GitLab

For each progress value report StatusBar::showProgress() calls
ProgressItem::setUsesBusyIndicator(), which eventually results in a call
to StatusbarProgressWidget::connectSingleItem(). The return value of
ProgressManager::instance()->singleItem() changes very rarely between
these calls. Let us avoid reconnecting the same signal/slot pair
repeatedly.

81220db1 eliminated such misleading
progress reports for items that use a busy indicator themselves, but
missed the case when a quick percent-progress-reporting task follows
right after a busy-indicator task.

For example, GrepJob's WorkCollectFiles step uses a busy indicator, then
its next WorkGrep step reports progress in percents. So when the
WorkGrep step is completed in less than a second (e.g. when there are
not many files to search in and the looked-for string is not present in
any of them), the bottom-right-corner progress bar shows
a busy indicator for 5 seconds.

Let us display 100% progress during these 5 seconds instead.

The fix was reverted in 5.6 because it broke the ABI. The ABI break is
acceptable for the master branch => revert the reverting commit.

This reverts commit 1a61900c.

Don't use QTest::newRow(qstrdup(...)), use QTest::addRow("%s", ...)
instead.

This was reported by LSAN in multiple tests.

Fixes UBSAN warning:
```
/home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/debugger/variable/variablecollection.cpp:517:39: runtime error: downcast of address 0x6190002fe480 which does not point to an object of type 'View'
0x6190002fe480: note: object is of type 'QWidget'
 16 02 00 7e  a8 ea fb a2 36 7f 00 00  40 20 0a 00 40 61 00 00  58 ec fb a2 36 7f 00 00  00 00 be be
              ^~~~~~~~~~~~~~~~~~~~~~~
              vptr for 'QWidget'
    #0 0x7f36b128079f in KDevelop::VariableCollection::viewDestroyed(QObject*) /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/debugger/variable/variablecollection.cpp:517
```

Fixes tons of leaks reported in test_projectcontroller.

We will have to do this everywhere, to prevent the vtable
from potentially accidentally being inlined.

Use std::unique_ptr for factory functions.

Fixes UBSAN warnings in some tests which install a custom controller
before initializing the core. E.g.:

```
********* Start testing of TestProjectController *********
Config: Using QtTest library 5.15.1, Qt 5.15.1 (x86_64-little_endian-lp64 shared (dynamic) release build; by GCC 10.2.0)
/home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/shell/core.cpp:336:15: runtime error: load of value 3200171710, which is not a valid value for type 'Setup'
    #0 0x7f3724cf4d43 in KDevelop::Core::setupFlags() const /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/shell/core.cpp:336
    #1 0x7f3724c996c7 in KDevelop::PluginControllerPrivate::initKTextEditorIntegration() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/shell/plugincontroller.cpp:297
    #2 0x7f3724c4f759 in KDevelop::PluginController::PluginController(KDevelop::Core*) /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/shell/plugincontroller.cpp:357
    #3 0x55fca7edb240 in FakePluginController::PluginController(KDevelop::Core*) /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/shell/tests/test_projectcontroller.cpp:140
    #4 0x55fca7edb240 in TestProjectController::initTestCase() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/shell/tests/test_projectcontroller.cpp:165
```

Fixes various LSAN reports in ./bin/test_sessioncontroller.

Probably a real leak, but within the thirdparty library, so let's ignore this for now.

This fix broke the ABI, kdev-python no longer compiles. Therefore it
must be reverted in the stable 5.6 branch.

This reverts commit 37ff587e.

Most notably, get rid of the `*_autogen_timestamp_deps` target
and ignore generated target source files in the `CMakeFiles` folder.
Furthermore skip the `*_autogen/timestamp` files, they aren't
interesting to us either.

Juraj Oravec authored Nov 07, 2019 and Milian Wolff committed Oct 17, 2020

Signed-off-by: Juraj Oravec <sgd.orava@gmail.com>

Efficient moving of ProjectFile is important for
ProjectFileDataProvider's performance.

GIT_SILENT

The user-declared copy constructors and copy assignment operators
prevented compiler from generating noexcept move constructors and move
assignment operators for these two class templates. Eliminating the
declarations of copy operations to follow the rule of zero is impossible
or at least confusing, because of other constructors and assignment
operators in the two class templates. Let us default the move operations
instead to improve performance.

