Use P0608's trick to detect convertibility without narrowing;
and now that we can depend on C++17, use its features.
First, this moves the burden of detecting a narrowing conversion on
the compiler, rather than us maintaining a complicated series
of checks. Of course, this exposes
* bugs in compilers (e.g. GCC < 9 thinks that float->bool is not
narrowing;
* behavior still not (widely) implemented (pointer to bool
conversions are narrowing, P1957);
* interesting compiler choices, e.g. GCC 9 thinks that unscoped
enumerations are non-narrowing convertible to a datatype big
enum to contain all the _enumerators_, even if the underlying
type of the enum (and/or its sizeof()) is wider than the target
datatype.
Second, it allows to detect conversions that have a narrowing
conversion as an intermediate step. Given a type like
struct Bad { operator double() const; };
then an object of type Bad is implictly convertible to a type
like int via a narrowing conversion. Therefore, a connection
is possible between a signal carrying a Bad and a slot accepting
an int. We can now detect and block this.
Tests regarding scoped enumerations have been dropped,
for the simple reason that a scoped enumeration is not
implictly convertible to an integral type, so we don't have
that detection (it would constantly fail). Scoped enumerations
do not take part in narrowing conversions anyhow, cf. [dcl.init.list].
[ChangeLog][QtCore][QObject] The detection of narrowing conversions
when calling QObject::connect() when
QT_NO_NARROWING_CONVERSIONS_IN_CONNECT now takes also
into account user-defined implicit conversions that undergo
through a narrowing conversion.
Change-Id: Ie09d59203fe6283378b36dfbc54de1d58098ef51
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Map setFilterWildcard() and setFilterFixedString() to now use
QRegularExpression.
Change-Id: I2dff2015234decb2badfd306975dcff8553cdd7f
Reviewed-by: Alex Blasche <alexander.blasche@qt.io>
The old implementation was either using CRC32 on modern processors
or a trivial, but rather slow implementation.
We can't continue with CRC32, as that implementation can only
give us 32bit hashes, where we now need to support 64bit in Qt 6.
Change the implementation to use MurmurHash, as public domain
implementation that is both very fast and leads to well distributed hashes.
This hash function is about as fast as the SSE optimized CRC32 implementation
but works everywhere and gives us 64 bit hash values.
Here are some numbers (time for 10M hashes):
14 char 16 char
QByteArray QString float
old qHash (non CRC32) 127ms 134ms 48ms
old qHash (using SSE CRC32 instructions 60ms 62ms 46ms
new qHash 52ms 43ms 46ms
Unfortunately MurmurHash is not safe against hash table DoS attacks, as
potential hash collisions are indepenent of the seed. This will get
addressed in followup commit, where we use SipHash or an SSE optimized
AES based hashing algorithm that does not have those issues.
Change-Id: I4fbc0ac299215b6db78c7a0a2a1d7689b0ea848b
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
At the same time use the opportunity to refactor the
insertion code inside the implementation of QHash to
avoid copy and move constructors as much as possible
and always construct nodes in place.
Change-Id: I951b4cf2c77a17f7db825c6a776aae38c2662d23
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
This is required, so that QHash and QSet can hold more
than 2^32 items on 64 bit platforms.
The actual hashing functions for strings are still 32bit, this will
be changed in a follow-up commit.
Change-Id: I4372125252486075ff3a0b45ecfa818359fe103b
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Make use of the new features available in QHash and do a more
performant implementation than the old one.
Change-Id: Ie74b3cdcc9871cd241aca205672093dc395d04a7
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
A brand new QHash implementation using a faster and more memory efficient data
structure than the old QHash.
A new implementation for QHash. Instead of a node based approach as the old
QHash, this implementation now uses a two stage lookup table. The total
amount of buckets in the table are divided into spans of 128 entries.
Inside each span, we use an array of chars to index into a storage area
for the span.
The storage area for each span is a simple array, that gets (re-)allocated
with size increments of 16 items. This gives an average memory overhead of
8*sizeof(struct{ Key; Value; }) + 128*sizeof(char) + 16 for each span.
To give good performance and avoid too many collisions, the array keeps its
load factor between .25 and .5 (and grows and rehashes if the load factor goes
above .5).
This design allows us to keep the memory overhead of the Hash very small, while
at the same time giving very good performance. The calculated overhead for a
QHash<int, int> comes to 1.7-3.3 bytes per entry and to 2.2-4.3 bytes for
a QHash<ptr, ptr>.
