| Age | Commit message (Collapse) | Author |
|---|
|
With Qt 6.5, QtWebEngine complains and aborts when it gets an empty argv:
https://bugreports.qt.io/browse/QTBUG-110157
https://codereview.qt-project.org/c/qt/qtwebengine/+/455347
https://codereview.qt-project.org/c/qt/qtwebengine/+/457045
https://www.riverbankcomputing.com/pipermail/pyqt/2023-March/045216.html
However, our custom qapp_args override for pytest-qt *does* return [] as
argv, causing all tests using QtWebEngine to fail.
We might decide to add sys.argv[0] to the given qapp_args in pytest-qt:
https://github.com/pytest-dev/pytest-qt/issues/483
But probably this should be fixed on the application-side, so let's do
exactly that.
See #7624
|
|
PyQt insists using the deprecated signal overload, seemingly nothing we can do
to change that:
https://www.riverbankcomputing.com/pipermail/pyqt/2023-March/045215.html
See #7624
|
|
The ELF file structure changed in a way that it seems impossible to
support reliably. Given that we have an API to get the version nowadays,
let's not bother with this. We might need to figure this out for PySide6
support, but not now.
See #7624, #3625
|
|
In doc/help/index.asciidoc, we have links like this:
* link:../quickstart{outfilesuffix}[Quick start guide]
That is correct in e.g. the GitHub file structure, as those files are
stored in e.g. doc/quickstart.asciidoc indeed.
It's *no* longer true when we view the built files via qute://help,
however: There, this turns into a flat file structure, with those pages
being at qute://help/index.html and qute://help/quickstart.html,
respectively.
It looks like QtWebEngine < 6.5 did just ignore the
<a href="../quicktart.html"> and pointed from qute://help/index.html to
qute://help/quickstart.html anyways, weirdly.
With QtWebEngine 6.5, however, this is now interpreted as a link
pointing to qute://quickstart.html instead, which is clearly wrong.
Until we have a less messy doc generation (probably as part of #345),
let's just patch the link to be correct.
See #7624
|
|
|
|
I don't quite understand why the value changed from #00000 to #121212
there with Qt 6.3, but it looks like the change is here to stay.
Instead of keeping #000000 the default and adding an override for every
new Qt release, let's just switch over and add an override for the old
versions (5.15 and 6.2), so this won't break again for every new release.
See #7624
|
|
See #7624
|
|
The fontMetrics().elidedText(...) call is copied from the base QTabBar
implementation of initStyleOption(). But here we are calling it with a
text_rect from our custom style class. In Qt6 QStyleSheetStyle no longer
calls into our custom Style class for the SE_TabBarTabText SubElement so
it ends up getting a text width from QCommonStyle which is wrong for
vertical tabs (`QCommonStylePrivate::tabLayout()` seems to transpose the
rect, maybe it thinks vertical tabs should have the text running
vertically too).
Also resolves the issue that when tabs.title.alignment is set to center
the leftmost character or two of titles wouldn't be visible when tabs
where small enough that the text got elided. Not 100% sure why but
presumably it wasn't picking up the padding etc logic in our custom
style class.
In the test I don't test the exact elided text when we don't have much
room because it might differ based on what monospace font is picked up
in the tests. So the amount elided might change. Also in the not-elided
one I'm only testing endswith() because the index should be at the start
of the string which I don't really care about.
Fixes: #7313
Fixes: #7205
|
|
Needed so that tests generating :help docs run properly.
Follow-up to af176f749fd7f5521d3978e268a1fad169ac2bd4 and
bbdc83f638ed1a8e263a3efac0da8a901c1a1420
(cherry picked from commit 3d56f57e502ffaad40936f19f44083a2b63f0690)
|
|
A quick tour through the PDF.js API:
v1.0.1040 introduced the ability to pass an object to PDFView.open()
instead of passing an URL:
https://github.com/mozilla/pdf.js/pull/5254
v1.6.210 (?) renamed PDFView to PDFViewerApplication:
ffa276a1826a95bf1f80b15f48cfbf2b9cfe6dad
v3.3.122 made it mandatory to pass an object (and also made originalURL
optional when doing so):
https://github.com/mozilla/pdf.js/pull/15972
We should probably properly get rid of old version support at some point
(see #7619), but until then, I *think* this approach should work with
older versions still.
