| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
| |
The former was just the private implementation of the latter. This way,
the code structure is more understandable to newcomers and follows
existing best-practices.
Change-Id: I07cf64370553f4c2de349b5f01e90b31112fee58
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
| |
Add space between type name and & or *.
Change-Id: I64bfe20510cb43ee0a0b6e08bd433fc657e925a0
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
|
|
|
|
|
|
| |
Change-Id: I92c5b00d5bbcc08a241ed0382c13b6bf2676ca6f
Reviewed-by: Milian Wolff <milian.wolff@kdab.com>
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This removes the custom WebSocket server implementation and replaces
it by a dependency on the QtWebSockets module.
Sadly, the QtWebSocket module does not yet support custom protocols.
Also, there is quite some boiler plate code required, something which
I want to simplify upstream in the QtWebSockets module later.
Change-Id: I8066418fb1857d23b8593c443bc9a98ded917a99
Reviewed-by: Kurt Pattyn <pattyn.kurt@gmail.com>
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This enables us to optionally use navigator.qt instead of a WebSocket,
which is nicer setup-wise and is also slightly faster:
navigator.qt:
284.0 msecs per iteration (total: 2,840, iterations: 10)
WebSocket:
295.8 msecs per iteration (total: 2,959, iterations: 10)
The baseline is ca. 203 msecs, which would mean a performance boost
of ca. 12.7%.
Furthermore, this sets the fundation to eventually add a WebEngine
transport mechanism. The WebViewTransport should also be removed and
instead the WebView itself should directly implement the
WebChannelTransportInterface.
Change-Id: I368bb27e38ffa2f17ffeb7f5ae695690f6f5ad21
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
| |
Change-Id: I7dc78213bc57cbfd63b021e00823c4aad105aa05
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
| |
Wrap everything in the QtWebChannel module with the Qt namespace or
use the Qt namespace where appropriate.
Change-Id: I1ef2b2f5eb22ec5e04ca76c034ef8ebf4043b899
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Some tests referenced Nokia in their license even though that was never
the case. The tests where written completely by me after Qt Nokia times.
What is missing are the examples which are still mostly original work
by Noam back then in Nokia times. The rest was (re-)written by me
completely since then anyways.
Change-Id: Ib423fb3459bcc1f7464a02de4fd82ddfd614d282
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This used to work but broke in one of the last few commits apparently.
Now we add a proper unit test to ensure it stays working.
The issue was that JSON only knows e.g. numeric types stored as double.
When we then try to call a method taking an int with the VariantArgument
that tries to convert the double to int, we failed and produced an
invalid QVariant which then converts to 0.
Now we use the appropriate API to convert the JSON data to the correct
target type before calling the method. Furthermore, it became clear
that we can greatly cleanup the VariantArgument thanks to that. It now
is reduced to just a QVariant wrapper class with an implicit cast
operator to QGenericArgument.
Change-Id: Ieaf60f548ea9584e7d760f9cd935da455787f376
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
The new registeredObjects list property is now preferred over the old
imparative registerObject/registerObjects API.
Items that are added to the list need an attached WebChannel.id property
which holds the identifier under which the object is published to remote
clients.
Change-Id: I96a8047b9a85e27f3fd48c900180c22ebd20eb35
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
QMake reported the following warnings which is fixed by this patch:
QtWebChannel: WARNING: .../qwebchannel/src/webchannel/qwebchannel.h
includes qwebchannelglobal.h when it should include
QtWebChannel/qwebchannelglobal.h
Change-Id: Ib6a330e372a9de32d9578aa17ea0d74257a23676
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Qt 5.3 propagates var-arguments of signals as QJSValue instead of as a
QVariant. This then fails to be serialized to QJson, failing our unit
tests.
Now, QJSValue types are manually casted to QJsonValues which makes
the tests pass again.
Change-Id: I730c595eee214ebe3d1f83009cd5605f66407f55
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
| |
Change-Id: I73a83380c571ed5a400b16cb255562bb8079eaac
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
This is achieved by hiding the MetaObjectPublisher completely as
private API. The QWebChannel is the only publisher API and now handles
both the socket as well as the publisher internally.
