The HB face caching mechanism introduced in 227e9a40cf
wasn't complete due that fact that HB-NG doesn't parse the entire
font table at once but rather references a table on-demand.
This incompleteness caused a crash in case the engine doesn't
get cached or when it removed from the cache and then re-used.
Task-number: QTBUG-36099
Change-Id: I7816836107655ce7cf6eb9683bb5dc7f892f9cd1
Reviewed-by: Lisandro Damián Nicanor Pérez Meyer <perezmeyer@gmail.com>
Reviewed-by: Michael Krasnyk <michael.krasnyk@gmail.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@digia.com>
One less obvious part of this patch: the fontCache pointer in engineData was not
safe. It isn't safe to rely on pointer addresses to verify we're cleaning up the
right thing, as a sequence of malloc()/free()/malloc() can return the same
pointer, and nothing was cleaning up the dangling pointer in engineData.
With this, it is possible to safely drop OpenGL contexts in QtQuick under all
conditions with no possibility of crashes.
Done-with: Aaron Kennedy <aaron.kennedy@jollamobile.com>
Change-Id: I7b91384251593730124323a74737d41333a05f59
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
This reverts commit 8c58640fdf.
The intent was to reduce the maintenance of parallel code paths,
but this had a negative impact on performance so we need to
enable it again.
Change-Id: Ic58f7781ff076055e7c48f20fc0e7a8a2c9e51a4
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
The glyph returned by loadGlpyh() must be freed manually when caching
is not enabled (that is, QT_USE_FT_CACHE is not set).
This change and https://codereview.qt-project.org/#change,65672
together should stop QGLWidget::renderText() from leaking memory
extensively.
Task-number: QTBUG-32792
Change-Id: Iff7e1591a1ea994a97b6a56bc16c64bf544e6713
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
This fixes rendering in the OpenGL paint engine,
simplifies our internal code paths and also produces
nicer text for non-90 rotated glyphs
Change-Id: Ib206065cd486ac29067f9ebf4a0d74189a673f1e
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Instead of loading HB face in QFreetypeFace::getFace(), defer allocation
until the first FT font engine for that face gets initialized;
then, QFreetypeFace "reparents" and manages the loaded HB face.
Change-Id: I2ac8ead4c6ed25d341af9c9cf0c34dfb979f8390
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
Commit b5922c89ba (Add support for retina glyph-based text drawing
in the GL engine), and commit 155a20628b (Use QPaintEngineEx to decide
if font is too big for using the glyph cache) together changed the
behavior of drawing scaled cached glyphs in the GL engine, producing
blocky text drawing when using the FreeType font engine.
Whereas before we would cache all glyphs without any transform, and
let the paint engine take care of the transform, commit b5922c added
support for a scaled GL glyph cache, resulting in a 2x cache, drawn
at 1x for example. The problem was that the FreeType engine claimed
to support producing glyphs at 2x, but did that using the QFontEngine
baseclass implementations, which use a simple fast-transform to scale
up the glyphs. The result was a 2x cache with horrible looking glyphs.
The first step in fixing this issue was to have the FreeType engine
claim to only support translations. This would then make the paint
engine choose path-based drawing for all scaled text, which is slow.
To restore the optimization that we would draw 0.5x-2.0x scaled text
using a smooth-scale in the GL engine (which was removed in 155a206),
we then needed to extend shouldDrawCachedGlyphs() and add a special
condition for this, coupled with a bit of logic in drawCachedGlyphs()
that ensures we don't propagate the painter scale to the glyph cache
if the engine is not able to produce scaled glyphs for it. This
means we get the old behavior of the GL engine doing a smooth scale
of the 1x glyph in the cache.
Finally, since the raster engine also checks if the font engine
supports the current transform, but for FreeType then ends up in
a separate code path when actually drawing the glyphs (as the
FreeType font engine supports internal glyph caching), we need
to add a corresponding check for hasInternalCaching() in the
shouldDrawCachedGlyphs() function, now that the FreeType engine
only reports that it supports translations.
Change-Id: Id03de896dec5f29535b281fb235332ef018045d8
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
QFontEngineFT::recalcAdvances uses design metrics if hinting is disabled
or slight. QFontEngine::doKerning only follows the QFontEngine::DesignMetrics
flag. This means in some instances the advances will be calculated in
subpixels but kerned in full pixels.
This patch makes QFontEngineFT decide if it should request design metrics
from QFontEngine::doKerning or not.
Change-Id: Ia0236efde2d7269623f690a6074afbe26e07c458
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
QFontEngine::FaceId::filename is intended to be local8Bit-encoded
but QFreetypeFace::getFace() always treats it like UTF-8-encoded.
Also replace explicit (to|from)Local8Bit and toLatin FaceId::filename
conversions with QFile::(en|de)codeName, where appropriate.
Change-Id: Ic7beabf0a160f2f02f1667bcb8e6067adaba1c16
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
For large fonts, we would get invalid metrics due to the metrics
being too large to fit in the signed chars of the cache struct.
