| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
| |
The pro-file is missing '+' when it is adding QtOpenVg support for sis
file.
Task-number: None
Reviewed-by: Janne Koskinen
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Destroying the control environment also pops items from the original
cleanup stack which is not owned by the control environment.
This is a platform issue (and does not occur in 9.3 & 9.4).
To avoid this causing a panic, the s60main is changed to not push items
on the cleanup stack (instead, it TRAPs the call to main(), and deletes
its allocations manually).
A further 9.2 cleanup stack crash was caused when application construction
fails (because of missing resource files). The resulting exception would
bypass deleting the control environment, and return from main with the
wrong cleanup stack active.
This is resolved by doing the TRAP / cleanup / throw manually instead of
using qt_trap_throwing (The control environment cannot be pushed to the
cleanup stack, because it owns the cleanup stack at that time).
This isn't a 9.2 specific bug, but rather is revealed by 9.2 because the
missing resource file is non fatal in 9.3 and 9.4.
Fixes many autotests which have crashed on S60 3.1 since the cleanup
stack swapping patch.
Reviewed-by: axis
|
| |
|
|
|
|
| |
Leaving function called without a TRAP
Reviewed-by: Sami Merila
|
| |
|
|
|
|
|
|
|
|
| |
Problem was caused by S60 closing a handle under our feet when QApplication
is destroyed. To avoid this, we open our own handle for the global lock
instead of using the shared one inside CFbsBitmap.
Also the global unlock/relock is not needed on S60 3.2 so this can be
eliminated and save a few cycles (the function call was a no-op before)
Reviewed-by: Jani Hautakangas
|
| |
|
|
|
|
|
|
|
|
|
| |
Symbian^1 - SV_SF_1
Symbian^2 - SV_SF_2
etc.
This is for the benefit of developers working with the Symbian master
codelines.
Reviewed-by: axis
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
QS60Style ignores toolButton's menu indicator rect. This means that the
button's menu cannot be triggered by the user, as the style does not
draw the indicator at all.
With this fix, style calculates a rect (including margins) for menu
indicator and draws that (arrow down) into toolButton (by making the
button bevel larger).
Task-number: QTBUG-5266
Reviewed-by: Alessandro Portale
|
| |
|
|
| |
RevBy: Trust me
|
| |
|
|
| |
RevBy: TrustMe
|
| |\ |
|
| | |
| |
| |
| |
| |
| |
| | |
fontDef.pixelSize is a qreal now and we should round it after
multiplication to avoid an accuracy loss.
Reviewed-by: mae
|
| | |
| |
| |
| |
| |
| |
| |
| |
| | |
Not sure if this is right fix.
We could also disable PLUGIN_PACKAGE_SIMPLE_HASH.
But this is automatically enabled when NETSCAPE_PLUGIN_API is
disabled.
Reviewed-by: thartman
|
| | |
| |
| |
| | |
Reviewed-by: trustme
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
qrand() uses global static seed storage, but if used
from destructors of other global static objects
(directly or via ~QSettings() or QMap::insert())
the global static seed storage can be already deleted.
Fallback to plain rand() in this case.
Task-number: QTBUG-5218
Reviewed-by: Justin McPherson
|
| | |\ |
|
| | | |\ |
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
MinGW doesn't have _aligned_malloc and posix_memalign doesn't have a
realloc function. So use our own implementation in all platforms.
Removing posix_memalign was reviewed by Brad, but I apparently lost
the commit.
Reviewed-by: Trust Me
|
| | | |\ \ |
|
| | | | | |
| | | | |
| | | | |
| | | | | |
Reviewed-by: Trust Me
|
| | | | | |
| | | | |
| | | | |
| | | | | |
Reviewed-by: Olivier Goffart
|
| | | | | |
| | | | |
| | | | |
| | | | | |
Reviewed-by: TrustMe
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
... as described in the documentation. Furthermore:
* use qt_safe_select to timeout correctly
* return immediately when timeout value is 0
Reviewed-by: Oswald Buddenhagen
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
This is future compatibility: we must rely on them being 0 in older
versions of Qt.
