Helmut Grohne [Thu, 28 May 2026 21:30:55 +0000 (21:30 +0000)]
call pkgconfig in order to be able to cross build qtbase with MySql.
Bug-Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=971604
Forwarded: not-needed Reviewed-by: Lisandro Damián Nicanor Pérez Meyer <lisandro@debian.org>
Qt's build system calls mysql_config... which won't work in a cross build
environment like Debian's, as it will throw an exec format error.
In order to solve this call pkgconfig and use mysqlclient.pc.
Pino Toscano [Thu, 28 May 2026 21:30:55 +0000 (21:30 +0000)]
Limit Linux-only code with Q_OS_LINUX
Forwarded: no
Last-Update: 2020-04-19
The QStorageInfo/QStorageIterator implementation used for Linux is used also
on Hurd, as it uses an interface provided by GNU libc.
QStorageIterator::device() tries to use PATH_MAX (unavailable on the Hurd)
to lookup a /dev/block/ path, which exists on Linux only; hence, perform that
check within a Q_OS_LINUX block.
This code makes Lintian unhappy. But we are really not using it, it only
gets inserted when building the online doc.
Anyways the best way to calm down Lintian is to simply remove it.
The commit being reverted assumes the mutex is only protecting 'nameMap'
and nothing else is mutable, which is false. The mutex is not only
protecting 'nameMap' but also protecting the containing value objects,
since even though the value object is accessed read-only, its
implementation mutates its internal states for 2-way conversion between
ByteArray and QString.
Commit 85e61297f7b02297641826332dbdbc845a88c34b ("restore
QProcessEnvironment shared data thread safety on unix") said that
implicit sharing together with 'mutable' is a time bomb and the bomb is
triggered by the reverted commit.
Returning physical DPI from logicalDpi() is problematic,
as explained in commit 77e04acb.
The most predictable implementation is to never return
physical DPI from QPlaformScreen::logicalDpi(). Other
platform plugins already do this, and this change
brings xcb in line with the rest of Qt.
We have the QPlatformScreen::physicalSize() API which
covers returning physical DPI (indirectly); Options
for selecting which one to use can be implemented on
top of these (see QT_USE_PHYSICAL_DPI).
The fix changes the relaxed load of d_ptr in lockFor{Read,Write} after
the acquire of the mutex to an acquire load, to establish
synchronization with the release store of d_ptr when converting from an
uncontended lock to a contended lock.
The loadRelaxed() at the beginning of tryLockForRead/tryLockForWrite
isn't enough to bring us the non-atomic write of the recursive bool.
Same issue with the std::mutex itself.
It is a precondition violation to call QByteArrayView::at() with
size() as argument. The code used that, though, as an implicit
end-of-string check, assuming == ' ' and == '=' would both fail for
null bytes. Besides, QByteArrays (but most certainly QByteArrayViews)
need not be null-terminated, so this could read even past size().
To fix, use higher-level API (startsWith()), consuming parsed tokens
along the way.
Add null checks to cover the cases where QAccessibleTableInterface::cellAt
returns nullptr (which happens e.g. when called with invalid indices via
AT-SPI) or where the cell object doesn't implement the
QAccessibleTableCellInterface, which would previously result in crashes.
Cherry-picked into 5.15 as it fixes a crash in popular accessibility client
software. Conflict resolution: remove C++17'isms (`if` with initializer).
This allows to have system tray support on the fly on Wayland at least
where only QDBusTrayIcon is possible and no need to fallback to
QSystemTrayIconSys
Gbp-Pq: Name dont_fallback_to_x11_tray_on_non_x11.diff
The original commit was made based on a KDE workaround for
libdbusmenu-qt crash, but Qt is not using libdbusmenu-qt, Qt is not
watching for StatusNotifierHost registration and Qt is not capable
to switch backends on the fly leading to tray support being
not detected on Plasma Wayland sessions and falling back to the poor
legacy protocol on X11.
Gbp-Pq: Name revert_statusnotifierhost_checking.diff
We have the encrypted() signal that lets users do extra checks on the
established connection. It is emitted as BlockingQueued, so the HTTP
thread stalls until it is done emitting. Users can potentially call
abort() on the QNetworkReply at that point, which is passed as a Queued
call back to the HTTP thread. That means that any currently queued
signal emission will be processed before the abort() call is processed.