This now allows us to create a proper QML api in the new QmlWebChannel.
Change-Id: I3096364af8485353ca9bc19df4a81a8e4552c3d7
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
The code now resides in a single qwebchannel.js file and there is only
a single callback-nesting required to setup a MetaObjectPublisher
connection.
The server-side will be simplified in the next step.
Change-Id: Ib5fc77a03c2b281c61af91713411eed571ec6108
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This reduces the traffic as the indices are usually much smaller than
the property names.
As such, the benchPropertyUpdates gets a speed boost of about 9% (or
10ms vs. 11ms). As we need to transmit the index during initialization
that degrades its performance slightly by ca. 4% (13ms vs. 12.5ms).
Considering that the initialization only takes place once whereas
the property updates potentially often, this is a good tradeoff.
Change-Id: If7df3e360f1528b7d7aa26c63ce851363ae9fd6a
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
| |
When handling the destroyed signal of a QObject, the QMetaObject of the
sender() will point to the global static QObject meta object. Thus, we
also cache its signal argument types. This way, we are able to properly
handle the destroyed signal with minimum effort.
Change-Id: Iba1a3fc94d55adad178302cc847fd4285815e689
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
Before, we constructed QVariant maps and lists and then converted them
to JSON to send the data to the webchannel.
By obsoleting the conversion step, benchInitializeClients shows a good
performance boost of ca. 19% (11.81ms vs 14.58ms).
Change-Id: Ief8e8127207a046f481488a478cd6a18fa0ebffe
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
| |
This method is quite slow, and can be compared to two calls of sender().
We now encode the signal index in the methodId and thus do not have to
call it anymore. The performance gain is about 27% (9.2 vs. 12.7ms).
Change-Id: Iaa75efa27a54a21e27e62994de25cafd8136159d
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Profiling shows that sender() and senderSignalIndex(), which again calls
sender() internally, are the hotspots in benchPropertyUpdates.
To speed things up a bit, call senderSignalIndex only once whenever a
signal is emitted. The performance gain is about 27% (16ms vs. 22ms).
While at it, also re-use the global s_destroyedSignalIndex static and
also call sender() only once even when assertions are enabled.
Change-Id: I90cd1a2b453e5c40d0f41276968f4545b42076bc
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
This allows us to remove the public API for the tests and allows for
more tests and benchmarks in the future.
To achieve this, we re-use the new qmetaobjectpublisher_p.h, which then
also must be exported.
Change-Id: I3c33b2f5be6cc674cd3092667151dd8da2263cf5
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This example shows how to use the (currently quite ugly) raw C++ API
to setup a webchannel without using QML at all. The HTML client is then
handled by the users default browser.
The example itself shows a simple chat between the HTML client and the
C++/Qt server, with a line edit for input and a text edit showing the
chat history.
Change-Id: I8baf14efb9d0c5f5880d99710cf6317fe9b887b9
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This module can hopefully be done in time for 5.3. This commit changes
the source structure and QMake files to adapt to typical Qt modules.
With this in place, we can now use QT += webchannel in qmake files to
link against the pure Qt/C++ QtWebChannel library.
The QML plugin is separated from it and can be loaded optionally, if
the quick module could be found. Also added is now a qmlplugindump
for tooling integration. Note that the Qt.labs namespace is removed.
The test file structure is also adapted to how its done in the
QtDeclarative module.
Note that this setup apparently does not support to run tests without
running make install first.
Change-Id: I1c15d72e7ab5f525d5a6f651f4e965ef86bc17bd
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This will allow us to create a stand-alone WebChannel C++ library,
without any QML dependencies, that can be used to publisher QObjects
to remote clients running in any browser engine supporting WebSockets.
The patch is large, as working with introspection through the
QMetaObject from C++ is more complicated compared to QML.