Reviewed-by: Bradley T. Hughes
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
QContiguousCache is a new type, so there are no binary compatibility
issues.
QHash and QMap didn't have any public qMalloc / qFree, so the entire
logic is contained in .cpp code. However, since old code will not
inform us of the alignment requirements, we need to add a bit to the
structure to indicate whether strict alignment is in use or not.
QList doesn't require any changes. For small, movable types, they're
all stored in the pointer array itself, so they're aligned. For larger
types, we use new(), so types with stricter requirements should define
their own operator new().
QLinkedList cannot be fixed. It uses new() on the QLinkedListNode,
which contains a T type. Sorry.
QVector did have public qMalloc / qFree. I've moved the calls to the
inner function and made it keep the old calls if the alignment
requirement is below a certain threshold. The idea is that, if it's
above, no one was using QVector anyway.
Reviewed-by: Bradley T. Hughes
|
| | | | | | |
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
The private inheritance ensures that we don't try to access the types
under the wrong pointer. If we did that, we'd cause strict aliasing
violations.
Reviewed-by: Olivier Goffart
|
| | | | | | |
|
| | | | | | |
|
| | | | | |
| | | | |
| | | | |
| | | | | |
We want people to not use this signal directly.
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
We were using this signal to update the signal hooks when the remote
service changed. That meant each Qt app received every single service
creation, change or destruction.
Now we only watch the services we're really interested in.
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
QDBusConnectionPrivate
I'll need to recurse into the signal connection mechanism in the next
commit.
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Use null services to indicate we're not interested in the owner, but
empty-but-not-null to indicate we don't know what the owner is. Since
empty service names are not valid, this will mean that this rule won't
match.
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
This allows us to listen only to the activations we're really
interested in.
|
| | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
matching.
The bus allows us to match string arguments when receiving messages.
This is very useful for the NameOwnerChanged signal, whose first
argument is usually what we're interested in.
By using these new functions, you can restrict receiving of signals to
those that you truly want, instead of receiving NameOwnerChanged for all
services registered/unregistered on the bus.
|
| | | | | |
| | | | |
| | | | |
| | | | | |
I'll use this feature to match the NameOwnerChanged signal from the bus.
|
| | | |\ \ \ |
|
| | | | | | |
| | | | | |
| | | | | |
| | | | | | |
Reviewed-by: João Abecasis
|
| | | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
This fixes a timeout that occurred on Mac with the gui event dispatcher:
we were waiting for a write notification, but timed out when we were in
closing state and still waiting for the socket engine to complete
writing.
Now we close the socket anyway after 2 seconds.
Reviewed-by: Thiago Macieira
|
| | | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
only disconnect from host when all bytes have been written; i.e. not
only check whether the write buffer is empty, but also check whether
the socket engine has still bytes to write. This is necessary for
HTTP and SOCKS5 socket engine, because they both contain an inner TCP
socket which also does buffering. For the native socket engine, there
is no difference with this patch.
Reviewed-by: Markus Goetz
Reviewed-by: Thiago Macieira
|
| | | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
When loading a library fails, the error message "The application or DLL
... is not a valid Windows image. Please check this against your
installation diskette." is shown, which is not very helpful.
Task-number: QT-2357
Reviewed-by: Marius Storm-Olsen
|
| | |\ \ \ \ \ |
|
| | | |\ \ \ \ \ |
|
| | | | |\ \ \ \ \ |
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
and removes THREAD from qfeature
Reviewed-by: Trust Me
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Reviewed-by: tom
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Reviewed-by: Trust Me
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Reviewed-by: tom
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Reviewed-by: tom
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Reviewed-by: tom
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Reviewed-by: Trust Me
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Reviewed-by: tom
|
