The DirectWrite font backend is an optional backend which
is planned to take over as the default on Windows. In order
to do this, though, a few gaps need to be filled in order
for it to pass all autotests.
The following things are covered by this:
1. Bitmap fonts are unsupported in DirectWrite. We enumerate
these using GDI and fall back to the GDI font engine when
loading them. As part of this, we introduce a new handle
type for fonts on Windows which can represent both
the DirectWrite and GDI engines.
2. "Legacy font names" where sub-family is embedded in the
family name is now enumerated together with the
typographic font name.
3. The DirectWrite font engine was not loading kerning pairs
from the font, like the other engines (omission which was
detected by the test)
4. Turning off antialiasing does not work with DirectWrite, so
we fall back to GDI for this.
5. Loading supported writing systems from application fonts
was not supported.
Task-number: QTBUG-119420
Change-Id: Icf6c351afb0d7487b2f4634199088d701a324aae
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Since 3f4d6279c4b0d04422efff478a5e2fb36259dbaa (khansen 2005) QMovie
calls QImageReader::read() until QImageReader::canRead() returns false.
That's apparently to support animated image formats in which the
frame count is not known at the beginnning (i.e. not specified in
the image format's metadata). But non-animated multi-frame formats
are expected to return valid values from QImageReader::imageCount();
and those also tend to keep returning true from canRead() regardless
of how many frames have been read (the interpretation of canRead()
is "does the file have the expected format?" rather than "are we
about to read too many frames?"). So, when a multi-frame image is abused
as an animation, QMovie was able to keep reading the same frame
repeatedly and pretend that the frame sequence goes on forever.
It also tended to read frames as fast as they could be decoded,
because nextImageDelay() is not usually provided, because multi-frame
image formats don't specify a frame rate in their metadata.
So now we change QMovie's behavior for image formats where
QImageIOHandler::supportsOption(Animation) returns false:
trust imageCount(), but not do not trust nextImageDelay().
But to actually jump to the next frame in this case, we also need to
call QImageReader::jumpToNextImage().
Altogether, this makes QMovie support "flipbook" animation for
multi-frame image formats, such as tiff and pdf.
Added "read frame x of c" logging in qt.gui.imageio category.
For testing, we use a pre-existing multi-frame Obj_N2_Internal_Mem.ico
file, to avoid depending on the tiff plugin.
[ChangeLog][QtGui][QMovie] QMovie now handles non-animated multi-frame
image formats (such as tiff): QImageIOHandler::imageCount() is observed,
and the default frame rate is 1 FPS.
Pick-to: 6.5 6.6 6.7
Fixes: QTBUG-117429
Change-Id: I6dad2a684e12c78a68288402e223a59427bf649e
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
2f6fe3a268 has made calls to
QApplicationPrivate::setActiveWindow() redundant.
Remove redundant calls.
Task-number: QTBUG-121488
Change-Id: I13ea8c0994eaeb764462af23ef66ea1a37659b77
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
Qt 5 streams cannot handle QBitArrays with more than INT_MAX bits,
even on 64-bit platforms, because of interface constraints (size_type
int).
Qt 6 can, so make sure to refuse serialization of oversized QBitArrays
to Qt-5-compatible streams.
[ChangeLog][QtCore][QBitArray] Now refuses to stream a QBitArray with
size() > INT_MAX to a Qt-5-compatible QDataStream.
Pick-to: 6.7 6.6 6.5 6.2
Change-Id: I263e27bd366757c8e0360dfd337948c44d00647a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
QCOMPARE() can report enum values by name just fine, no need to
laboriously convert them to strings. While comparing all tags in one
go did allow a more comprehensive report, it's enough to know we
failed; this is testing cross-platform code, so a debugger can tell us
all those extra details if we get a failure.
Testing qHash() doesn't distinguish equal things is fairly low value;
at least avoid duplicating the construction of the reference value.
Replace a bunch of other QVERIFY()s with the new cousins of QCOMPARE()
for ordered and different comparisons.
In the process, mark some of the QLocale objects as const.