Fixes #7618
(cherry picked from commit 924a7a5907124a0a33e2cc9454e389b7a2634244)
|
|
I'm not 100% sure we need this. But there is an existing test matrix
that I want to active for now. We can review all the configuration later
I guess?
|
|
|
|
|
|
|
|
|
|
|
|
v0.9.0 seems to add SC2317 ("Command appears to be unreachable."):
https://www.shellcheck.net/wiki/SC2317
This adds various false-positives in password_fill, let's just ignore
those...
Also see: https://github.com/koalaman/shellcheck/issues/2613
|
|
|
|
|
|
Update dependencies
|
|
qt6: Delay showing windows
|
|
|
|
|
|
|
|
|
|
Fixes #7352, closes #6793
|
|
|
|
See https://tox.wiki/en/4.0.3/faq.html#tox-4-changed-ini-rules
|
|
|
|
Closes #812
This does *not* seem to help with the crashes in #5385 or #5124, weirdly
enough, but it still seems simpler, so why not do it.
|
|
If we shut down qutebrowser directly, in some rare situations, we seem
to get into some trouble related to a QWidget trying to call into a
Q(Common|StyleSheet)Style at shutdown (checking the style hint
SH_Widget_ShareActivation, to figure out the palette to use when the
window is hidden, or somesuch?).
With the recent changes for #7504, those crashes suddenly started
happening much more often. After staring at the related changes and gdb
for over a day, I still have no idea what actually causes them and why
it's crashing - it smells like some kind of PyQt/Qt cleanup order bug...
However, what I *did* find out (after a lot of painful debugging and
messing around with things) is that delaying the shutdown done from
:quit seems to reliably get rid of the issue. I don't know *why*, but
I haven't seen it happen anymore with the latest changes...
Fixes #5385 and #5124, both speculatively.
Closes #7504 since this was the last change required to make it all
work.
|
|
None is the default value for instance.shutdown()'s "session", so we can
simplify this a bit.
|
|
With pytest main, pytest now reuses temporary directories...
|
|
In bleeding tests, we started to get a segfault on the second test in
tests/unit/mainwindow/test_prompt.py, with a stacktrace like:
Thread 1 "python" received signal SIGSEGV, Segmentation fault.
0x00007ffff7b55912 in _PyFunction_Vectorcall (func=<function at remote 0x7fffc4368ca0>, stack=0x7fffd74656a8, nargsf=<optimized out>, kwnames=0x0) at Objects/call.c:341
341 if (((PyCodeObject *)f->fc_code)->co_flags & CO_OPTIMIZED) {
(gdb) bt
#0 0x00007ffff7b55912 in _PyFunction_Vectorcall (func=<function at remote 0x7fffc4368ca0>, stack=0x7fffd74656a8, nargsf=<optimized out>, kwnames=0x0) at Objects/call.c:341
#1 0x00007ffff4e0b3f1 in PyQtSlot::call(_object*, _object*) const (this=0x555556c759c0, args=('/tmp/pytest-of-florian/pytest-70/test_simple_completion_1_next_0/test',), callable=<function at remote 0x7fffc4368ca0>)
at ../../qpy/QtCore/qpycore_pyqtslot.cpp:247
#2 PyQtSlot::invoke(void**, _object*, void*, bool) const (this=0x555556c759c0, qargs=<optimized out>, qargs@entry=0x7fffffff86c0, self=<optimized out>, self@entry=0x0, result=result@entry=0x0, no_receiver_check=<optimized out>)
at ../../qpy/QtCore/qpycore_pyqtslot.cpp:159
#3 0x00007ffff4e12213 in PyQtSlot::invoke(void**, bool) const (no_receiver_check=<optimized out>, qargs=0x7fffffff86c0, this=<optimized out>) at ../../qpy/QtCore/qpycore_pyqtslot.cpp:78