In the case of HTTP2 it is a little special since it is multiplexed and
the code is built to start requests as they are available. This means
that, while the code worked fine for HTTP1, since one connection only
has one request, it is not working for HTTP2, since we try to send more
requests in-between the encrypted() signal and the abort() call.
This patch changes the code to delay any communication until the
encrypted() signal has been emitted and processed, for HTTP2 only.
It's done by adding a few booleans, both to know that we have to return
early and so we can keep track of what events arose and what we need to
resume once enough time has passed that any abort() call must have been
processed.
Using O_PATH requires correctly specifying whether the fd is writable or
not. Stating that the fd is writable without it actually being writable
results into rejection on xdg-desktop-portal side. Other implementations
like xdg-open or gtk have also moved away from O_PATH fds so this will
make a matching implementation and avoid possible rejections from xdp.
Accessibility actually works when running applications as root, but we
would never properly connect, since the enabledChanged signal would be
emitted from the constructor in this case. So after connecting the
signal, check the value by hand to make sure not to miss the
notification.
Only applications running as root would be affected, because all other
applications would go through the asynchronous pattern of getting the
bus address from dbus instead.
Since ODBC does not have a direct way finding out if unicode is
supported by the underlying driver the ODBC plugin does some checks. As
a last resort a sql statement is executed which returns a string. But
even this may fail because the select statement has no FROM part which
is rejected by at least Oracle does not allow. Therefore add another
query which is correct for Oracle & DB2 as a workaround. The question
why the first three statements to check for unicode availability fail
is still open but can't be checked since I've no access to an oracle
database.
This hint is not really needed in the first place and only causes
problems in some environments.
For example in KDE, the compositor animates changes in position and size
for all ToolTip windows. However, this is not wanted here because we use
this window as a thumbnail for a drag-and-drop operation.
Before this patch the dragged element would lag significantly behind the
cursor. Now it works as expected, i.e. the dragged element follows the
cursor immediately.
Qt wayland platform plugin has improved quite a lot and it is now pretty
much usable on Gnome. It also improves user experience a lot on HiDPI
displays.
On some platforms, such as X11 and Wayland with some compositors,
QMenu could be a popup window, which should be set a transient parent
to get relative position, which is requested by Wayland.
Added transientParentWindow() for QMenuPrivate like QDialogPrivate.
The fusion style did ignore the QIcon::On icon because it reset
State_On to avoid the visual shift of a pressed button.
But it's not needed to reset this flag - the shift does not happen
because the fusion style does return 0 as offset for
PM_ButtonShiftHorizontal/PM_ButtonShiftVertical so no shifting will
happen.
When multiple globs match, and the result from magic sniffing is
unrelated to any of those globs, globs have priority and one of them
should be picked up.
Helmut Grohne [Tue, 5 May 2026 13:51:20 +0000 (16:51 +0300)]
call pkgconfig in order to be able to cross build qtbase with MySql.
Bug-Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=971604
Forwarded: not-needed Reviewed-by: Lisandro Damián Nicanor Pérez Meyer <lisandro@debian.org>
Qt's build system calls mysql_config... which won't work in a cross build
environment like Debian's, as it will throw an exec format error.
In order to solve this call pkgconfig and use mysqlclient.pc.
Pino Toscano [Tue, 5 May 2026 13:51:20 +0000 (16:51 +0300)]
Limit Linux-only code with Q_OS_LINUX
Forwarded: no
Last-Update: 2020-04-19
The QStorageInfo/QStorageIterator implementation used for Linux is used also
on Hurd, as it uses an interface provided by GNU libc.
QStorageIterator::device() tries to use PATH_MAX (unavailable on the Hurd)
to lookup a /dev/block/ path, which exists on Linux only; hence, perform that
check within a Q_OS_LINUX block.
This code makes Lintian unhappy. But we are really not using it, it only
gets inserted when building the online doc.
Anyways the best way to calm down Lintian is to simply remove it.
