Squeak-dev May 2000

squeak-dev@lists.squeakfoundation.org

137 participants
680 discussions

Re: SimpleGeneticAlgorithm implementation
by Dean_Swan＠Mitel.COM 03 May '00

03 May '00

From: Dean Swan@MITEL on 04/25/2000 11:56 AM >Okay, I took a crack at it. The blocks can be removed without making >things messy. In general, I divided up SimpleGeneticAlgorithm into 3-4 >classes. Here are the particulars: Lex, Thanks. I'll try to take a look at what you've done over the next few days. >I toyed with it about 15 minutes, to try and duplicate the trick >in #fieldAt:width:, but ultimately I gave up. :| I'm sure someone >who knows BitBlt better could fix this right up and get performance >back up. That took me quite a while to get right, several hours, at least. The inspiration for this was taken from BitBlt>>bitPeekerFromForm:. I imagine the put could be done in a similar fashion, using BitBlt>>bitPokerToForm: as a model, and it is *very* worthwhile, because the speed difference for fields that are less than 30 bits wide is dramatic. As I said, I'll have to play around with what you've done. Speed is extremely important for genetic algorithms, since for non-trivial applications, you may have to run it for thousands or even millions of generations (assuming you start from a completely random population) to get a usable answer. >...several methods can get simpler--for example, swapChromosome:with: > is now just a couple of gets and a couple of sets, instead of 3 BitBlt > operations. I'm not to sure this is a good idea. Certainly, this is conceptually cleaner, but gets and sets ultimately end up sending fieldAt:width: and fieldAt:width:put:, which end up doing a lot more than three BitBlts, even if the puts are optimized in the same way that the gets are. It also involves more message sends, which are a major bottleneck for Squeak, because even my fastest Mac (my wife's iMac G3/333), and my fastest PC (PII/450) only crank a little over a million sends/second. ...And, BitBlt is fast. That's why I went with encoding the chromosomes in a form to begin with, in spite of my misgivings about the obtuseness of the implementation. I have to admit the need to reference the original instance of the SimpleGeneticAlgorithm has always bothered me. It's a kind of circular reference thing that just feels "messy". I'll see if I can do anything to get the speed back up, but my original whack at it was primarily focused on making the code small and fast, at the expense of transparency of design and readability. >Changing the block to a method didn't seem to hurt things, and >the non-local features of blocks turned out not to be needed. Keep in mind that the class comment demonstrates an absolutely trivial application. My real application does involve some other non-local references to objects that represent the search target, and they are *very* expensive to compute, so they would have to be made visible to the evaluator method by some means (class pools? - ick) because recomputing them for every reference is just out of the question. I still haven't found very good answers for my target problem, and I've let this thing crank for days on my fastest machines. Exchanges like this is one of the many reasons I'm so attracted to Squeak. 99% of the code I write never gets read by anybody but me. My employer only cares if it "meets the spec", and gets done on time. (They're not really that brash, but those really are the major things that count come annual review time). Actually, they also care about cost targets, so I suppose I get indirectly graded on target hardware design also. Such is life developing embedded firmware. It also leads to code which places performance and resource minimization at the top of the priority list, which often leads to less than totally transparent and obvious designs. I'm still inclined to think that blocks are one of the more powerful (and thus dangerous) constructs in Smalltalk, and I for one don't want to see them go away. Thanks Again! -Dean Swan dean_swan(a)mitel.com

2 1

RE: BitBlt >> #pixelAt:put: speed
by Raab, Andreas 03 May '00

03 May '00

Yes - use a cache form for drawing the points directly and pass the entire image to FormCanvas, ala: "Create some random points" rnd _ Random new. points _ (1 to: 5000) collect:[:i| (rnd next @ rnd next * 400) asIntegerPoint]. "Make us a canvas" fc _ FormCanvas on: Display. "Create a cached form for drawing all these points" cacheForm _ Form extent: 400@400 depth: Display depth. "Setup BB" bb _ BitBlt toForm: cacheForm. bb combinationRule: Form over. bb width: 1; height: 1. bb fillColor: Color red. "Now draw these points offscreen" points do:[:pt| bb destOrigin: pt; copyBits]. "And finally paint the entire image" fc paintImage: cacheForm at: 0@0. This should be a lot faster than anything else. - Andreas -----Original Message----- From: Gerardo Richarte [mailto:core.lists.squeak@core-sdi.com] Sent: Tuesday, May 02, 2000 6:20 PM To: squeak(a)cs.uiuc.edu Cc: recipient list not shown Subject: Re: BitBlt >> #pixelAt:put: speed Andres Valloud wrote: > > Is there anything faster I can do? for example using Balloon somehow? > > Are you trying to do that directly on the World's canvas? Did you try to > draw them in a buffer and then draw the buffer form? Eh! Andres! (que haces? como va todo por ahi?) I'm not sure what am I doing, I'm just drawing on the canvas I get as argument in Morph >> #drawOn: So you think I should use an intermediate Canvas for this? Thanks! Liencistas Bye! (just for spanish speaking people, sorry) Richie++ -- A390 1BBA 2C58 D679 5A71 - 86F9 404F 4B53 3944 C2D0 Investigacion y Desarrollo - CoreLabs - Core SDI http://www.core-sdi.com <http://www.core-sdi.com> --- For a personal reply use gera(a)core-sdi.com

2 1

FW: FW: Time to make a Freshmeat kind of Squeak site ?
by Stephen Pair 03 May '00

