On Sat, Jan 16, 2016 at 4:03 PM, Dimitris Chloupis
<kilon.al...@gmail.com> wrote:
> Hello Ben :)
>
> First thanks for investing 1 hour of your precious time to look through my
> code. I really appreciate it . If there is one thing I love about pharo is
> the super amazing community.
>
> Yes definetly my fault as always :D
>
> I did not know about this profiling, I used it for the creating of my Morphs
> but it did not occur to me to google like you did and go with profiling all
> processes.
>
> A huge thank you to latsaben and the rest of you guys. I downgraded down to
> Pharo 4 and again I agree 5 is the unstable version anyway but I loved using
> the new tools and could not wait till April for the Pharo 5 release :)
>
> The problem is battling without documentation is very hard and I have spent
> at least 10 hours trying to understand how taskbar works and I still fail.
>
> Your code solves my problem , apart from 1. Loading those pngs is sloowwwww,
> one would expect that accesing the files themselves to get the binary data
> is what is slow for pharo but no, thats is fast enough. What is slow ,
> because I profiled this is converting the binary data to form is consuming
> a ton of time. The result is that to load a GUI with only 1 mb of pngs takes
> 1 second.
What is the practical effect of this that is most concerning?
1. The time taken loading it the first time after installing from
Catalog Browser ?
2. The time taken to start it *each* time from World > Chronos
Manager, if you open/close it a lot?
3. Distributing Chronos Manager as a packaged application?
each time you want to open it
Not sure if the following will suit your needs, but in the definition
of ChrGUIMorph try moving 'secondsTimerFormCollection' from
instanceVariableNames: to classVariableNames:
then test by opening/closing the app a few times. After taking a
while opening the first time, subsequent openings seem to be faster.
>
> Also with your solution I lose the ability to update my taskbar icon, by the
> way the non optimised code I was using is a direct copy from SystemWindow
> class, which means that class should be optimised too.
So consider that solution just a problem-isolation step on the way ;).
I had another look because I was intrigued that even though you used
LRUCache for the /icons/ variable, the ifAbsent: block in
#logotinyIcon was being executed anyway. I thought that must be cache
size too small and items bumped out, so as a test I replace LRUCache
with Dictionary, but the the ifAbsent: block was *still* getting
executed (note, this is at steady-state, not startup). So I thought
/icons/ must be getting reset. Investigating by putting the following
at the top of #initializeIcons...
Transcript crShow: thisContext printString ,
' <-- ', thisContext sender printString,
' <-- ', thisContext sender sender printString.
So now this executes every time I open the World menu. And after
ChronosManager is opened, the Transcript scroll rapidly. So everywhere
you do something like this...
icon: ChrStopwatchSettingsPNG new logotinyIcon
where logotinyIcon returns an object, the object created by #new is
thrown away. Worse, #new invokes #initialize and hence
#initializeIcons wipes out your cache every cycle.
Try making /icons/ a class variable and moving methods like
logotinyIcon to the class side, so instead you do...
icon: ChrStopwatchSettingsPNG logotinyIcon
Also move #initializeIcons to class side and add a class side
#initialize method to call it. Note that will only be executed when
the class is first loaded, so you will probably need to execute it
manually in your image.
One thing I'm interested in the choice of LRUCache over Dictionary for
the cache? Did you find Dictionary was taking too much space?
I didn't try implementing the above yet, but I guessing you'll get the
performance fix while maintaining the flexibility you want.
cheers -ben
> Another problem is the scaling of images, really bad with morphic though
> Athens must be better.
>
> Auto completion problems are well knows, sometimes autocompletion does not
> show all methods, sometimes autocompletion shows every character you type so
> for example ChronosManager open will show
> ChronosManager o
> ChronosManager op
> ChronosManager ope
> ChronosManager open
>
> Sometimes it shows methods that do not even belong to the class forcing me
> to use Spotter to find the correct method. Its a a mess.
>
> I am not abandoning pharo, I love it even with its flaws. But I try to
> outsource as much as I can my workflow from pharo to external libraries and
> apps that are way more mature and efficient for what I am trying to do.