#4 PyQtSlotProxy::unislot(void**) (qargs=0x7fffffff86c0, this=0x555557193e20) at ../../qpy/QtCore/qpycore_pyqtslotproxy.cpp:205
#5 PyQtSlotProxy::unislot(void**) (qargs=0x7fffffff86c0, this=0x555557193e20) at ../../qpy/QtCore/qpycore_pyqtslotproxy.cpp:186
#6 PyQtSlotProxy::qt_metacall(QMetaObject::Call, int, void**) (this=0x555557193e20, _c=<optimized out>, _id=0, _a=0x7fffffff86c0) at ../../qpy/QtCore/qpycore_pyqtslotproxy.cpp:170
#7 0x00007ffff48bd91d in doActivate<false>(QObject*, int, void**) (sender=0x555556b3d680, signal_index=28, argv=0x7fffffff86c0) at kernel/qobject.cpp:3945
#8 0x00007fffeff96aca in QFileSystemModel::directoryLoaded(QString const&) (this=<optimized out>, _t1=<optimized out>) at .moc/moc_qfilesystemmodel.cpp:272
#9 0x00007ffff48b0be0 in QObject::event(QEvent*) (this=this@entry=0x555556b3d680, e=e@entry=0x7fff0c004af0) at kernel/qobject.cpp:1347
#10 0x00007fffeff962ab in QFileSystemModel::event(QEvent*) (this=this@entry=0x555556b3d680, event=event@entry=0x7fff0c004af0) at dialogs/qfilesystemmodel.cpp:1748
#11 0x00007ffff070995c in sipQFileSystemModel::event(QEvent*) (this=0x555556b3d680, a0=0x7fff0c004af0) at /usr/src/debug/pyqt5/PyQt5-5.15.7/build/QtWidgets/sipQtWidgetsQFileSystemModel.cpp:376
[...]
#23 0x00007ffff4da94ce in meth_QCoreApplication_processEvents(PyObject*, PyObject*, PyObject*) (sipArgs=<optimized out>, sipKwds=0x0) at /usr/src/debug/pyqt5/PyQt5-5.15.7/build/QtCore/sipQtCoreQCoreApplication.cpp:590
[...]
from pytest-qt in Python:
Current thread 0x00007fb366207740 (most recent call first):
File ".../pytestqt/plugin.py", line 209 in _process_events
File ".../pytestqt/plugin.py", line 171 in pytest_runtest_setup
[...]
File ".../pytest/src/_pytest/runner.py", line 115 in pytest_runtest_protocol
[...]
File ".../pytest/src/_pytest/main.py", line 317 in pytest_cmdline_main
[...]
Introduced by this change in pytest, which now deletes the temporary
directory after the test by default:
https://github.com/pytest-dev/pytest/pull/10517
It sounds like something odd like the directory removal causing (Py)Qt
to call the lambda, but the underlying Python object already being gone
or something along those lines. With a proper Qt slot (@pyqtSlot),
PyQt seems to do the right thing, so let's just use that instead...
|
|
See https://archlinux.org/packages/extra/x86_64/mesa/download/
|
|
With Qt 6.4, QtWebEngine closes/reopens the main window to switch the
RHI rendering mode when a QWebEngineView gets added:
https://github.com/qt/qtbase/blob/v6.4.1/src/widgets/kernel/qwidget.cpp#L10706-L10709
To avoid this, we need to make sure we only call .show() *after* adding
a tab, similarly to what Qt did too:
https://code.qt.io/cgit/qt/qtwebengine.git/commit/?id=d7e0fd5304ebdb12c6f809cdbcf8193b49b9ecd2
See #7504
----
This commit handles changes around app.py/mainwindow.py, which was
probably the most complex one of the whole set (and also involving the
oldest code I bet...).
This changes a couple of intertwined things:
- app._process_args() now needs to take track of the window it opened
(if any), so that it can call .show() on it *after* opening pages.
* If a session was loaded instead, sessions.py takes care of showing
it.