DUChainPointer(const DUChainPointer& rhs) was not defaulted, but did
exactly what the defaulted copy constructor would do as d is the only
DUChainPointer's data member. Let us make this more obvious by
defaulting this copy constructor.

The pseudo-copy constructor, the defaulted copy assignment operator and
destructor prevented compiler from generating noexcept move constructor
and move assignment operator. Eliminating them follows the rule of zero
and lets the compiler generate efficient default operations.

Path(const Path& base, const QString& subPath = QString()) acts as an
inefficient copy constructor. When only one Path argument is passed to
this constructor, the only effect is in the member initializer list:
`: m_data(other.m_data)`. This is exactly what the defaulted copy
constructor would do as m_data is the only Path's data member. But the
compiler-generated copy constructor is faster, because it is inline and
it doesn't perform various checks on subPath.

This should improve performance in many places. For example, it makes
ProjectFileDataProvider::projectOpened() and
ProjectFileDataProvider::projectClosing() somewhat faster.

GIT_SILENT

300ea014 introduced
DocumentControllerPrivate::shuttingDown variable. It is set to false in
constructor and to true in DocumentController::cleanup(). However
TestDocumentController::init() calls DocumentController::initialize()
before each test function; TestDocumentController::cleanup() calls
DocumentController::cleanup() after each test function. So the second
test function fails to open a document after shutdown.

Let us set DocumentControllerPrivate::shuttingDown to false in
DocumentController::initialize() to fix the test failure.

Use std::unique_ptr to prevent the leaks.

Fix the (broken) usage of operator++ in test_kdevhash to make it
compile. Just flip/flop between the values instead.

GIT_:SILENT

GIT_SILENT

GIT_SILENT

Do not expect QtQuick 1.0 by default, that's long deprecated.
Furthermore, improve the module code completion by listing all
library paths.

Don't rely on `next` bringing us to the expected line. Instead,
request the line explicitly and continue until we land there.
Apparrently some compiler change triggered a failure of the
test then otherwise, as we ended up in the A::A ctor in line 11
while we actually wanted to print the pointerList in loc 46.

Fixes UBSAN warning:
```
/home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/document.cpp:73:70: runtime error: downcast of address 0x6030000bcd40 which does not point to an object of type 'Document'
0x6030000bcd40: note: object is of type 'QObject'
 24 00 00 18  e0 a4 64 8d 8a 7f 00 00  a0 52 03 00 80 60 00 00  d0 5b 09 00 40 60 00 00  d0 d1 00 00
              ^~~~~~~~~~~~~~~~~~~~~~~
              vptr for 'QObject'
    #0 0x7f8a8f5e99d5 in operator() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/document.cpp:73
    #1 0x7f8a8f5e99d5 in call /usr/include/qt/QtCore/qobjectdefs_impl.h:146
    #2 0x7f8a8f5e99d5 in call<QtPrivate::List<QObject*>, void> /usr/include/qt/QtCore/qobjectdefs_impl.h:256
    #3 0x7f8a8f5e99d5 in impl /usr/include/qt/QtCore/qobjectdefs_impl.h:443
    #4 0x7f8a8d3eb035  (/usr/lib/libQt5Core.so.5+0x2eb035)
    #5 0x7f8a8d3e427f in QObject::destroyed(QObject*) (/usr/lib/libQt5Core.so.5+0x2e427f)
    #6 0x7f8a8d3e9356 in QObject::~QObject() (/usr/lib/libQt5Core.so.5+0x2e9356)
    #7 0x7f8a8f5ebf91 in Sublime::Document::~Document() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/document.cpp:76
```