The commit being reverted assumes the mutex is only protecting 'nameMap'
and nothing else is mutable, which is false. The mutex is not only
protecting 'nameMap' but also protecting the containing value objects,
since even though the value object is accessed read-only, its
implementation mutates its internal states for 2-way conversion between
ByteArray and QString.
Commit 85e61297f7b02297641826332dbdbc845a88c34b ("restore
QProcessEnvironment shared data thread safety on unix") said that
implicit sharing together with 'mutable' is a time bomb and the bomb is
triggered by the reverted commit.
Returning physical DPI from logicalDpi() is problematic,
as explained in commit 77e04acb.
The most predictable implementation is to never return
physical DPI from QPlaformScreen::logicalDpi(). Other
platform plugins already do this, and this change
brings xcb in line with the rest of Qt.
We have the QPlatformScreen::physicalSize() API which
covers returning physical DPI (indirectly); Options
for selecting which one to use can be implemented on
top of these (see QT_USE_PHYSICAL_DPI).
The fix changes the relaxed load of d_ptr in lockFor{Read,Write} after
the acquire of the mutex to an acquire load, to establish
synchronization with the release store of d_ptr when converting from an
uncontended lock to a contended lock.
The loadRelaxed() at the beginning of tryLockForRead/tryLockForWrite
isn't enough to bring us the non-atomic write of the recursive bool.
Same issue with the std::mutex itself.
It is a precondition violation to call QByteArrayView::at() with
size() as argument. The code used that, though, as an implicit
end-of-string check, assuming == ' ' and == '=' would both fail for
null bytes. Besides, QByteArrays (but most certainly QByteArrayViews)
need not be null-terminated, so this could read even past size().
To fix, use higher-level API (startsWith()), consuming parsed tokens
along the way.
Add null checks to cover the cases where QAccessibleTableInterface::cellAt
returns nullptr (which happens e.g. when called with invalid indices via
AT-SPI) or where the cell object doesn't implement the
QAccessibleTableCellInterface, which would previously result in crashes.
Cherry-picked into 5.15 as it fixes a crash in popular accessibility client
software. Conflict resolution: remove C++17'isms (`if` with initializer).
This allows to have system tray support on the fly on Wayland at least
where only QDBusTrayIcon is possible and no need to fallback to
QSystemTrayIconSys
Gbp-Pq: Name dont_fallback_to_x11_tray_on_non_x11.diff
The original commit was made based on a KDE workaround for
libdbusmenu-qt crash, but Qt is not using libdbusmenu-qt, Qt is not
watching for StatusNotifierHost registration and Qt is not capable
to switch backends on the fly leading to tray support being
not detected on Plasma Wayland sessions and falling back to the poor
legacy protocol on X11.
Gbp-Pq: Name revert_statusnotifierhost_checking.diff
We have the encrypted() signal that lets users do extra checks on the
established connection. It is emitted as BlockingQueued, so the HTTP
thread stalls until it is done emitting. Users can potentially call
abort() on the QNetworkReply at that point, which is passed as a Queued
call back to the HTTP thread. That means that any currently queued
signal emission will be processed before the abort() call is processed.
In the case of HTTP2 it is a little special since it is multiplexed and
the code is built to start requests as they are available. This means
that, while the code worked fine for HTTP1, since one connection only
has one request, it is not working for HTTP2, since we try to send more
requests in-between the encrypted() signal and the abort() call.
This patch changes the code to delay any communication until the
encrypted() signal has been emitted and processed, for HTTP2 only.
It's done by adding a few booleans, both to know that we have to return
early and so we can keep track of what events arose and what we need to
resume once enough time has passed that any abort() call must have been
processed.
Using O_PATH requires correctly specifying whether the fd is writable or
not. Stating that the fd is writable without it actually being writable
results into rejection on xdg-desktop-portal side. Other implementations
like xdg-open or gtk have also moved away from O_PATH fds so this will
make a matching implementation and avoid possible rejections from xdp.
Accessibility actually works when running applications as root, but we
would never properly connect, since the enabledChanged signal would be
emitted from the constructor in this case. So after connecting the
signal, check the value by hand to make sure not to miss the
notification.
Only applications running as root would be affected, because all other
applications would go through the asynchronous pattern of getting the
bus address from dbus instead.