The new implementation also completely splits the QHash and QMultiHash classes.
One behavioral change to note is that the new QHash implementation will not
provide stable references to nodes in the hash when the table needs to grow.
Benchmarking using https://github.com/Tessil/hash-table-shootout shows
very nice performance compared to many different hash table implementation.
Numbers shown below are for a hash<int64, int64> with 1 million entries. These
numbers scale nicely (mostly in a linear fashion with some variation due to
varying load factors) to smaller and larger tables. All numbers are in seconds,
measured with gcc on Linux:
Hash table random random random random reads full
insertion insertion full full after iteration
(reserved) deletes reads deletes
------------------------------------------------------------------------------
std::unordered_map 0,3842 0,1969 0,4511 0,1300 0,1169 0,0708
google::dense_hash_map 0,1091 0,0846 0,0550 0,0452 0,0754 0,0160
google::sparse_hash_map 0,2888 0,1582 0,0948 0,1020 0,1348 0,0112
tsl::sparse_map 0,1487 0,1013 0,0735 0,0448 0,0505 0,0042
old QHash 0,2886 0,1798 0,5065 0,0840 0,0717 0,1387
new QHash 0,0940 0,0714 0,1494 0,0579 0,0449 0,0146
Numbers for hash<std::string, int64>, with the string having 15 characters:
Hash table random random random random reads
insertion insertion full full after
(reserved) deletes reads deletes
--------------------------------------------------------------------
std::unordered_map 0,4993 0,2563 0,5515 0,2950 0,2153
google::dense_hash_map 0,2691 0,1870 0,1547 0,1125 0,1622
google::sparse_hash_map 0,6979 0,3304 0,1884 0,1822 0,2122
tsl::sparse_map 0,4066 0,2586 0,1929 0,1146 0,1095
old QHash 0,3236 0,2064 0,5986 0,2115 0,1666
new QHash 0,2119 0,1652 0,2390 0,1378 0,0965
Memory usage numbers (in MB for a table with 1M entries) also look very nice:
Hash table Key int64 std::string (15 chars)
Value int64 int64
---------------------------------------------------------
std::unordered_map 44.63 75.35
google::dense_hash_map 32.32 80,60
google::sparse_hash_map 18.08 44.21
tsl::sparse_map 20.44 45,93
old QHash 53.95 69,16
new QHash 23.23 51,32
Fixes: QTBUG-80311
Change-Id: I5679734144bc9bca2102acbe725fcc2fa89f0dff
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Clear Linux containers running as root may have no /etc/passwd. But
they'll have /etc/machine-id because systemd creates that. Also test
/proc/version (a Linux-specific file) because that isn't writeable even
by root.
Take the opportunity to check with access() instead of assuming root and
only root can write to the file.
Change-Id: Ibdc95e9af7bd456a94ecfffd1603e8359604752b
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
This was never tested. The infinite loop in QCborContainerPrivate::grow
is the proof.
[ChangeLog][QtCore][QCborArray] Fixed an infinite loop when operator[]
was called with with an index larger than the array's size plus 1.
Change-Id: Ibdc95e9af7bd456a94ecfffd1603df3855c73f20
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Similar to the QJsonObject issue of the previous commit (found with the
same tests, but not the same root cause). One fix was that copying of
byte data from the QByteArray to itself won't work if the array
reallocates. The second was that
assign(*that, other.concrete());
fails to set other.d to null after moving. By calling the operator=, we
get the proper sequence of events.
[ChangeLog][QtCore][QCborMap] Fixed some issues relating to assigning
elements from a map to itself.
Note: QCborMap is not affected by the design flaw discovered in
QJsonObject because it always appends elements (it's unsorted), so
existing QCborValueRef references still refer to the same value.
Task-number: QTBUG-83366
Change-Id: Ibdc95e9af7bd456a94ecfffd1603df846f46094d
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
The refactoring to use CBOR missed two places where we could assign from
the same object and thus cause corruption. In fixing this issue, I found
a design flaw in QJsonObject, see Q_EXPECT_FAILing unit test and task
QTBUG-83398.
[ChangeLog][QtCore][QJsonObject] Fixed a regression from 5.13 that
incorrect results when assigning elements from an object to itself.
Fixes: QTBUG-83366
Change-Id: Ibdc95e9af7bd456a94ecfffd1603df24b06713aa
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Normally people shouldn't create temporary files on /, but if you're
running as root, why not?