Fixes UBSAN warning:
```
/home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/document.cpp:91:72: runtime error: downcast of address 0x60300038b6f0 which does not point to an object of type 'View'
0x60300038b6f0: note: object is of type 'QObject'
 22 00 80 03  e0 c4 ad 14 57 7f 00 00  a0 fa 1a 00 80 60 00 00  50 b7 38 00 30 60 00 00  00 00 00 00
              ^~~~~~~~~~~~~~~~~~~~~~~
              vptr for 'QObject'
    #0 0x7f5716a8026e in operator() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/document.cpp:91
    #1 0x7f5716a8026e in call /usr/include/qt/QtCore/qobjectdefs_impl.h:146
    #2 0x7f5716a8026e in call<QtPrivate::List<QObject*>, void> /usr/include/qt/QtCore/qobjectdefs_impl.h:256
    #3 0x7f5716a8026e in impl /usr/include/qt/QtCore/qobjectdefs_impl.h:443
    #4 0x7f571487d035  (/usr/lib/libQt5Core.so.5+0x2eb035)
    #5 0x7f571487627f in QObject::destroyed(QObject*) (/usr/lib/libQt5Core.so.5+0x2e427f)
    #6 0x7f571487b356 in QObject::~QObject() (/usr/lib/libQt5Core.so.5+0x2e9356)
    #7 0x7f5716b351b2 in Sublime::View::~View() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/view.cpp:60
```

Fixes UBSAN report:
```
/home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/area.cpp:158:66: runtime error: downcast of address 0x6030000bd7f0 which does not point to an object of type 'Area'
0x6030000bd7f0: note: object is of type 'QObject'
 2f 00 00 7b  e0 a4 1c 83 62 7f 00 00  20 59 03 00 80 60 00 00  00 df 00 00 b0 60 00 00  00 00 00 00
              ^~~~~~~~~~~~~~~~~~~~~~~
              vptr for 'QObject'
    #0 0x7f628507ca99 in operator() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/area.cpp:158
    #1 0x7f628507ca99 in call /usr/include/qt/QtCore/qobjectdefs_impl.h:146
    #2 0x7f628507ca99 in call<QtPrivate::List<QObject*>, void> /usr/include/qt/QtCore/qobjectdefs_impl.h:256
    #3 0x7f628507ca99 in impl /usr/include/qt/QtCore/qobjectdefs_impl.h:443
    #4 0x7f6282f6b035  (/usr/lib/libQt5Core.so.5+0x2eb035)
    #5 0x7f6282f6427f in QObject::destroyed(QObject*) (/usr/lib/libQt5Core.so.5+0x2e427f)
    #6 0x7f6282f69356 in QObject::~QObject() (/usr/lib/libQt5Core.so.5+0x2e9356)
    #7 0x7f6285096fcb in Sublime::Area::~Area() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/area.cpp:161
    #8 0x7f628509815e in Sublime::Area::~Area() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/area.cpp:161
    #9 0x56049f3ba6c5 in TestAreaOperation::cleanup() /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/sublime/tests/test_areaoperation.cpp:144
```

Free the raw string we get when we call QTest::toString on the
KTextEditor::Range internally.

Fixes LSAN warning:
```
Direct leak of 76 byte(s) in 4 object(s) allocated from:
    #0 0x7f1cde10b0c1 in operator new[](unsigned long) /build/gcc/src/gcc/libsanitizer/asan/asan_new_delete.cpp:102
    #1 0x7f1cc5d2d518 in qstrdup(char const*) (/usr/lib/libQt5Core.so.5+0x124518)
    #2 0x7f1cc7eace1d in char* QTest::toString<KTextEditor::Range>(KTextEditor::Range const&) /home/milian/projects/kf5/src/frameworks/ktexteditor/src/utils/range.cpp:105
    #3 0x559204f76e07 in char* QTest::toString<KDevelop::DocumentRange>(KDevelop::DocumentRange const&) /home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/language/editor/documentrange.h:66
```

While at it, use QByteArray::fromRawData too.

Fixes UBSAN warning:

```
QWARN  : TestProblems::cleanupTestCase()      1.561 ?[31mwarning: KDevelop::EmbeddedTreeRemoveItem::newItemCount[/home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/util/embeddedfreetree.h:814]?[0m: 77 VS 32
/home/milian/projects/kf5/src/extragear/kdevelop/kdevelop/kdevplatform/util/embeddedfreetree.h:816:17: runtime error: shift exponent 32 is too large for 32-bit type 'unsigned int'
```

Use QCoreApplication instead

Without this setting, one cannot use a clang-format pre-commit
hook, as it would frequently pull in unrelated changes. We should
fix the sorting once in the whole code base and then re-enable
this setting eventually.