* The setActiveWindow call was also moved for simplicitly. Before, it
was called even with --nowindow given, but that doesn't really make
much sense.
* I'm not actually sure why the .setActiveWindow() call is there. Qt
docs say "Warning: This function does not set the keyboard focus to
the active widget. Call QWidget::activateWindow() instead.".
It was added back in 2014: ae44aa01a6d7d6f2e2b2e97a48f49a579f342c9f
("Set initial focused window correctly."). It's possible it's not
needed anymore, but that's for a separate commit.
- app.process_pos_args() now gets a MainWindow (rather than win_id)
from mainwindow.get_window(), and is responsible for calling .show()
and .maybe_raise() on it when nothing else does (like open_url called
from it).
* To preserve existing behavior (and not fail tests), it also still
calls .maybe_raise() when a command was given. Maybe we should get
rid of this to fix #5094, but that's for a separate commit.
* Note it does *not* call .show(). That's taken care of later in
_process_args() already.
- app.open_url() can directly show the window, because it already opens
a URL. It now still returns the MainWindow object for consistency.
Also gets rid of some questionable objreg usage just to access a
tabbed browser, as a tiny step towards #640.
- Similarly, app._open_startpage() could have stayed, but now also
takes a MainWindow and uses objreg a bit less.
- open_desktopservices_url also takes care of the show/raise (as this
gets called isolated from Qt), and also avoids objreg stuff where an
attribute access would have sufficed...
- mainwindow.get_window() now doesn't show the window anymore, and
returns the mainwindow object instead of a window ID.
* However, this means it can't actually raise the window anymore
(it needs to be shown first).
- To keep the decision about whether to raise the window in a central
place, MainWindow now has a new should_raise attribute and
maybe_raise() method. mainwindow.get_window() sets should_raise
accordingly, and the caller is responsible to call .maybe_raise()
after showing.
|
|
I got an exception there while running tests with the refactored code.
I don't quite understand how/why it changed, but it seems sensible to
skip this too if we skipped calling calling .activateWindow() due to
it being deleted.
See 2fd21bb16bdc1f0c3f719c572271cc730c02c7a6
|
|
With Qt 6.4, QtWebEngine closes/reopens the main window to switch the
RHI rendering mode when a QWebEngineView gets added:
https://github.com/qt/qtbase/blob/v6.4.1/src/widgets/kernel/qwidget.cpp#L10706-L10709
To avoid this, we need to make sure we only call .show() *after* adding
a tab, similarly to what Qt did too:
https://code.qt.io/cgit/qt/qtwebengine.git/commit/?id=d7e0fd5304ebdb12c6f809cdbcf8193b49b9ecd2
See #7504
----
This commit handles browser.commands, in particular the
"_new_tabbed_browser" convenience method (which is now, sadly, likely
less convenient).
|
|
With Qt 6.4, QtWebEngine closes/reopens the main window to switch the
RHI rendering mode when a QWebEngineView gets added:
https://github.com/qt/qtbase/blob/v6.4.1/src/widgets/kernel/qwidget.cpp#L10706-L10709
To avoid this, we need to make sure we only call .show() *after* adding
a tab, similarly to what Qt did too:
https://code.qt.io/cgit/qt/qtwebengine.git/commit/?id=d7e0fd5304ebdb12c6f809cdbcf8193b49b9ecd2
See #7504
----
This commit handles various simple and isolated cases. Things requiring
a bit more refactoring are handled in a separate commit.
|
|
|
|
Update dependencies
|
|
|
|
|
|
|
|
Still some stuff broken sadly, see #7314
|
|
(cherry picked from commit 39e577c36eef9892165a6979c403e9b2d9bc239c)
|
|
|
|
|
|
We don't support Qt 5.12 anymore, so those can go.
|
|
Trivial to polyfill, needed by various pages in the wild,
and only natively available with Qt 6.2.
Closes #7501
Also see #7237 and 726d5e614b1f80c339084eba7d1c240bf467fa9b
(cherry picked from commit ce64e5d899666bd469bf07bd0fecbc75d4bfed3a)
|