Caught when running tst_qtemporaryfile as root:
openat(AT_FDCWD, "", O_RDWR|O_CLOEXEC|O_TMPFILE, 0600) = -1 ENOENT (No such file or directory)
Change-Id: Ibdc95e9af7bd456a94ecfffd1603ebfc17cea220
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
Reviewed-by: David Faure <david.faure@kdab.com>
Conflicts:
examples/opengl/doc/src/cube.qdoc
src/corelib/global/qlibraryinfo.cpp
src/corelib/text/qbytearray_p.h
src/corelib/text/qlocale_data_p.h
src/corelib/time/qhijricalendar_data_p.h
src/corelib/time/qjalalicalendar_data_p.h
src/corelib/time/qromancalendar_data_p.h
src/network/ssl/qsslcertificate.h
src/widgets/doc/src/graphicsview.qdoc
src/widgets/widgets/qcombobox.cpp
src/widgets/widgets/qcombobox.h
tests/auto/corelib/tools/qscopeguard/tst_qscopeguard.cpp
tests/auto/widgets/widgets/qcombobox/tst_qcombobox.cpp
tests/benchmarks/corelib/io/qdiriterator/qdiriterator.pro
tests/manual/diaglib/debugproxystyle.cpp
tests/manual/diaglib/qwidgetdump.cpp
tests/manual/diaglib/qwindowdump.cpp
tests/manual/diaglib/textdump.cpp
util/locale_database/cldr2qlocalexml.py
util/locale_database/qlocalexml.py
util/locale_database/qlocalexml2cpp.py
Resolution of util/locale_database/ are based on:
https://codereview.qt-project.org/c/qt/qtbase/+/294250
and src/corelib/{text,time}/*_data_p.h were then regenerated by
running those scripts.
Updated CMakeLists.txt in each of
tests/auto/corelib/serialization/qcborstreamreader/
tests/auto/corelib/serialization/qcborvalue/
tests/auto/gui/kernel/
and generated new ones in each of
tests/auto/gui/kernel/qaddpostroutine/
tests/auto/gui/kernel/qhighdpiscaling/
tests/libfuzzer/corelib/text/qregularexpression/optimize/
tests/libfuzzer/gui/painting/qcolorspace/fromiccprofile/
tests/libfuzzer/gui/text/qtextdocument/sethtml/
tests/libfuzzer/gui/text/qtextdocument/setmarkdown/
tests/libfuzzer/gui/text/qtextlayout/beginlayout/
by running util/cmake/pro2cmake.py on their changed .pro files.
Changed target name in
tests/auto/gui/kernel/qaction/qaction.pro
tests/auto/gui/kernel/qaction/qactiongroup.pro
tests/auto/gui/kernel/qshortcut/qshortcut.pro
to ensure unique target names for CMake
Changed tst_QComboBox::currentIndex to not test the
currentIndexChanged(QString), as that one does not exist in Qt 6
anymore.
Change-Id: I9a85705484855ae1dc874a81f49d27a50b0dcff7
When a class has multiple QProperty members to implement functionality,
it is common to have functions in the class that react to changes. For
example to emit a compatibility signal, in case of Qt Quick to mark the
scene graph as dirty, etc. etc.
To faciliate this use-case, this patch adds an internal
QPropertyMemberChangeHandler template that allows connecting a QProperty
field to a member function callback.
At the moment that callback is still 3 * sizeof(pointer). This could in
theory be reduced to 2 by eliminating the back-pointer (prev) as the
observer lives as long as the property. That however belongs into maybe
a future patch.
In order to get a pointer back to the surrounding object that holds the
QProperty as well as provides the callback function, the property system
was changed to pass through the address of the QProperty member at
run-time, and at compile time the delta from the QProperty member to the
beginning of the surrounding class is calculated. Through subtraction we
obtain the pointer to the owning object.
Change-Id: Ia2976357053f474ff44d0d6f60527c3b8e1f613a
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
std::function does not have deduction guides in older libc++ (presumably older
than version 10). Omitting the template parameter isn't essential for the test,
so just give it.
Change-Id: Ia9bb91f961b0928203737ec976913effd06433e0
Reviewed-by: Jüri Valdmann <juri.valdmann@qt.io>
The name of the option may cause confusion due to the fact
that it's not _fully_ anchoring the match, only anchoring it
at the offset passed to match() -- in other words, it's a
"left" anchoring. Deprecate the old name and introduce
a new one that should explain the situation better.