Change-Id: Ic93b8ed60c6f2cc846fbba428983778896d61291
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
They were expanding as simple blocks, so their uses didn't end in
semicolon, which looks wrong when reading the code.
Pick-to: 6.7 6.6 6.5
Change-Id: Ibea7b01ac165045604b6eb7a838765b2061c368a
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
(This turns out to be identical to v44, for our purposes.)
The CLDR license has been revised at v44 to "UNICODE LICENSE V3",
which is now included (as LICENSES/UNICODE-3.0.txt) in addition to the
old license (still in use, presumably, by UCD - at least until its
next update). Some new QLocale::Language entries are needed. There is
no change to the time-zone data.
Some tests needed changes:
* Various Arabic locales now use U+0623 (Arabic letter aleph with
hamza above) in exponent separator, replacing plain U+0627 (Arabic
letter aleph); it is still followed by U+0633 (Arabic letter seen).
* Where likely sub-tags used to fill in world, 001, as territory for a
language, they now (e.g. for Prussian and Yiddish) give specific
countries.
* Tamil locales now have something of a mix of inherited and localized
forms for AM/PM, which looks a lot like a mistake in CLDR.
* New likely sub-tag rules fix ctor(und_US) and ctor(und_GB), which
previously failed.
[ChangeLog][Third-Party Code] Updated QLocale's data extracted from
the Unicode Common Locale Data Repository (CLDR) to v44.1. The license
changed to Unicode License V3.
Pick-to: 6.7 6.6 6.5
Fixes: QTBUG-121485
Task-number: QTBUG-121325
Change-Id: Ide1a68016129526d7a5aa3fc67f1a674858696bc
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
The method takes a pointer, so the code shouldn't crash when passed a
nullptr.
QFileInfoGatherer::getInfo() still needs to generate a descriptive
string for the file, so we refactor QAbstractFileIconProvider::type()
to put the implementation into a reusable static function
QAbstractFileIconProviderPrivate::getFileType(const QFileInfo &info).
This unfortunately involves constructing a QMimeDatabase on the fly,
but the docs say that is fine.
Drive-by change: use nullptr instead of `0` for pointers.
Pick-to: 6.5 6.6 6.7
Fixes: QTBUG-99178
Change-Id: Ia32ea0a26701d593e74fbecced7be8d9e0aa0f52
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
installEventFilter() prepends new objects to the eventList, so that
events that are posted while processing events are left to the next
round of event processing.
This is a baseline test to check that subsequent commits preserve the
current behavior.
QCOMPARE_GT is available since Qt6.4, so make the check backportable to
older releases too.
Pick-to: 6.7 6.6 6.5 6.2 5.15
Task-number: QTBUG-120779
Change-Id: I5ed5e9c2917a9be62de4af19c3b72889399b4fe6
Reviewed-by: Jarek Kobus <jaroslaw.kobus@qt.io>
Upgrading emscripten to 3.1.50 breaks this test, so we disable the time
for time being. After emscripten update this test is to be enabled
and fixed.
Change-Id: Ic48d81e2285ed8f7639bf20c6c29b2b9e402a591
Reviewed-by: Morten Johan Sørvig <morten.sorvig@qt.io>
As a result of using QWidget::setVisible to show the child widgets we
would end up also setting ExplicitShowHide. This is not in line with
the intent of ExplicitShowHide, which is to flag a widget as explicitly
shown/hidden by the developer, which in turn prevents Qt Widgets from
toggling WState_Hidden when the widget is reparented.
By using QWidgetPrivate::setVisible instead, we can show the child
without setting ExplicitShowHide.
As side effect of this is that we no longer reset WA_WState_Hidden
from QWidgetWindowPrivate::setVisible(). This is an issue when the
setVisible call comes as a result of destroying the QWidgetWindow,
as that is an implicit hide, and should not result in the widget
having WA_WState_Hidden. QWidget handles this case in hideChildren
by not calling QWidgetPrivate::setVisible -- instead doing its own
reset of WA_WState_Visible. We don't want to untangle this just yet,
so as a workaround we detect that the widget is already !isVisible(),
thanks to hideChildren having hidden it, and then skip the call
to QWidgetPrivate::setVisible that results from QWindow::destroy().