On the other hand, the move to C++ allows a much more performant
implementation of quite some parts of the publisher. One thing is
that signal and method invocations can be handled via numeric indices,
where before we needed to transmit the string identifier of the
signal or method.
Eventually this can now also be applied to properties, further
decreasing the size of messages between server and clients.
Note that this patch contains quite some TODOs and rough edges,
such as the invokable bench_* helper functions in the public API
of the MetaObjectPublisher. These are to be seen as temporary, and
will be cleaned up in followup commits later. This is done to prevent
a further blow-up of this already big patch.
Change-Id: I57e788d8a19edd410651611382d912f9ad6660c9
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
By handling more logic on the C++ side, such as serializing the JSON
data to QByteArray, we can save up to 6% in the propertyUpdates
benchmark.
Furthermore, this allows for some more code cleanup and obsoletes the
WebChannel.qml file.
Also, it is a first step towards removing the QML dependency and
making it optional alltogether.
Change-Id: Id610b5f2652da4a7ad867aef576fabcc40d3d92c
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
The map-initializations are not required anymore, as the properties
are now properly initialized.
Furthermore, Qt.Debug messages are only handled by the
MetaObjectPublisher, thus the webChannel.debug message should only be
available to HTML clients which use qobject.js.
Change-Id: Iae11a73d33d5eec3a90a264bf0418a5781523a22
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
| |
To do so, we need to remember the connected functor and pass that to
disconnect, in both the client side as well as the QML server side.
Change-Id: Ic61fc5d2a203212278c23471c216683e309e2c9f
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
| |
Change-Id: If26e033dcb031678815277890be4ad2ebad3fd4b
Reviewed-by: Milian Wolff <milian.wolff@kdab.com>
Reviewed-by: Thomas McGuire <thomas.mcguire@kdab.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
An HTML client can trigger the creation of new objects, or a published
object on the server side might return other objects.
These are then wrapped on the fly and can be used like the other
objects. Note though that the HTML client can call deleteLater on
these objects. Also, it does not yet work to wrap objects on the fly
in signal arguments or property values.
Change-Id: I92aa8a3e52f42d5325dd0771bbf9e2ae213e88f9
Reviewed-by: Arvid Nilsson <anilsson@blackberry.com>
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
|
| |
No other changes seem to be required.
Change-Id: Ie15639f74f09cbd303828c96f75d29283ec4d562
Reviewed-by: Arvid Nilsson <anilsson@blackberry.com>
Reviewed-by: Chris H-C <chutten@blackberry.com>
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This includes two optimizations which have been forgotten to be
upstreamed before:
a) For the common case of property notify signals named "...Changed",
the string name is now not send anymore to the client. Instead, these
cases are special-cased and constructed on the fly. This drastically
reduces the size of Qt.init responses and property update messages.
b) Furthermore, do not list signals as methods, further reducing the
size of Qt.init response messages.
Together this shows a noticeable reduce of CPU instructions in the
benchmarks as recorded with perf:
benchmark_classInfo: down ~10%
benchmark_initializeClients: down ~2%
benchmark_propertyUpdates: down ~1%
Change-Id: I01e59f5c1dceedb893f7a3e3e127acb493baaa7f
Reviewed-by: Michael Bruning <michael.bruning@digia.com>
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
|
| |
Instead of initializing the property maps in Component.onCompleted, use
"var foo: ({})" instead. This looks suprising, but the "var foo: {}"
syntax won't work - it's an empty binding expression closur, not a map.
Change-Id: I9edeeeeeabb3a871a46d8bb8a991a4155040497a
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
| |
For performance reasons, we group property updates and send a single
batch update notification. This is now tested to work as intended.
This also uncovered a bug in webchannel.js when multiple functions
subscribe to the same id.
Change-Id: Ic8648d664dd1fe54df7e25fade6a6088386af992
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This tests the functionality of publishing a plain QtObject from
QML to the HTML client.
It tests property binding, i.e. reading and writing of an objects
property on the client side, as well as change notification tracking.
Furthermore a server-side method is invoked from the client and
signal submission from the server to the client is tested.