[ChangeLog][QtCore][QRegularExpression] The AnchoredMatchOption
match option has been deprecated in favor of
AnchorAtOffsetMatchOption, which should better describe
that the match is only anchored at the offset.
Change-Id: Ib751e5e488f2d0309a2da6496378247dfa4648de
Reviewed-by: Samuel Gaist <samuel.gaist@idiap.ch>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
In particular, this changed the US currency formats for negative
amounts to be parenthesised versions of the positive amount forms,
rather than having a minus sign after the $ sign. Test updated.
[ChangeLog][QtCore][QLocale] Currency formats are now based on CLDR's
accounting formats, where they were previously mostly based (more or
less by accident) on standard formats. In particular, this now means
negative currency formats are specified, where available, where they
(mostly) were not previously.
Task-number: QTBUG-79902
Change-Id: Ie0c07515ece8bd518a74a6956bf97ca85e9894eb
Reviewed-by: Cristian Maureira-Fredes <cristian.maureira-fredes@qt.io>
Added QFuture::onFailed() method, which allows attaching handlers for
exceptions that may occur in QFuture continuation chains.
Task-number: QTBUG-81588
Change-Id: Iadeee99e3a7573207f6ca9f650ff9f7b6faa2cf7
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Replaced the internal ExceptionHolder for storing QException* by
std::exception_ptr. This will allow to report and store exceptions
of types that are not derived from QException.
Task-number: QTBUG-81588
Change-Id: I96be919d8289448b3e608310e51a16cebc586301
Reviewed-by: Vitaly Fanaskov <vitaly.fanaskov@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
We should not implicitly convert a QString to a QLocale object. It can
easily create unwanted side effects.
Change-Id: I7bd9b4a4e4512c0e60176ee4d241d172f00fdc32
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
QFuture's original design pre-dates C++11 and its
introduction of move semantics. QFuture is documented
as requiring copy-constructible classes and uses copy
operations for results (which in Qt's universe in general
is relatively cheap, due to the use of COW/data sharing).
QFuture::result(), QFuture::results(), QFuture::resultAt()
return copies. Now that the year is 2020, it makes some
sense to add support for move semantics and, in particular,
move-only types, like std::unique_ptr (that cannot be
obtained from QFuture using result etc.). Taking a result
or results from a QFuture renders it invalid. This patch
adds QFuture<T>::takeResults(), takeResult() and isValid().
'Taking' functions are 'enabled_if' for non-void types only
to improve the compiler's diagnostic (which would otherwise
spit some semi-articulate diagnostic).
As a bonus a bug was found in the pre-existing code (after
initially copy and pasted into the new function) - the one
where we incorrectly report ready results in (rather obscure)
filter mode.
Fixes: QTBUG-81941
Fixes: QTBUG-83182
Change-Id: I8ccdfc50aa310a3a79eef2cdc55f5ea210f889c3
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
QByteArray doesn't like it.
Apply the same protection to QString, which we know uses the same
backend but uses elements twice as big. That means it can contain
slightly more than half as many elements, but exact half will suffice
for our needs.
Change-Id: Iaa63461109844e978376fffd15f9d4c7a9137856
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
QtQml needs the private just for one detail which nobody else should
need it for: Tracking additional dependencies and marking the binding as
dirty. Exporting the private requires hiding some variables and
providing accessors, to compile with MSVC - including the removal of
QVarLengthArray usage. Upside: The binding structure shrinks by 8 bytes
and the encapsulation makes it a little easier to change things without
breaking declarative, ... in the unlikely event ;-)
Also remove setDirty() from the public API as it's not needed by QtQml
and using it is dangerous, because it means that there's a risk of
somebody keeping a reference (count) to the untyped binding from within
the binding closure, which introduces a memory leak.
Change-Id: I43bd56f4bdf218efb54fa23e2d627ad3acfafeb5
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Speed up the QSortFilterProxyModel filtering by only updating the
source_to_proxy entries which are really changed - When proxy intervals
are added or removed, it is not needed to update the proxy_to_source
indexes which were not touched.
Change-Id: I35459ff1b04f4610ec74f4b01d58a71832a9ae22
Reviewed-by: David Faure <david.faure@kdab.com>
Increasing the sample size of randomly generated test samples reduces
the probability of small deviations from the expected uniform
distribution.
On my machine with the new values the test fails approximately once per
3000 consecutive runs, instead of failing once per 300.