Since ODBC does not have a direct way finding out if unicode is
supported by the underlying driver the ODBC plugin does some checks. As
a last resort a sql statement is executed which returns a string. But
even this may fail because the select statement has no FROM part which
is rejected by at least Oracle does not allow. Therefore add another
query which is correct for Oracle & DB2 as a workaround. The question
why the first three statements to check for unicode availability fail
is still open but can't be checked since I've no access to an oracle
database.
This hint is not really needed in the first place and only causes
problems in some environments.
For example in KDE, the compositor animates changes in position and size
for all ToolTip windows. However, this is not wanted here because we use
this window as a thumbnail for a drag-and-drop operation.
Before this patch the dragged element would lag significantly behind the
cursor. Now it works as expected, i.e. the dragged element follows the
cursor immediately.
Qt wayland platform plugin has improved quite a lot and it is now pretty
much usable on Gnome. It also improves user experience a lot on HiDPI
displays.
On some platforms, such as X11 and Wayland with some compositors,
QMenu could be a popup window, which should be set a transient parent
to get relative position, which is requested by Wayland.
Added transientParentWindow() for QMenuPrivate like QDialogPrivate.
The fusion style did ignore the QIcon::On icon because it reset
State_On to avoid the visual shift of a pressed button.
But it's not needed to reset this flag - the shift does not happen
because the fusion style does return 0 as offset for
PM_ButtonShiftHorizontal/PM_ButtonShiftVertical so no shifting will
happen.
When multiple globs match, and the result from magic sniffing is
unrelated to any of those globs, globs have priority and one of them
should be picked up.
Helmut Grohne [Thu, 29 Jan 2026 19:25:23 +0000 (22:25 +0300)]
call pkgconfig in order to be able to cross build qtbase with MySql.
Bug-Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=971604
Forwarded: not-needed Reviewed-by: Lisandro Damián Nicanor Pérez Meyer <lisandro@debian.org>
Qt's build system calls mysql_config... which won't work in a cross build
environment like Debian's, as it will throw an exec format error.
In order to solve this call pkgconfig and use mysqlclient.pc.
Pino Toscano [Thu, 29 Jan 2026 19:25:23 +0000 (22:25 +0300)]
Limit Linux-only code with Q_OS_LINUX
Forwarded: no
Last-Update: 2020-04-19
The QStorageInfo/QStorageIterator implementation used for Linux is used also
on Hurd, as it uses an interface provided by GNU libc.
QStorageIterator::device() tries to use PATH_MAX (unavailable on the Hurd)
to lookup a /dev/block/ path, which exists on Linux only; hence, perform that
check within a Q_OS_LINUX block.
This code makes Lintian unhappy. But we are really not using it, it only
gets inserted when building the online doc.
Anyways the best way to calm down Lintian is to simply remove it.
The commit being reverted assumes the mutex is only protecting 'nameMap'
and nothing else is mutable, which is false. The mutex is not only
protecting 'nameMap' but also protecting the containing value objects,
since even though the value object is accessed read-only, its
implementation mutates its internal states for 2-way conversion between
ByteArray and QString.
Commit 85e61297f7b02297641826332dbdbc845a88c34b ("restore
QProcessEnvironment shared data thread safety on unix") said that
implicit sharing together with 'mutable' is a time bomb and the bomb is
triggered by the reverted commit.
Returning physical DPI from logicalDpi() is problematic,
as explained in commit 77e04acb.
The most predictable implementation is to never return
physical DPI from QPlaformScreen::logicalDpi(). Other
platform plugins already do this, and this change
brings xcb in line with the rest of Qt.
We have the QPlatformScreen::physicalSize() API which
covers returning physical DPI (indirectly); Options
for selecting which one to use can be implemented on
top of these (see QT_USE_PHYSICAL_DPI).
The fix changes the relaxed load of d_ptr in lockFor{Read,Write} after
the acquire of the mutex to an acquire load, to establish
synchronization with the release store of d_ptr when converting from an
uncontended lock to a contended lock.
The loadRelaxed() at the beginning of tryLockForRead/tryLockForWrite
isn't enough to bring us the non-atomic write of the recursive bool.