Task-number: QTBUG-121398
Pick-to: 6.7
Change-Id: Ib5b4d9c84f0569124c5f3ca2169cabff18934e2d
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
The nativeHiddenChild is not used for anything, and shouldn't be
needed to trigger the failure condition.
That said, I was not able to reproduce the test failure on macOS
14 with the test neither pre or post patch, nor with any of the
test cases mentioned in 51300566ff,
nor with 51300566ff itself, so this
has seemingly been fixed or worked around some other way in the
meantime.
Task-number: QTBUG-121398
Pick-to: 6.7
Change-Id: I299e7f4b71ebdb17870348a3d5b0c49a93228c8b
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
This is cleaner as the test dirs/files are created in a QTemporaryDir
that is cleaned up automatically, so less manual cleanup in the
unittest.
This also fixes issues for build environments where the source dir is
read-only.
entrylist dir in the source dir is only needed for the construction of
the QResource in the CMakeLists.txt.
Task-number: QTBUG-117449
Change-Id: I3e6389ff730c3a617854e85318f80838e012f2c8
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Don't call the native function directly, call the Java function that
calls the native function.
Pick-to: 6.7
Change-Id: Icdcf250313a38f6e4bc2b90fb7b0adbfa5a890fb
Reviewed-by: Juha Vuolle <juha.vuolle@qt.io>
Since 6.7 we consider the BUILD_SHARED_LIBS when creating libraries
using Qt CMake API. This change may affect the user projects that rely
on the old strategy of selecting the default library type. To preserve
the old behavior this change introduces the QTP0003 policy that allows
user to control whether the BUILD_SHARED_LIBS should or shouldn't be
considered in library creation process.
The policy doesn't affect Qt repos, we assume that we want the NEW
behavior by default.
Fixes: QTBUG-121707
Pick-to: 6.7
Change-Id: I4bcfbd8966839731624e3f7ef9e0d6bb2782ac50
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
Reviewed-by: Leena Miettinen <riitta-leena.miettinen@qt.io>
Amends 83e5d74864.
A model index returned by QFileSystemModel might become invalid when
events get processed, so don't store the result of setRootPath to re-use
it in a QTRY_COMPARE function. Instead, always ask for a fresh model
index.
Also, use std::chrono::duration::count correctly; it returns the "tick",
not the corresponding milliseconds, so (10s).count() returns 10 instead
of 10000. Explicitly use 10000ms here.
Un-blacklist the test on macOS again.
Task-number: QTBUG-74471
Pick-to: 6.7 6.6 6.5
Change-Id: Ic98bb53c696441131bbc1055b64822faf2aec96f
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@qt.io>
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Since QVersionNumber doesn't have an existing way to modify individual
segments, provide only const_iterator.
[ChangeLog][QtCore][QVersionNumber] Added (const) iterators over
segments (begin()/end(), incl. c- and r- variants).
Change-Id: Ia9af70c2a9c59f630123894ad2c9f38031ef5b8f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
In Qt 5, a QBitArray could not contain more than INT_MAX bits, because
the then-size_type, int, cannot represent more, even if the underlying
storage could hold 8x as much, and the serialisation format, using
unsigned int, could represent 2x.
Therefore, reject old formats with sizes that exceed INT_MAX elements
as corrupt.
Likewise, the Qt 6 serialisation format unconditionally uses 64-bit
sizes, but 32-bit platforms still cannot represent more than
numeric_limits<qsizetype>::max() (= INT_MAX) bits in memory. This is a
valid stream for 64-bit platforms, though, so ideally, this should be
using SizeLimitsExeeded, which, however, is only available from Qt
6.7. So, for now, and until we have SizeLimitsExeeded, mark the stream
as corrupt here, too.
[ChangeLog][QtCore][QBitArray] Fixed undetected overflows in the
deserialisation (opertor>>()) from QDataStream.
