Using strings to identify equal events is terribly inefficient. By
keeping a global list of event types we can assign each event a
numerical type index and use that to compare them.
We can also avoid excessive string copying and data duplication by
referring to the global type list where event type information is
needed.
Task-number: QTCREATORBUG-11823
Change-Id: I837bd5d0f5395b0003002ef8dd278fb27679c65d
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
We never have multi-string event data and dragging around the list
everywhere just adds noise.
Change-Id: I4c73543464abea01d342e3f0a296ed1b05ee2a88
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
All models are single-category now, so we can merge that into the base
class.
Change-Id: I4eab43ac2b4a77402121f004097e51a29b379aa4
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
With only one category per model we only have one eventType per model
and thus we don't need to differentiate per type anymore when
calculating the nesting.
Change-Id: Ic42a1c5c056f3480b7842a57fbff66a5e907abfb
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
This simplifies the code a lot and allows for more flexibility
when interacting with the data.
Change-Id: I69630071eee66840e905fcd95ba8c708742d58b6
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
This allows more precise specification of which event is supposed to be
selected.
Task-number: QTCREATORBUG-11945
Change-Id: Iff2e9bb8569711cc5df72a5ca55956e0091d6163
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
There is no reason to express colors as QVariantList.
Change-Id: I8c816e547b2f9be0f02a9d275791021ac06cff4f
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
Generally save both the Message and RangeType attributes so that we
avoid clashes between those types. Also keep all the types in one
place and make their names follow qtdeclarative's conventions.
Change-Id: I811bfcc4b72aaa2a0142babc92d96968ed2d4007
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
Having multiple views not only is bad for performance but also creates
difficult to debug problems on some hardware configurations and is
fairly confusing.
Task-number: QTBUG-38222
Change-Id: I885e800b1ededab9137874105e3b2f9ec88a06e8
Reviewed-by: Jocelyn Turcotte <jocelyn.turcotte@digia.com>
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
If there are too many events it takes too long to draw the overview.
By drawing only part of the events in each paint callback and
scheduling multiple paints we can mitigate that. The result is that a
rough outline is visible immediately and the details are filled in bit
by bit.
Task-number: QTCREATORBUG-12341
Change-Id: Id94d914e2926be01cb1635fbd8814c4d2f57d23b
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
If start and end times are initialized by clear() the change signals
may or may not be sent.
Change-Id: Ib50a2bbecf58791c5f93348e7450f4c0324f9d90
Reviewed-by: Daniel Teske <daniel.teske@digia.com>
Like this no one will get the idea that the socket state represents
the connection state and we can safely replace the underlying
implementation with something not derived from QAbstractSocket.
All the logging is retained but the connection creates the messages
now.
Change-Id: If84ff42f1fa9785254fbd49c75be867b9f663c83
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
The isConnected() method was misleading as it only checked for the
network socket to be connected. Sending data before the hello message
arrives is a pretty stupid idea and the debug clients need a way to
detect that. The new isOpen method stresses that it's about the
logical protocol, not the network socket.
Change-Id: I84f093d91d7a21d20a8f3d14730c9f189ec35bde
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
There is no reason for any external class to access that. If the call
to flush() from the qml profiler had any effect then it hid a bug that
should be fixed.
Change-Id: I8f655733bd3bff808653cd56b1904aa87c31f6c0
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
"state" is the more widely used term and we should follow that
convention. We also need to distinguish between the state of the
underlying network socket and the client itself. The change makes this
explicit.
As preparation for the upcoming centralized debug support the "State"
enum of the debug client is also moved into the QmlDebugClient class.
Change-Id: Ib9d7e03d23528f16ed696ed3518e813d11ea1c32
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
The context is stateful and there could be any kind of junk left in it.
In particular, the W3C spec says that it should be set up with a default
clipping path on creation, but it doesn't say what should happen if the
canvas is resized. Thus, we can never be sure that the clipping path is
sane.