Change-Id: I4d1815504c353290a2fb350b3fd1cbb802f8d559
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
A simple 16k file can produce deep enough recursion in Qt to cause stack
overflow. So prevent that.
I tested 4096 recursions just fine on my Linux system (8 MB stack), but
decided 1024 was sufficient, as this code will also be run on embedded
systems that could have smaller stacks.
[ChangeLog][QtCore][QCborValue] fromCbor() now limits decoding to at
most 1024 nested maps, arrays, and tags to prevent stack overflows. This
should be sufficient for most uses of CBOR. An API to limit further or
to relax the limit will be provided in 5.15. Meanwhile, if decoding more
is required, QCborStreamReader can be used (note that each level of map
and array allocates memory).
Change-Id: Iaa63461109844e978376fffd15fa0fbefbf607a2
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
If we detected that the OS supports a version of system forkfd (Linux
pidfd, FreeBSD procdesc), the forkfd_wait() function was using only the
system waiting implementation, which of course can't work for file
descriptors created with FFD_USE_FORK. So just detect EBADF and attempt
again.
If the file descriptor is neither one of our pipes nor a system forkfd,
bad things will happen...
Fixes: QTBUG-82351
Change-Id: I4e559af2a9a1455ab770fffd15f59fb3160b22eb
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Add some overloads where (I thought) it makes sense for QDir and QFile
to accept std::filesystem::path objects. Currently my thinking is to
not add overloads for static functions where std::filesystem can already
do the same job, e.g. create directory or file.
Template and enable_if is needed due to both QString and
std::filesystem::path being able to be constructed from string literals.
The common shared code is currently in QFile because QDir had an
implicit include of QFile, made explicit in this patch, and QFileInfo
has an include to QFile as well.
The QT_HAS_STD_FILESYSTEM macro is visible in user-code which I
currently take advantage of in the tests, and users could too.
Change-Id: I8d05d3c34c6c17e20972a6a2053862b8891d6c3c
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Test fails on MinGW 8.1 x86, but not on MinGW 8.1 x86_64.
Task-number: QTQAINFRA-3304
Task-number: QTBUG-69947
Change-Id: Ie9a35bd6d5a8481028cd0ea426d1cf00bd7cf093
Reviewed-by: Tony Sarajärvi <tony.sarajarvi@qt.io>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
A generic binding allows implementing the binding function in a way that
enables the QML engine to run binding scripts and convert the V4::Value
into a QVariant and then assign the value to the property with the help
of QMetaType::construct.
Change-Id: Id4807be92eee7e3501908e6c5e4c861cfcb7772a
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
If we want to filter away a column without changing the filtering
for the rows calling invalidateFilter() is wasteful because it will
call filterAcceptsRow() for all rows even though that is not needed.
This commit add two functions, invalidateRowsFilter() and
invalidateColumnsFilter() that work the same way as invalidateFilter()
except that they will invoke respectively only filterAcceptsRow() and
filterAcceptsColumn().
Change-Id: Ib4351cf08c229bd97bbbfee6da92397dca579a84
Reviewed-by: David Faure <david.faure@kdab.com>
QSet and std::(unordered_)set were so far not treated as appendable, as
they lack a push_back method. We do however need support for this in
declarative to enable converting back from QJSValue arrays to sets.
We achieve this by testing for and using the insert method. While vector
has also such a method, it doesn't take a single value, but rather a
position or iterator + value, so the template specialization is not
ambiguous.
Task-number: QTBUG-82743
Change-Id: I74fc7b1b856d9bcd38100b274ba2b69578ea8bbb
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
A common pattern in declarative is to use the unused bits in linked list
next pointers for additional information storage. The "next" pointer is
typically then a tagged pointer of the containing class, which is not
fully defined yet. Therefore alignof() can't be used at tagged pointer
instantiation time. This patch delays the calls to alignment, etc. until
the corresponding functions are used, as in principle the tagged pointer
is just a quintptr and no additional information should be needed until
operating on it.
Change-Id: I87a3578ee921d471e1b60ed5903b549ef0610b97
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
The call of _control87 would crash because of the previous test.
Change-Id: I254efe9c2e9892a473a02663e5ff7016791d5d6d
Reviewed-by: Tony Sarajärvi <tony.sarajarvi@qt.io>
Useful for attaching small bits of information in the alignment bits of
a naked pointer. For use in the new property system as well as in
qtdeclarative (where currently a similar class exists as private API).
Change-Id: Idf9b93e714e15129f302e16425dbeda94bcd207b
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