Pick-to: 6.7 6.6 6.5 6.2 5.15
Change-Id: Ib24cf9218c06a3a05185723c77d4313611c2dd40
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
- Match QCOMPARE logic when printing stringlists that don't match, it's
"actual" then "expected"
- Use row tag names that don't interfere with the shell; e.g. using
`tst_qdiriterator iterateResouce:':/testdata'` doesn't work for some
reason, so just add a qrc prefix
Change-Id: I3d556b83ec34f2dab15ea3178273af10fb2c62e6
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
cxx11_future flag should be disabled for VxWorks. VxWorks still does
have some bugs related to this std features and some parts of code needs
to be excluded. At least till the 24.03 VxWorks release as for now this
is the expected release that should contain fixes.
Task-number: QTBUG-115777
Change-Id: Ic652403697d727f4ae05ae7287ff8285075d3802
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This test uses socketpair which is not available on vxworks, so it fails
to compile.
Task-number: QTBUG-115777
Pick-to: 6.7
Change-Id: I46f3b4ebd937dc4ca2511381dd58451434e6a2fc
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The boundingRect and controlPointRect did not use the start point from
the `QPainterPath(const QPointF &startPoint)` constructor until the
`dirtyBounds` or `dirtyControlBounds` member variables were set to true.
Those two are false on construction. This bug was fixed by adding a new
constructor for QPainterPathPrivate that initializes the `elements`,
`bounds` and `controlBounds` member variables with the start point from
the constructor.
There is also an autotest to verify that the top left of the
boundingRect and controlPointRect are at the same position as
elementAt(0) when the start point constructor is used.
[ChangeLog][QtGui][QPainterPath] boundingRect() and controlPointRect()
now use the start point from QPainterPath(const QPointF &startPoint).
Pick-to: 6.7 6.6 6.5
Change-Id: I7bf30364406c14ed60f75d24b78a9a5535f75d93
Reviewed-by: Eirik Aavitsland <eirik.aavitsland@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Unlike other containers, a QBitArray's size() is not limited by
storage, but, esp. on 32-bit platforms, its size_type: A INT_MAX
size() QBitArray only requires 256MiB of storage.
So we can't rely on "won't happen in practice" here and need to avoid
the potential UB (signed overflow) in the (size + 7) / 8
logical-to-storage-size calculation by using unsigned arithmetic.
Use the opportunity to Extract Methods storage_size() and
allocation_size(), which differ by one (d[[0] contains the size() mod
8), making it clear what's what.
[ChangeLog][QtCore][QBitArray] Fixed a bug with QBitArrays whose
size() came within 7 of the size_type's maximum.
Pick-to: 6.7 6.6 6.5 6.2 5.15
Change-Id: I5d94bae9c9c210ba1e36f8cf03609125c81bd15d
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
I've noticed that on some Windows systems, the resolver refuses to
resolve IPv6-only addresses. But it can still do reverse resolution from
IPv6.
Change-Id: I5dd50a1a7ca5424d9e7afffd17adfe739ded932e
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Our ideal scenario was to run python to run its getnameinfo()... so just
use getnameinfo() directly. This also avoids the problem of Python not
being present and our falling back to nslookup.
Pick-to: 6.7
Change-Id: I5dd50a1a7ca5424d9e7afffd17adfc830386306f
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
These exhibit the problem described by a recent bug report.
Task-number: QTBUG-121485
Change-Id: Ia09acfa22e687ba096091a73f30df1ffd22a6e32
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
There were three things going on in this test (itself a sufficient
reason to split it up):
* Some reporting and checking of the default locale; the reporting is
duplicated in defaulted_ctor() and the check fitted more naturally
there.
* Checks that various combinations of language, script and territory
got resolved according to likely-subtag rules. These were handled
via a macro and natural candidates to become data-driven.
* A test that territory is preserved when it's the only given tag
(with a few known exceptions); broken out as a steparate test.
In the process, give the data-rows of the likely-subtag parts names
that let me extend their testing to also test construction from
string. The territory-only cases can't support that, as QLocale
doesn't support und_* forms of tags for unspecified language.