Change-Id: I62e544cddf4483b57535a9bc1e05a36105ec6622
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This removes a lot of obsolete files and simplifies the build system
of the examples.
Furthermore, the examples can now be run without running make install
first. It reuses the same import path as the test does.
Note that the examples are not installable anymore now though. If this
is required, it can be added again.
Change-Id: Ic7ff80f734b035a03fb1a11a2df492c97298ceff
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
This uncovered a bug in webchannel.js, which stringified strings leading
to duplicated quoting. This is also fixed now.
Furthermore, some QMake changes are required to make it possible to run
the tests without first installing QWebChannel.
Change-Id: If7e8f73a748f86f2d5c7d39000e90612367038af
Reviewed-by: Zeno Albisser <zeno.albisser@digia.com>
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: Idf344f46aa09a13dfe4db00203b7644006fbf944
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
| |
This is required for factory-like methods on the C++/QML side,
which we want to access from the HTML side as well.
Change-Id: I2852bbc9c8effb6d6f49b5be784241a6e2320823
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is a big code drop - sorry for that. The benefits are worth it
though, I'm sure. The optimizations were required to make the
WebChannel useable even on a low-end embedded device with medium
amount of traffic.
The changes in this patch can be grouped into different parts:
a) Do more in C++: Esp. by leveraging e.g. the new classInfoForObjects
in QtMetaObjectPublisher (on the C++ side) one can greatly reduce
the time required for initialization of the webchannel.
b) Property Caching: Instead of requiring a socket roundtrip whenever
a property is read on the HTML side, we now cache the property values
on the HTML side. Note that for this to work properly, one needs to
add proper notify signals to the property declarations, as otherwise
the cache will not get updated.
c) Grouping: Instead of sending separate messages to the clients
for every property update, these signals are grouped by a 50ms timer,
and then send aggregated to the client. This reduces the socket
traffic, as more boiler plate can be shared.
d) Compression: Some data was previously send repeatedly, such as
property name and notify signal. This is now compressed internally
where possible (i.e. for the ${propName}Changed naming scheme).
e) Message Flood Prevention: Previously, one could easily flood an
HTML client by sending data to it. If it could not work off the
incoming stream one would freeze the HTML client. Now, we wait for an
idle signal of the client prior to sending new data to it. Paired
with the message grouping and property cache mentioned above, we
are able to only send the newest data once the HTML client becomes
active again. I.e. we discard now-obsolete property updates etc.
Change-Id: I8f3ae16ed6c1f6a89b644acdce7efbf0f07fc786
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: Idd32243173775ec49b4a51a55faa85e47e11a4f1
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: I47345fc52677b68ae75d018484b495fc81949054
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: Id4400b63f0bd5180194523f1efbac8b82c4dbe91
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: Id7b35aab5012e1eba84fb3685b3bc6619aa92580
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
| |
For channels with multiple clients we used to create it once for
every client which is not needed.
Change-Id: Ib1be0c9f7bc78c0415fe2e9f6f8aa5112d0156c6
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: I18a596c18530b18a833a58e734ce484caa6ae68f
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
| |
In principle everything worked before already, with the difference
that responses to channel.exec could not properly be distinguished.
This is now fixed by adding a UUID to the WebChannel and sending that
along with the integer exec ID and comparing it in response messages.
Change-Id: I954716b04c1d9a41fffb8b9bb736a1fa45fec7f2
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: I5948de7edff3aa8a58c9cc6e3789c4e7fffb7260
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: I269ded913b25bb3d3869f2da85ce0ff7449b534b
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
| |
Change-Id: Id825cc0095398d72922700fa5c0d4f30cf19353c
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|
|
|
|
|
|
|
|
| |
Enums that are marked by Q_ENUMS in a Q_OBJECT or Q_GADGET are thus
accessible in JavaScript via obj.EnumName.EnumKey.
Change-Id: Ia3e92da9bc05e06011f250ec8f5cf6ac26a3b0f4
Reviewed-by: Pierre Rossi <pierre.rossi@gmail.com>
|