This in turn would lead to the bounding box reported by the
font engine being wrong, and subsequently, the algorithm which
decides whether any part of the glyph is visible in the raster
engine would make the wrong decision, hiding glyphs that should
not be hidden.
Task-number: QTBUG-30026
Change-Id: I59b613ce889fcac0bd50ffec9d369728068d0263
Reviewed-by: Samuel Rødal <samuel.rodal@digia.com>
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
This affects HB_Font and HB_Face.
As of now, the Hurfbuzz API usage is concentrated in qfontengine(|_ft).cpp
and qtextengine.cpp, thus it is a lot easier to switch to Hurfbuzz-NG.
Change-Id: Ie06959efd5d6080fe44c407d9f5de0a07dd1c210
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Josh Faust <jfaust@suitabletech.com>
We ask grayscale or mono bitmaps from FreeType but in some cases treat
the output as ARGB without conversion. This surfaces using QGLWidget as
a viewport to QDeclarative content. The offending glyphs are then
generated through QTextureGlyphCache::textureMapForGlyph. This adds a
fix for converting to the expected ARGB32 data.
Change-Id: Ia219582ebd76b7e4e9379111a42312b4d97718de
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@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>
Store the result in a temporary QByteArray and continue
working with that one.
Change-Id: I24bc243f0f3dfb37d840faf7592b3383bd37c7e2
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Reviewed-by: Gunnar Sletta <gunnar.sletta@nokia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
These flags are specific to font engine(s) and has nothing
to do with the text engine or the text layout.
Change-Id: I4bb793c3c634b3cf0ae0a8a8c23b946fad5874b6
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Scaling hinted glyphs looks ugly, and it makes smooth scaling
animation not possible. Since nothing will work as intended
in hinted mode, we should disable hinting automatically when
glyphs are loaded with scaling transformation.
Task-number: QTBUG-24846
Change-Id: Id7fb5f5bdc2d00be157b0c5d55c316473571473c
Reviewed-by: Thorbjørn Lund Martsum <tmartsum@gmail.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@nokia.com>
For high resolution or extremely large font sizes, the advance
cached here is likely to overflow, since FreeType returns 26.6
fixed point value and we only take signed char here for advance.
In those cases we should skip caching because there won't be
that many big glyphs after all.
Also move the metrics caching block a bit down to take glyph
embolden and oblique into account.
As a result we also don't need to increase the linearAdvance
size because any linearAdvance less than 128 should fit in
the old 10.6 fixed format.
Change-Id: Ic4920ada49954ce1e0a8673c9f33f30e385e3046
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@nokia.com>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
The only use for storing charset loaded by fontconfig in font engines
is for determine if a font supports certain codepoint, however FreeType
already does that. The charset loaded is sometimes not complete, for
instance in fontconfig 2.9.0 it removed Apple Roman platform support
for cmap loading, thus results a regression in common symbol fonts
(Wingdings, Webdings) rendering. Because those symbol fonts produced
by Monotype only contain two cmap tables: Apple Roman and Microsoft
Symbol, since the Microsoft Symbol table has a weird 0xF000 offset,
we always fallback to the Apple Roman cmap table.
Removing freetype charset cache also make each font engine consuming
less memory.
Change-Id: I88f3f44981f3a1c517b84809a3f5b834ffff7681
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
If we only need to access the font metrics (like in scenegraph) for
layout, we don't need to render glyphs with FreeType at all.
Increase cached linearAdvance to 22 bits so that no overflow will
happen: FreeType returns 16.16 fixed point linearHoriAdvance, but
QFixed is 26.6, we store
glyph->linearAdvance = linearHoriAdvance >> 10
Apparently 'short' is not enough since it's only 16 bits.
Change-Id: Id14eafa19f01a687de11997526281f9e7e860482
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
Cached font was used regardless of the format, resulting in incorrect
advance in some cases when default format differed from the cached
format.
Task-number: QTBUG-24188
Change-Id: I39e4156bd9ba743afa7e106e934c90227fbf2b8b
Reviewed-by: Jiang Jiang <jiang.jiang@nokia.com>
Report https://bugs.webkit.org/show_bug.cgi?id=79561 shows that
this commit will cause significant slow down in text handling
operations. Though the exact reason is unknown we should revert
it first and find out the reason later.
This reverts commit 692064bcfd.
Change-Id: I16a56c3093bdfa2119ab6a6e9049ef2925468e29
Reviewed-by: Simon Hausmann <simon.hausmann@nokia.com>
Neither rendering with outline nor fetchMetricsOnly requires the
rendering from FreeType so we don't need to render them or cache
it. It should speed up recalcAdvances() quite a lot.
Change-Id: I0f623cb4f79da2edf6e9c9634a2f22fb0c66823c
Reviewed-by: Samuel Rødal <samuel.rodal@nokia.com>
Historically, we've calculated font height as ascent+descent+1.
In Qt 4, a patch was added to work around this by subtracting
1 from the descent of the font engines. We now remove the +1 and
the work arounds.