Change-Id: Id9f0fc46f30eb887b47931bad1619255acb44266
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Replace various QFETCH()/QCOMPARE() pairs with QTEST().
Just because it's terser.
Change-Id: I8496a293e3634991dcb33b8c7939f1c3028a63c5
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Instead of separate QJsonObject and QJsonArray types.
Resulted from API-review
Pick-to: 6.7
Change-Id: I977d692d3709a8c3aa872683ddda54a143c25e67
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
This naming should make it clearer that (successful) calls to
readJson(), readBody(), and readText() consume the data received
so far.
Resulted from API-review
Pick-to: 6.7
Change-Id: I09ca9eac598f8fc83eecb72c22431ac35b966bf5
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Note: documentation will be updated in a follow-up commit
This commit makes QRestReply and QRestAccessManager
classes lighter, non-owning wrappers. Furthermore their
APIs don't duplicate the wrapped QNetwork* APIs.
This makes it easier to use / opt-in to these helpers
in pre-existing applications which are based on
QNetworkAccessManager and QNetworkReply.
Since APIs are no longer duplicated, the QRest
classes are more obviously a convenience _wrapper_,
as opposed to being an alternative vertical stack.
In practice this change consists of:
- QRestAM never instantiates QNetworkAccessManager,
but accepts it via constructor. It does not take
ownership of the QNetworkAccessManager.
- QRestReply accepts QNetworkReply via constructor. It
does not take ownership of the QNetworkReply
- Signals and most duplicated functions are removed
from both QRestAM and QRR.
- QRestReply is no longer a QObject
- Since QRestAM doesn't have much to report anymore,
the debug operator is dropped.
Resulted from API-review
Pick-to: 6.7
Change-Id: Ib62d9cc2df41cac631396a84bb7ec4d2d54b0c8c
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
If the line width is negative, then we might exit the layout
loop before consuming any text, and thus the loop will never
finish. This is a side effect of a change for maximumWidth:
991c056438.
49a63d3759 fixed this issue for
QTextLayout::setFixedSize(), but I forgot to do the same in
the overload of QTextLayout::setNumColumns() which includes
an alignment width and therefore sets the line width in addition
to the column count.
Basically, we just make sure the line width is never negative so
that the width > line.width condition also means the width > 0.
Pick-to: 6.5 6.6 6.7
Fixes: QTBUG-115459
Change-Id: If904bfc64cd74e819a0864db55fa9555073d0781
Reviewed-by: Vladimir Belyavsky <belyavskyv@gmail.com>
Reviewed-by: Lars Knoll <lars@knoll.priv.no>
Add rbeing/rend overload, relevant typedefs, and decrement operators.
As a drive-by, add noexcept to begin/end functions.
Found during header review.
Pick-to: 6.7
Task-number: QTBUG-119952
Change-Id: I32d9a7d50a1f03550944c2247516c455d4822fe7
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Augment test case. Found during header review.
Pick-to: 6.7
Task-number: QTBUG-119952
Change-Id: I326395397167edb05ff1f45f7151614c02b7e7eb
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Add missing nested typedefs for both the QJniArray container and
the QJniArrayIterator. Expand test case to make sure that some standard
algorithms (such as std::distance and ranged for) work with those
types.
Found during header review.
Pick-to: 6.7
Task-number: QTBUG-119952
Change-Id: I96f348215c6f1e0e1ce777d9bdd2f172d7e52974
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Version 883ced078a83f9d79a98933145425c221a5e51f0, fetched on
2024-01-25.
[ChangeLog][Third-Party Code] Updated the public suffix list to upstream
SHA 883ced078a83f9d79a98933145425c221a5e51f0.
Pick-to: 6.7 6.6 6.5 6.2 5.15
Task-number: QTBUG-121325
Change-Id: I95e82b5c351218c1641e11e7a166ae21be2c5ad6
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Actually triggered by investigating QTBUG-121485, where the issue
turns out to be at construction time, but the complete lack of any
testing of endonyms clearly needed addressed in any case. In the
process, break up the long list of private slots in the test-class
declaration.