Same issue with the std::mutex itself.
It is a precondition violation to call QByteArrayView::at() with
size() as argument. The code used that, though, as an implicit
end-of-string check, assuming == ' ' and == '=' would both fail for
null bytes. Besides, QByteArrays (but most certainly QByteArrayViews)
need not be null-terminated, so this could read even past size().
To fix, use higher-level API (startsWith()), consuming parsed tokens
along the way.
Add null checks to cover the cases where QAccessibleTableInterface::cellAt
returns nullptr (which happens e.g. when called with invalid indices via
AT-SPI) or where the cell object doesn't implement the
QAccessibleTableCellInterface, which would previously result in crashes.
Cherry-picked into 5.15 as it fixes a crash in popular accessibility client
software. Conflict resolution: remove C++17'isms (`if` with initializer).
This allows to have system tray support on the fly on Wayland at least
where only QDBusTrayIcon is possible and no need to fallback to
QSystemTrayIconSys
Gbp-Pq: Name dont_fallback_to_x11_tray_on_non_x11.diff
The original commit was made based on a KDE workaround for
libdbusmenu-qt crash, but Qt is not using libdbusmenu-qt, Qt is not
watching for StatusNotifierHost registration and Qt is not capable
to switch backends on the fly leading to tray support being
not detected on Plasma Wayland sessions and falling back to the poor
legacy protocol on X11.
Gbp-Pq: Name revert_statusnotifierhost_checking.diff
We have the encrypted() signal that lets users do extra checks on the
established connection. It is emitted as BlockingQueued, so the HTTP
thread stalls until it is done emitting. Users can potentially call
abort() on the QNetworkReply at that point, which is passed as a Queued
call back to the HTTP thread. That means that any currently queued
signal emission will be processed before the abort() call is processed.
In the case of HTTP2 it is a little special since it is multiplexed and
the code is built to start requests as they are available. This means
that, while the code worked fine for HTTP1, since one connection only
has one request, it is not working for HTTP2, since we try to send more
requests in-between the encrypted() signal and the abort() call.
This patch changes the code to delay any communication until the
encrypted() signal has been emitted and processed, for HTTP2 only.
It's done by adding a few booleans, both to know that we have to return
early and so we can keep track of what events arose and what we need to
resume once enough time has passed that any abort() call must have been
processed.
Using O_PATH requires correctly specifying whether the fd is writable or
not. Stating that the fd is writable without it actually being writable
results into rejection on xdg-desktop-portal side. Other implementations
like xdg-open or gtk have also moved away from O_PATH fds so this will
make a matching implementation and avoid possible rejections from xdp.
Accessibility actually works when running applications as root, but we
would never properly connect, since the enabledChanged signal would be
emitted from the constructor in this case. So after connecting the
signal, check the value by hand to make sure not to miss the
notification.
Only applications running as root would be affected, because all other
applications would go through the asynchronous pattern of getting the
bus address from dbus instead.
Since ODBC does not have a direct way finding out if unicode is
supported by the underlying driver the ODBC plugin does some checks. As
a last resort a sql statement is executed which returns a string. But
even this may fail because the select statement has no FROM part which
is rejected by at least Oracle does not allow. Therefore add another
query which is correct for Oracle & DB2 as a workaround. The question
why the first three statements to check for unicode availability fail
is still open but can't be checked since I've no access to an oracle
database.
This hint is not really needed in the first place and only causes
problems in some environments.
For example in KDE, the compositor animates changes in position and size
for all ToolTip windows. However, this is not wanted here because we use
this window as a thumbnail for a drag-and-drop operation.
Before this patch the dragged element would lag significantly behind the
cursor. Now it works as expected, i.e. the dragged element follows the
cursor immediately.
Qt wayland platform plugin has improved quite a lot and it is now pretty
much usable on Gnome. It also improves user experience a lot on HiDPI
displays.
On some platforms, such as X11 and Wayland with some compositors,
QMenu could be a popup window, which should be set a transient parent
to get relative position, which is requested by Wayland.
Added transientParentWindow() for QMenuPrivate like QDialogPrivate.
projects / qtbase-opensource-src.git / log