34cd1b7235cb35cf26b7e349d92
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
The limitation of four glyph caches per font engine was to prevent memory
spikes during rotation of text. But in an application with multiple top
level OpenGL windows, each with their own context, we'd end up trashing
the list of glyph caches during rendering, even if each window just drew
the same static text.
Having a shared context between the windows helped a bit, but had other
performance issues due to the globally shared state, so the better approach
is to limit the caches to four per context.
This brings the multiwindow manual test from a grinding 4fps on macOS
to a smooth 60fps for 20 concurrent windows.
Task-number: QTBUG-52372
Change-Id: I26edd5f6edb5c7818e14b2203af062df19ae7127
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
Same change as in QColor (d38f86e50b).
relocs: -239
text: +2248B
data: -3872B
(optimized GCC 6.1 Linux AMD64 build). The text size increase means
we pushed the data into an immutable section.
Change-Id: Iad10c877d4a4877878dded56a7ef1e14ff92c996
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Same change as in QColor (d38f86e50b),
but much more dramatic effect, due to 5x more elements in the array:
relocs: -657
text: -3416B
data: -10528B
(optimized GCC 6.1 Linux AMD64 build)
Change-Id: Ia266ab68f2d309743374ac2034a69f58bf556adf
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
This is a runtime setting, as the user's freetype version can differ from
the developer's, and freetype already safely handles it internally when
not available.
Task-number: QTBUG-55345
Change-Id: I26e73728196d60ae26e5f1919ecd0dadac393890
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
QImage may have a null d-pointer, and convertToFormat_helper was the
only method not protected against it.
Change-Id: Ibfe5b139e3922f2a0c8284c6ae1932c2150efc66
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
The code for handling of the rarely used "restore to previous" frame
disposal option would mistakenly clear the entire existing frame. It
looks like this was a copy/paste mistake in the code; the intention
obviously was to clear the newly created backingstore instead.
This fixes the rendering of
http://media1.giphy.com/media/9Jevgp0CgvAnm/giphy.gif
Task-number: QTBUG-55141
Change-Id: I4d344b7733edd62346656154e215c21727f9b6bb
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
As of version 10.12 (Sierra), the name of Apple's desktop operating
system will be macOS. Change the occurrences where the Mac platform
is discussed to use a macro \macos, which expands to 'macOS'. This
helps with adapting to future renaming.
Update the instructions on mac-specific Q_OS_* macro usage.
Add a \target for the old 'Qt for OS X' topic to keep links working
for other documentation modules that try to link with the old name.
Change-Id: Id33fb0cd985df702a4ae4efb4c5fd428e77d9b85
Reviewed-by: Leena Miettinen <riitta-leena.miettinen@qt.io>
This doesn't cover FBOs bound directly with glBindFramebuffer(),
but it's perfect to create a fast path for code we know uses QOGLFBO,
thus avoiding expensive glGetIntegerv() driver calls.
The use case is to use this in QSG24BitTextMaskShader::activate(), where
we need to check if the current FBO is sRGB capable.
Change-Id: I434eeeb7e6a3d16be9327315536ad7280245085d
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Reviewed-by: Sean Harmer <sean.harmer@kdab.com>
26238aca8c causes double deletion of the
QTouchDevice in case the post routine already cleaned up the list by the
time the touch handler gets to do it.
Just check the list of devices to see if the one we hold is still there.
If not, the pointer is likely to be a dangling one so do nothing.
This will avoid dying with bus error or similar on application exit.
Task-number: QTBUG-51562
Change-Id: I50c1edee7405aad308274538219698388c2cc9f9
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
Before: HexRgb: 0.00230 ms per iteration, HexArgb: 0.00290 ms per iteration
After: HexRgb: 0.00051 ms per iteration, HexArgb: 0.00061 ms per iteration
This showed up as a relevant optimization when profiling KIconLoader
which uses QColor::name() as part of the key -- thanks to Mark Gaiser for
the investigation and first suggestion of a solution. I have also seen
customer code writing a replacement for QColor::name() because it was
too slow to be used as a hash key.
Change-Id: I009ccdd712ea0d869d466e2c9894e0cea58f0e68
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
The unneeded ';' triggered warnings in pedantic compilation mode.
Change-Id: Id2324823e138560bb25234306601253d7bbd713e
Reviewed-by: Richard J. Moore <rich@kde.org>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
A constraint ensuring we do not sample beyond the current scan-line
was missing in the SSE2 optimized sampling.
Discovered with lancelot.
Change-Id: Ib0ece8f8bfaa034733873dc5b8baaaad5d4c0380
Reviewed-by: Erik Verbruggen <erik.verbruggen@qt.io>
Also make use of qdoc's \note command.
Change-Id: I276300cfcfde06e82b04793dbf25df8ec73e9838
Reviewed-by: Leena Miettinen <riitta-leena.miettinen@qt.io>
Reviewed-by: Topi Reiniö <topi.reinio@theqtcompany.com>
Instead of storing a pointer to a string, store the string
in the RGBData struct. It's not as efficient as in other
such cases, because one string is particularly long, but
it's still more than acceptable.
Text size increases slightly, but data size decreases a lot
(can't say by how much, exactly, as I'm on a UBSan build).
Change-Id: I1df2985fd1ebfccd84b48315d8d319dd9e25c8e7
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>
The Freetype cache was almost completely disabled by
134c6db858 because after that
change, the lockedAlphaMapForGlyph() function would no longer
cut off early for empty glyphs like spaces, but rather go
through all alpha map functions before it realized that there
was nothing to render. This would in turn invalidate the cache
for every empty glyph, causing all glyphs to be rerendered for
every isolated word.
This change adds back a cut off. This is only needed in the
lockedAlphaMapForGlyph() function, since the superclass implementation
of the other alpha map functions already contains a cut off for
width/height == 0.
[ChangeLog][Qt Gui][Text] Fixed a performance regression in Freetype
engine that was introduced in Qt 5.5.
Change-Id: I381285939909e99cc5fb5f3497fecf9fa871f29a
Task-number: QTBUG-49452
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
hjk
Alexandru Croitor
Alexandru Croitor
Alexander Volkov
Allan Sandfeld Jensen
Christian Strømme
Frederik Schwarzer
Marc Mutz
Eskil Abrahamsen Blomfeldt
Laszlo Agocs
Friedemann Kleint
Liang Qi
Konstantin Tokarev
Frederik Gladhorn
David Schulz
Jake Petroules
Andy Shaw
Tor Arne Vestbø
Samuel Gaist
Edward Welbourne
Eirik Aavitsland
Topi Reinio
Sérgio Martins
Jesus Fernandez
David Faure
Kai Pastor
Kai Koehne