Task-number: QTBUG-121485
Change-Id: I49021f78d3bea2e1e55b2755a45943ab3fc23722
Reviewed-by: Øystein Heskestad <oystein.heskestad@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Every test row was following its locale-specific test with tests for
Irish and Greek day names, repeated regardless of the locale-specific
data. Express these tests as rows in their own right and shorten the
per-row testing, so we only do each once.
Change-Id: I8f919b50ac54423bacab6e5a9d34254b7db59a55
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
In the process, change it to use QCOMPARE(A, B) rather than QVERIFY(A == B).
Change-Id: I26c64100b08aef43c56f5266c0de71f5cde12816
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
... to complement QTEST_THROW_ON_FAIl/SKIP environment variables.
This allows to conveniently test both modes by running each test
twice.
Task-number: QTBUG-66320
Change-Id: I8b2810e8345061c98472d846017de910a11e0657
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Add exception classes and use them to change the control flow for
QTEST_{FAIL,SKIP}_ACTION from return'ing from just the immediate
function to the full way to the QTestLib infrastructure, here we
filter them out.
There are three modes:
- If QT_NO_EXCEPTION, then we return
- If QTEST_THROW_ON_... is also defined, #error out
- Otherwise, if QTEST_THROW_ON_... is defined, always throw
- Otherwise, the decision is made at runtime (with defaults read from
QTEST_THROW_ON_... environment variables).
Three selftests depend on the old behavior, as they explicitly check
that multiple FAIL SKIP etc emerge, which the new framework, of
course, prevents. Locally disable throwing at the test function level.
Add initial docs and enable exceptions in all of the selftest
subprograms to facilitate switching between the two runtime-selectable
modes.
[ChangeLog][QtTest] Added QTEST_THROW_ON_FAIL and QTEST_THROW_ON_SKIP
C++ macros and environment variables that, when defined, change how
QCOMPARE/QVERIFY/QSKIP etc exit the test function on failure. Instead
of a return, exiting only the immediately-surrounding function, they
throw a special exception instead, thereby exiting from subfunctions
of the test function, all the way to QtTestLib.
Fixes: QTBUG-66320
Change-Id: I96c38d2a1dcdd9de84942cf448a8bbf3ab6d3679
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Tooling can then use this information to find the correct base class,
even absent C++ scoping information.
Pick-to: 6.7 6.6 6.5
Task-number: QTBUG-101141
Change-Id: I5350da8d2d9aaf5ec86027357131ebac1eb50372
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
The definition of iterator_t, and, therefore, of is_compatible_range
depends on this, otherwise say, 0, is being treated as a valid range
and hits a hard error in adl_begin() when trying to call begin(int&).
TIL: decltype(auto) does _not_ SFINAE.
Fix by calculating the return type manually, re-enabing SFINAE.
Pick-to: 6.7 6.6
Change-Id: Icacd70554f4050ecaeb396c9ae60bc4f21a220c9
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Given a dir A that has some files and a subdir B, when a model is set to
only show files, setting the root path to A, the filters are initially
respected. Setting the root path to B then back to A, the filters would
be ignored and B would be visible in the model.
Traversing the path elements in node() led to dir B getting added to
the bypassFilters hash table, which made filtersAcceptNode() bypass the
filters.
I couldn't find a commit explaining the logic behind using bypassFilters
(the trail goes cold at the 'Qt 4.5' mega commit). The only clue I found
was the "// always accept drives" comment in the code, which hints at
this being useful only on Windows(?).
Fixes: QTBUG-74471
Pick-to: 6.7 6.6 6.5
Change-Id: Icb9055524a28990c591e924634a63e29a49835aa
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Wladimir Leuschner
Eskil Abrahamsen Blomfeldt
Shawn Rutledge
Frédéric Lefebvre
Marc Mutz
Edward Welbourne
Ahmad Samir
Piotr Wierciński
Tor Arne Vestbø
Volker Hilsheimer
Alexey Edelev
Krzysztof Sommerfeld
Łukasz Matysiak
Noah Davis
Thiago Macieira
Juha Vuolle
Mårten Nordheim