Task-number: QTBUG-38297
Change-Id: If04597234e71574573888ce14deac5cf50df4a3f
Reviewed-by: Eike Ziller <eike.ziller@digia.com>
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
By passing signals with identification information of varying accurary
between the different views the event being selected could be changed
while it was selected. By checking the current selection against the
information given in the signal and not reselecting when it matches the
situation is improved.
Also, the selection methods are given more appropriate names. We hardly
ever want to select the "next" event, but rather the "best fitting" one.
Task-number: QTCREATORBUG-11945
Change-Id: I659b4929cb88f4c931a0893aa95a3bc92da5a23b
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
Using the extra window the ZoomControl keeps makes it impossible to
increase contentWidth to a point where it overflows. The drawback is that
the position on the scrollbar doesn't reflect the real position of the
visible part of the trace anymore if you zoom in to such a depth.
Task-number: QTCREATORBUG-11879
Change-Id: I6649f3c139f76c242a91d60364a28a4a00c9acee
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
This window is dynamically sized so that no integer overflows can occur
when using it in the timeline view instead of the whole trace time.
Task-number: QTCREATORBUG-11879
Change-Id: Id86faaf614b5f833e47ce26bb859b63eb09e7547
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
As most of the numbers involved are single precision floats rounding
errors can easily lead to such numbers. Instead of ignoring it set the
range to the smallest allowable value.
Task-number: QTCREATORBUG-11879
Change-Id: If8d08b27cc9e4cf2a63ff4973f519b0d3363178d
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
This is insignificant at lower zoom levels but becomes really annoying
when closer to the 500ns limit.
Task-number: QTCREATORBUG-11879
Change-Id: Ide0069f7c6b135aa31262a4396559fa3ba89f8e3
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
When zooming in the x offset and width of events can be larger than 2^31.
We can restrict them to the visible area, though.
Task-number: QTCREATORBUG-11879
Change-Id: I841300b55cdd583d5c3fa58b196101038f6f6036
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
This prevents some overflows when zooming into the timeline.
Task-number: QTCREATORBUG-11879
Change-Id: I968c4737af8c64798d196a1463268d86146864e7
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
This function is by a large margin the most often called one when
loading a trace. Passing it as a pointer to qSort certainly doesn't
help. Also, qSort is deprecated.
Task-number: QTCREATORBUG-11823
Change-Id: I98d744d1615733de93a8d35bccaa338643a2f6f4
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
Previously, if you selected an event with the same ID from a different
model the rangeDetails wouldn't be updated. Now modelId and itemId can only
be updated together and a single signal is emitted for that. This signal is
then used to update the details.
Change-Id: Ie1e971f5ac8c041b49df347fa0fbb401d5422766
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
The selectedEventChanged signal was never emitted and aliasing the
selectedItem property was unnecessary.
Change-Id: I2a15ad82bbf20a7ae9a42e17935242fbac7b5129
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
The iteration and comparison done by that takes significant time for
large traces.
Task-number: QTCREATORBUG-11823
Change-Id: I706b42f64ef0fd8b89229f51e52f0faaaf61d87a
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
Saving a trace is so much faster like this.
Task-number: QTCREATORBUG-11823
Change-Id: I5c68a16739d8bc49bfc0e3bb923ab23058aab6d0
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
Several QML properties need extra treatment to avoid stale data
after clearing. Toggling the visibility of the timeline renderer
doesn't really help there.
Task-number: QTCREATORBUG-11833
Change-Id: I1903d7bd69d6fe31ecc4cadca8e30fa2104fa09a
Reviewed-by: Christian Stenger <christian.stenger@digia.com>
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
That is what the code in the main view expects of an empty trace and
it's also logically more coherent.
Task-number: QTCREATORBUG-11833
Change-Id: I78b90ca5332bfb004cabbb32c4eb3e7e83fd7187
Reviewed-by: Christian Stenger <christian.stenger@digia.com>
Reviewed-by: Kai Koehne <kai.koehne@digia.com>