Change-Id: I7e25d49b97ac892015d3328f32d70eb9a7c2d88f
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@nokia.com>
Reviewed-by: Lars Knoll <lars.knoll@nokia.com>
Don't indirect over internal API.
With QAbstractFileEngine and friends on the way out, QFile is losing its
(marked internal) virtual fileEngine() function and the file engine
can't be queried for LocalDiskFlag. That information is now available
through QFileInfo, instead.
Change-Id: I48827a96fd8cd748055ad5f075912fc8e1c5ef7f
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@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>
QFontEngineFT::cloneWithSize, which is used among other things by
QRawFont::setPixelSize, should create a new font engine using the font
definition from the current font engine.
Change-Id: I9415422bbecc2bf7b7acf105c12c1c83a894526c
Reviewed-by: Simon Hausmann <simon.hausmann@nokia.com>
glyphMargin() support for QTextureGlyphCache is implemented in
respective font engines, thus this function is platform dependent.
Before Qt 5 the code is guarded in macros like #ifdef Q_WS_MAC,
now we should move them into QFontEngine and its subclasses.
So far only Windows font engines support it. FreeType and Core Text
based font engines all ignore it.
Change-Id: Ia14016533d8fbfaacf848a7d3bc928f8197318f5
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
So that it won't just return "Helvetica" that may not exist in
target system. Providing it seems to fix QTBUG-21833 as well,
thus we can remove the QEXPECT_FAIL statement in qlistview test.
'_' appears to reach the minimum right bearing in some fonts.
Task-number: QTBUG-21833
Change-Id: I2340af6ee736083ec4f6575e469840ebec4e545e
Reviewed-by: Jiang Jiang <jiang.jiang@nokia.com>
It appears that the symbol_map we obtained with FreeType doesn't
work for common symbol fonts like Wingdings, in previous X11 code
we use the code path with normal charmap, to enable it we need to
make sure QT_NO_FONTCONFIG is not defined. And since the FcCharset
functions doesn't really operates on the font database, we can
safely skip the mutex calls.
Change-Id: I8c484de200be1d47c053b10be2c51d2273dcd359
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
Move subpixelPositionForX from QTextureGlyphCache to QFontEngine,
since some font engines like QFontEngineFT may need a custom
implementation or tweak it a little bit.
In QRasterPaintEngine::drawCachedGlyphs, do not add aliasedCoodinate
to x offset as that will break subpixel positioning.
Change-Id: Idbcec617509459b80965220ceb07b17737649bbf
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
If the face is unlocked in this case, calling unlockAlphaMapForGlyph after doing
the operation will cause an assert.
Change-Id: Icb2310a3b139e5bedf24f9c3794bd7d3fd38cfea
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
Before commit 070d9c00c4, the raster paint engine used
to call loadGlyphs() on the FT engine, which ensures that the glyphs are loaded
and (more importantly) that the correct transformation matrix is applied on the
freetype face. After commit 070d9c0 lockedAlphaMapForGlyph is called for each glyph,
which unfortunately doesn't respect the provided transformation matrix. Therefore
when drawing scaled text, it was never actually scaled.
This patch applies the combination of the fontengine transform and the requested
transform on the freetype face actually used, similarly to loadGlyphs.
Change-Id: I0956a9e71784582db6bb90475a001a63800773f4
Reviewed-by: Jiang Jiang <jiang.jiang@nokia.com>
As windows' fontdatabases also uses functionalities from
qbasicunixfontdatabase it was renamed to qbasicfontdatabase. But instead
of iterating over the font directories' files it uses system calls to
obtain the list of fonts and uses registry values to find the according
filenames to add.
The native font engine was still kept. It can be activated by adding
fontengine=native as platformargument.
Change-Id: I7197bed5d18b8a33d4aa97ce91bfa1cd281b80ea
Reviewed-on: http://codereview.qt-project.org/5839
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
Sanity-Review: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@nokia.com>
This is a regression caused by 070d9c00. If the defaultFormat
is set to Format_None, we should default to Format_A8 like before
the change.
Change-Id: I8ad07abff96fc1dfa03c0c5b49d13beb36b9de70
Reviewed-on: http://codereview.qt-project.org/5540
Reviewed-by: Qt Sanity Bot <qt_sanity_bot@ovi.com>
Reviewed-by: Jiang Jiang <jiang.jiang@nokia.com>
Konstantin Ritt
J-P Nurmi
Robin Burchell
Gunnar Sletta
Oswald Buddenhagen
Laszlo Agocs
Tor Arne Vestbø
Allan Sandfeld Jensen
Eskil Abrahamsen Blomfeldt
Frederik Gladhorn
Sergio Ahumada
Florian Hänel
Iikka Eklund
Sergio Ahumada
Marc Mutz
Jiang Jiang
Miikka Heikkinen
Eskil Abrahamsen Blomfeldt
João Abecasis
Kai Koehne
Jason McDonald
Pierre Rossi
Friedemann Kleint
Oliver Wolff
Simon Hausmann
Olli Werwolff