Raul,
Yes—I forgot. Modern browsers’ rendering engines are concurrent and will
recruit additional cores, so WebView is not entirely dependent on its “home”
core. Still, there is apparently something that the home core needs to do
before a page can be loaded. Below is a script that demonstrates the problem.
1) Evaluate starve’'
2) Resize the window that appears
3) Type in a URL and press enter. It should load in the right pane.
4) Select the “Draw Frames” check box. This turns on animation, which tries
(and fails, on my machine) to hit 100 fps. It’s doing a lot of work per frame;
the core is pegged.
5) Type in a URL and press enter. On my machine it did not load until
I—eventually—turned off “Draw Frames”.
This (and your comments) suggest to me that if I thread off all of the “data
wrangling” (which is considerable and could be a lot smarter) and just leave a
shared data structure that a relatively lightweight main-thread rendering
routine can turn into gl2 calls, that might fix it. Unfortunately, the only
way to know for sure is to build it.
Thank you.
Ed
load 'gl2'
coinsert 'jgl2'
buildForm =: 3 : 0
wd 'pc vizform;'
wd 'bin h;'
wd 'bin v;'
wd 'bin h;'
wd 'cc drawFrames checkbox;cn Draw Frames;'
wd 'cc urlBox edit;'
wd 'bin z;'
wd 'cc vocContext isigraph;'
wd 'bin z;'
wd 'cc browser webview;'
wd 'bin z;'
)
layoutForm =: 3 : 0
'w h' =. ". wd 'getp wh'
winW =. w - 20
winH =. h - 20
wd 'set drawFrames maxwh ' , (": (<. winW * 0.15) , 30) , ';'
wd 'set urlBox maxwh ' , (": (<. winW * 0.3) , 30) , ';'
wd 'set vocContext maxwh ' , (": (<. winW * 0.5) , winH) , ';'
wd 'set browser maxwh ' , (": (<. winW * 0.5) , winH) , ';'
)
vizform_resize =: 3 : 0
layoutForm ''
)
vizform_close =: 3 : 0
wd 'timer 0'
wd 'pclose'
)
sys_timer_z_ =: 3 : 0
wd 'set vocContext invalid'
)
vizform_urlBox_button =: 3 : 0
smoutput 'Loading' ; urlBox
wd 'set browser url *' , urlBox
)
vizform_drawFrames_button =: 3 : 0
if. drawFrames = '1' do.
wd 'timer 10'
else.
wd 'timer 0'
FrameTimeStamps =: ''
wd 'set vocContext invalid'
end.
)
vizform_vocContext_paint =: 3 : 0
glfill 255 255 255 255
k =. 1e7 ?@$ 1000
drawFrameRate ''
)
FrameTimeStamps =: ''
drawFrameRate =: 3 : 0
FrameTimeStamps =: (t =. (6!:1) '' ) , FrameTimeStamps
fps =. +/ (t - 1) < FrameTimeStamps
glfont 'arial bold 24'
glrgb 0 0 0
gltextcolor ''
gltextxy 10 10
gltext (": fps) , ' fps'
)
starve =: 3 : 0
buildForm ''
layoutForm ''
wd 'pshow'
wd 'set urlBox text "https://www.cnn.com";;'
)
> On Mar 22, 2023, at 10:25 PM, Raul Miller <rauldmil...@gmail.com> wrote:
>
> If only it were that simple.
>
> Sadly, Qt's design is such that all UI activity must happen on the main
> thread.
>
> https://doc.qt.io/qt-5/thread-basics.html
>
> Looking at how the system works, I believe that this was not the
> original plan, but instead was an organic response to some of the very
> reasonable expectations about how a user interface works.
>
> Here, it's perhaps also worth noting that a typical web browser, while
> it will use multiple threads, also uses multiple processes.
> Effectively, giving control over some part of screen real estate or
> some underlying form of infrastructure to an independent program. And,
> the webview control is, roughly speaking, a web browser.
>
> --
> Raul
>
> On Wed, Mar 22, 2023 at 6:36 PM Ed Gottsman <edward.j.gotts...@gmail.com>
> wrote:
>>
>> Hello.
>>
>> I've got a Window Driver application whose left side is an isigraph control
>> and whose right side is a webview control. The webview loads pages in
>> response to interaction with the isigraph.
>>
>> The isigraph side is quite snappy despite almost zero optimization effort (a
>> tribute to the developers of the J runtime, the gl2 library, etc.).
>>
>> The webview…is much less responsive, even when loading locally-cached HTML.
>> In fact, as long as the mouse is moving and frames are being generated in
>> the isigraph, the webview is unresponsive. Only when the mouse pauses (and
>> no frames are generated) does the webview load a page.
>>
>> I *think* that what's happening is that the two Window Driver child controls
>> share a core, and the isigraph side steals all of the available cycles to
>> service mouse-move events by generating frames, leaving no cycles for the
>> webview to parse and render HTML.
>>
>> With that in mind, it seems possible that I could improve the responsiveness
>> of the webview by threading off some or all of the (considerable) work that
>> goes into generating an isigraph frame, leaving most of the shared core's
>> cycles for the webview.
>>
>> I hesitate, however, before re-arranging the code because 1) I'm not sure
>> about the diagnosis and 2) I've worked in graphics environments where you're
>> not allowed to update GUI controls off the main thread and 3) maybe there's
>> a much simpler approach (can I somehow thread off the webview, e.g.?).
>>
>> Comments on any part of this would be greatly appreciated.
>>
>> Thank you.
>>
>> Ed
>> ----------------------------------------------------------------------
>> For information about J forums see http://www.jsoftware.com/forums.htm
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