>
> I always inteded to make Blender work with Pharo, I accomplished that now I
> want to make Unreal engine to work with pharo, so I can make a triangle of
> love, Blender for asset creation, Unreal for real time rendering of 2d and
> 3d graphics , Pharo as the brain of the logic and advanced scripting of the
> other 2. This is was my dream and goal all along. So I will be moving my GUI
> to Unreal that will allow me much more advanced and performance orientated
> features that mix 2d with 3d but I will keeping my main logic in pharo.
>
>
> On Sat, Jan 16, 2016 at 6:35 AM Ben Coman <b...@openinworld.com> wrote:
>>
>> On Sat, Jan 16, 2016 at 7:28 AM, Dimitris Chloupis
>> <kilon.al...@gmail.com> wrote:
>> > taskbarTask (self valueOfProperty: #noTaskbarTask ifAbsent: [ false ])
>> > ifTrue: [ ^ nil ]. ^ nil "TaskbarTask morph: self state: self
>> > taskbarState
>> > icon: self taskbarIcon label: self taskbarLabel"
>> >
>> > The uncommented part was the one that was slowing me down, its a copy
>> > from
>> > SystemWindow, on a new image of Pharo consumption drops to 15% but still
>> > have issues with Nautilus etc.
>> >
>> > latsabben at Slack also recommended caching which helped also
>> >
>> > taskbarTask "myTask := nil." myTask ifNil: [ myTask := TaskbarTask
>> > morph:
>> > self state: self taskbarState icon: self taskbarIcon label: self
>> > taskbarLabel ]. myTask label: self taskbarLabel. ^myTask
>> >
>> > Anyway I decided to port my project to C++ and Unreal Engine because I
>> > have
>> > many issues with Pharo speed wise
>>
>> See my other post, in about an hour I moved your App from 55% cpu
>> usage on my machine to 7%, only 1% above our 6% idle. We *do* need
>> to address that minimum idle, but its at the VM level since in-Image
>> profiling shows 90% time in ProcessorScheduler class>>idleProcess. So
>> to me the Image seems not too bad performance wise
>>
>> > and stability wise with Pharo 5.
>>
>> Well, you are talking about bleeding edge alpha software.
>>
>> > Plus many IDE features I miss like proper auto completion etc.
>>
>> You've probably mentioned this somewhere previously, but in another
>> thread could you leave us with a summary of what is missing from auto
>> completion.
>>
>> > To be fair I tried to make custom gui with python and it was even slower
>> > in
>> > the past.
>> >
>> > So its clear I need a high performance language + API, because I will be
>> > building a very heavy GUI (many more animations) and I would like also
>> > some
>> > fast 3d functionality too.
>>
>> good luck with it.
>> cheers -ben
>>
>>
>> >
>> > On Sat, Jan 16, 2016 at 1:07 AM Sven Van Caekenberghe <s...@stfx.eu>
>> > wrote:
>> >>
>> >>
>> >> > On 15 Jan 2016, at 23:30, Dimitris Chloupis <kilon.al...@gmail.com>
>> >> > wrote:
>> >> >
>> >> > taskbar was the problem, damn pharo gui is a huge pain in the hat.
>> >>
>> >> How so ?
>> >>
>> >> > On Fri, Jan 15, 2016 at 11:32 PM Dimitris Chloupis
>> >> > <kilon.al...@gmail.com> wrote:
>> >> > ITs not the step, I removed the step as I said in my first post.
>> >> > Still
>> >> > 30% cpu consumption
>> >> >
>> >> > The images are PNGs and RGBA , 8bit
>> >> >
>> >> >
>> >> > On Fri, Jan 15, 2016 at 10:54 PM Hilaire <hila...@drgeo.eu> wrote:
>> >> > It depends on what you are doing in a step, but 1s step should not
>> >> > hurt.
>> >> > May be the problem is somewhere else.
>> >> > With DrGeo, I noted Athens is faster to BitBlt with bitmap operations
>> >> > (in my case, only scaling and displaying a From in a DrGeo canvas).
>> >> > Also, do your bitmaps come with 32 bits depth?
>> >> >
>> >> > --
>> >> > Dr. Geo
>> >> > http://drgeo.eu
>> >> >
>> >> >
>> >> >
>> >>
>> >>
>> >
>>
>
