nice wrote:

Even a simple

MessageTally spyOn: [Transcript cr; show:'hello'].

exhibits a factor 2x slowdown between 3.8 and 3.10

Nicolas

And the main difference lies in the drawing of the container, not the contents as you can test it:

Smalltalk garbageCollect. theTranscriptWindow := SystemWindow allInstances detect: [:e | e label = 'Transcript']. "(I know, bad code, but cannot remember correct API)"

Time millisecondsToRun: [theTranscriptWindow drawDropShadowOn: World canvas]. => 4 36 Time millisecondsToRun: [World canvas drawMorph: theTranscriptWindow ]. => 4 21 Time millisecondsToRun: [theTranscriptWindow drawSubmorphsOn: World canvas]. => 18 11

Scores are given in 3.8 and 3.9/3.10 (where they are dependent on theTranscriptWindow bounds extent). Notice the transparency and background gradient in 3.9/3.10...

Don't know how this connects to number of opened Browsers seriously impacting speed, but maybe that's a clue...

Nicolas

Matthias Berth wrote:

...

Hello,

can anyone tell me were to find the earliest available 3.9 alpha image?

I want to identify the cause of the slowdown of browsers when you have many of them opened.

http://article.gmane.org/gmane.comp.lang.smalltalk.squeak.general/108359

I am using the benchmark from Andreas. So far I see the problem in the current squeak-dev 3.9.1, Squeak3.9b-7029.image. Browser speed is OK in Squeak3.8-6665.image (Squeak3.8-6665-basic.zip).

Cheers

Matthias

On Thu, May 01, 2008 at 09:04:52PM +0200, Matthias Berth wrote:

I've found Squeak3.9a-6678.zip on ftp://st.cs.uiuc.edu/pub/Smalltalk/Squeak/3.9alpha :)

Thank you for finding that.

karl wrote:

Matthias Berth wrote:

...

Hello,

can anyone tell me were to find the earliest available 3.9 alpha image?

I want to identify the cause of the slowdown of browsers when you have many of them opened.

Much of the slowdown is the pane resize handles from the look enhancement. There is arguably a bug in Morph>>privateAddMorph: that sometime adds submorphs to the top left corner of the screen instead of to the top left corner of the owner morph. This causes massive screen redraws. Also the adding of the pane resizes causes whole window redraws more frequently than before they were added.

Karl

Thanks, that's helpfull. This can be visualized with Preferences / debug / debugShowDamage

Nicolas

nice wrote:

karl wrote:

...

Matthias Berth wrote:

...

Hello,

can anyone tell me were to find the earliest available 3.9 alpha image?

I want to identify the cause of the slowdown of browsers when you have many of them opened.

Much of the slowdown is the pane resize handles from the look enhancement. There is arguably a bug in Morph>>privateAddMorph: that sometime adds submorphs to the top left corner of the screen instead of to the top left corner of the owner morph. This causes massive screen redraws. Also the adding of the pane resizes causes whole window redraws more frequently than before they were added.

Karl

But does that explain that Browser get redrawed even if obscured? Change the script a little to set the position near top left rather than the extent:

win := Browser openBrowser topView. "win extent: 600@350." win position: 10@10. br := win model. br setClassOrganizer. br systemCategoryListIndex: 2. World doOneCycle.

Transcript cr; show: [1 to: 10 do:[:i| br classListIndex: i. World doOneCycle. ]] timeToRun.

very linear in 3.10: 1611 2459 3288 4122 4976 5820

slower downgrading in 3.8: 1258 1308 1367 1427 1483 1552 1627

Nicolas

Here is a workaround to I made a while back. I'm not sure how much speedup you will get.

Karl