| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
| |
None of the other code handles it like a multi hash, so let's change the
calls from unite() to insert since we now have an option which won't
silently create a multi hash.
Change-Id: I520295bde81c4ad1b0665308fcaf5fe41c6c3039
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
|
|\
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
src/imports/imports.pro
src/src.pro
Change-Id: Icdc39b6169d15b2102acd0e4d550a8d91e4b0744
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Just because we need a data stream version for QQmlDebugPacket, we don't
need to add such a huge dependency. Using a template here adds some
boiler plate to all users of QQmlDebugPacket, but this is worth it.
Task-number: QTBUG-60996
Change-Id: Ibade8757e28ac772942129c22c8721b504da3aee
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|/
|
|
|
|
|
| |
All known QML profiler clients track locations by now.
Change-Id: I1613da87dab22cfadfc25a35678d3e816ecda189
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
|
|
|
|
| |
QQmlDebugPacket is moved to PacketProtocol, and the other headers to
src/qml/debugger. This makes it easier to develop debug plugins in
external repositories.
Change-Id: I1ab099a304dad793ec13aaef414071cab6d7d408
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We have to make sure all profiler calls to one adapter are done from the
same thread.
It turns out that all the calls from QQmlTypeLoader are done from the
type loader thread. By using a separate adapter for that, we avoid any
extra locking.
Task-number: QTBUG-62987
Change-Id: I5cffcc30b84dd4154daa0c56312a1141ce2033e1
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
|
|
|
|
|
|
|
| |
It benchmarks QElapsedTimer and QDataStream. We should do this in qtbase
if we need to do it.
The test was not in the parent qml.pro, but it was mentioned in some
outdated comments in the QML profiler plugins. Remove those comments, too.
Change-Id: I0d1341c32f4a2e02a04a958f76be015fe8d927fb
Reviewed-by: Erik Verbruggen <erik.verbruggen@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
They are decoded as ints, as in
QPacket &operator>>(QPacket &ds, QmlObjectProperty &data)
{
int type;
ds >> type >> data.name >> data.value >> data.valueTypeName
>> data.binding >> data.hasNotifySignal;
data.type = (QmlObjectProperty::Type)type;
return ds;
}
So we should serialize the same, since the required operator<< and >>
don't exist.
Change-Id: I33dc971f005a4848bb8ffffd1478edcf2e916dfc
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
|
|
|
|
|
|
|
| |
This way QtQml doesn't need to know the ctors.
Change-Id: Ie74049092b5eb9837537591c0cf37ad1487e4066
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
| |
Change-Id: I6746b777f73d047f5cf610bfca9b320ac1e13676
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
|
|
| |
This saves time when serializing the data to be sent.
Change-Id: Ic8c534d55445934a64dd253273099194b27d98af
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As the various file names are actually not kept as QStrings in the
respective objects being profiled, our saving them as QStrings in each
and every profiling event is not implicitly shared and causes a huge
memory overhead. Avoid that by saving each location only once, indexed
by a disguised pointer to the object it refers to.
Normally, objects could disappear during the profiling session, and new
objects could be allocated in their place, which would mess up our
indexing system. We prevent that by referencing the objects when we
index them, thus preventing them from getting auto-destructed.
Mind that those are not JavaScript objects but rather functions,
bindings, components and the like. So, this will only cause a memory
leak if you're compiling and dropping QML components over and over.
Task-number: QTBUG-52937
Change-Id: Ia4dfb09a71a5c9a2d6ce25c3811bbe2a1036c1c1
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
There can only be one detail type for compressed set of messages. This
is unlikely to change anytime soon. Also, be more careful about not
detaching the data when sending the messages.
Task-number: QTBUG-52937
Change-Id: I859b12889b0766543dac00b21784e14b76e458b6
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This reduces memory churn as we don't have to reallocate the various
QPacket, QBuffer, QByteArray objects and their private classes all the
time. Also, subsequent packets often have similar sizes, which makes it
beneficial to preallocate the underlying byte array to the size of the
previous packet on clear(). In order not to carry the extra reserved
space in the lists passed to the debug server, we squeeze the byte
arrays before inserting them into the lists. That detaches and copies
them, which would have been done anyway when reusing the same packet
for the next message.
The result is a reduction in the number of temporary memory allocations
from QQmlProfilerServiceImpl::sendMessages() by about 75%.
Change-Id: Ief98cd7a93a2e8e840e111dee6b346cae00e06bf
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
This enables more parallel data processing and limits the memory
usage. Benchmarks with the "planets" example from canvas3d show that
this change reduces the time between the profiling being stopped and
the last events arriving in the profiling client by about 50%.
Change-Id: Iea16f9e2ae3adf584ec4a3c7fc766eaa21740f98
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
| |
From Qt 5.7 -> LGPL v2.1 isn't an option anymore, see
http://blog.qt.io/blog/2016/01/13/new-agreement-with-the-kde-free-qt-foundation/
Updated license headers to use new LGPL header instead of LGPL21 one
(in those files which will be under LGPL v3)
Change-Id: Ic36f1a0a1436fe6ac6eeca8c2375a79857e9cb12
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We cannot use the same data stream version for the client and server
versions of QPacket and QPacketProtocol should not deal with QPackets
but with simple byte arrays because the underlying QDataStream is hard
to copy.
The new QQmlDebugPacket picks its data stream version from
QQmlDebugConnector now, which adjusts it when connecting. As there can
only ever be one QQmlDebugConnector, we can keep the version static.
The clients need to query the connection for the correct version. We
may connect to several different servers sequentially or we may have a
server running while using a client, and we don't want to confuse the
versions between those.
With this in place, all remaining occurrences of QDataStream are
replaced with QPacket or QQmlDebugPacket.
Change-Id: I3f6ba73fcbfad5e8df917c5feb9308116738a614
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
| |
This saves some code in QtQuick and allows us to split up QPacket into
different versions for client and server.
Change-Id: I9b8c723274fb11d6321c5002910148b193aa6b40
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
| |
There is only one type of binding left and as all clients check atEnd()
before trying to get the bindingType from the stream we can just remove
all of this.
Change-Id: I4d50ef4a691630b521292cb622ef539b3494a970
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
We need it in 3 places in qtdeclarative and we could also use it in
QtCreator. We don't want to bundle it with the debug client code as it
is also necessary for the server.
QPacket replaces QQmlDebugStream as it has the same purpose. This
also fixes the inconsitent handling of data stream versions.
Change-Id: I650fae353f267511c551b427d9169f4d718aa7f2
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The compilation of .js resources was missed and the compile times were
underestimated because the dependency resolution and error checking
wasn't factored in. In particular we only profiled the done() step for
QQmlTypeData. However, JavaScript compilation also occurs in the
dataReceived() step of QQmlScriptBlob and QQmlTypeData will already
parse the program into an AST at that step.
Compile steps can be nested now, but considering the fact that
significant time may be spent before and after compiling dependencies
for a parent module, this seems to be the best way to model them.
Furthermore, in order to not needlessly convert QUrl to QString at
runtime, the compilation profiler saves the files now as QUrl.
Change-Id: I215a87787f9117c069ecd77b2d913cc0b0ff3c89
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|
|
Change-Id: I12627a07ceedea4aceafa6f0e630c0cab69d156d
Reviewed-by: Simon Hausmann <simon.hausmann@theqtcompany.com>
|