03 May '00

Lex Spoon wrote: > Editors, yes. But why does it need to be controlled? IMHO, let's just > try it with volunteer editors (whoever first goes to the Swiki and > starts organizing things), and then see if we run into any serious > disagreements. > > One thing that Wiki and Swikis have shown, is that leaving control wide > open can really work in a lot of circumstances. In many groups--even in > classrooms where students are competing with each other--people just > don't screw up each others' work. But doesn't that in fact contribute to the problem?...people don't feel that they should significantly re-organize other people's contributions. Therefore, we get many names for the same thing (goodies, tools, packages), things scattered...important things buried and not so important things in an inappropriate place, inconsistent verbage and navigation, etc. As for the issue of being controlled, I think it's needed for the same reason that www.squeak.org is controlled. Which begs the question, why is it controlled? Perhaps some other people could give some good reasons for this. I think there is a need for both squeak.org and a wide open Swiki. And, I think what people are arguing is the squeak.org (and possible the Swiki) should be more friendly to the newbie as well as better organized for the expert. - Stephen

1 0

[BUG]Polygon fill color
by Karl Ramberg 03 May '00

03 May '00

When you draw in PaintBoxMorph and use the Star or the Polygon tools they fill the wrong nuance of color. Bug seems to be that the PolygonMorph uses a different mode when it draws the color. Karl

1 0

[BUG][FIX]Halo Menu
by Karl Ramberg 03 May '00

03 May '00

The halo menu was displaying debug... two times. I commented out a line to fix this. Small artifact: When in Novice mode the halo menu has a line at the bottom :-( Could not find that line in the mess ;-) Karl 'From Squeak2.8alpha of 19 February 2000 [latest update: #2052] on 3 May 2000 at 2:47:07 pm'! !Morph methodsFor: 'menus' stamp: 'kfr 5/3/2000 14:22'! addCustomMenuItems: aCustomMenu hand: aHandMorph "Add morph-specific items to the given menu which was invoked by the given hand." aCustomMenu add: 'add drop-shadow' action: #addDropShadow. (self isKindOf: SystemWindow) ifFalse: [aCustomMenu add: 'put in a window' action: #embedInWindow]. aCustomMenu addUpdating: #stickinessString target: self action: #toggleStickiness. aCustomMenu add: 'adhere to edge...' action: #adhereToEdge. "Preferences noviceMode ifFalse: [self addDebuggingItemsTo: aCustomMenu hand: aHandMorph]." ! !

1 0

[Question]Converting colorForms from 32/16 to 8
by Karl Ramberg 03 May '00

03 May '00

How do I convert a colorForm from 32/16 bit color depth to 8 bit color depth ? Karl

1 0

[ENH]Removed items from NewMorph menu
by Karl Ramberg 03 May '00

03 May '00

Items that caused a direct debugger. FlapTab MethodMorph StringMorphEditor TextMorphForEditView ViewerFlapTab Karl 'From Squeak2.8alpha of 19 February 2000 [latest update: #2052] on 3 May 2000 at 12:56:47 pm'! !FlapTab class methodsFor: 'as yet unclassified' stamp: 'kfr 5/3/2000 12:51'! includeInNewMorphMenu "Not to be instantiated from the menu" ^ false! ! !MethodMorph class methodsFor: 'as yet unclassified' stamp: 'kfr 5/3/2000 12:52'! includeInNewMorphMenu "Not to be instantiated from the menu" ^ false! ! !StringMorphEditor class methodsFor: 'as yet unclassified' stamp: 'kfr 5/1/2000 13:41'! includeInNewMorphMenu "Not to be instantiated from the menu" ^ false! ! !TextMorphForEditView class methodsFor: 'as yet unclassified' stamp: 'kfr 5/1/2000 13:41'! includeInNewMorphMenu "Not to be instantiated from the menu" ^ false! ! !ViewerFlapTab class methodsFor: 'as yet unclassified' stamp: 'kfr 5/3/2000 12:52'! includeInNewMorphMenu "Not to be instantiated from the menu" ^ false! !

1 0

[ENH]SystemWindow SimpleButtonMorph
by Karl Ramberg 03 May '00

03 May '00

Here is the most recent and probably best functioning change set for the SystemWindow and SimpleButtonMorph enhancements I've done. Warning !! The enhancement to the SystemWindow is not backwards compatible. The enhancement to SimpleButtonMorph is a way to send update info to submorphs while the button is pressed in the way of #actWhilePressed. Probably not the best name but it does the job :-) Karl 'From Squeak2.8alpha of 19 February 2000 [latest update: #2052] on 30 April 2000 at 3:14:35 pm'! RectangleMorph subclass: #SimpleButtonMorph instanceVariableNames: 'target actionSelector arguments actWhen actWhilePressed oldColor ' classVariableNames: '' poolDictionaries: '' category: 'Morphic-Widgets'! MorphicModel subclass: #SystemWindow instanceVariableNames: 'labelString stripes label closeBox collapseBox activeOnlyOnTop paneMorphs paneRects collapsedFrame fullFrame isCollapsed menuBox mustNotClose labelWidgetAllowance updatablePanes allowReframeHandles collapseBoxSubMorph ' classVariableNames: 'TopWindow ' poolDictionaries: '' category: 'Morphic-Windows'! !SimpleButtonMorph methodsFor: 'initialization' stamp: 'kfr 4/29/2000 17:04'! initialize super initialize. self borderWidth: 1. self cornerStyle: #rounded. self color: (Color r: 0.4 g: 0.8 b: 0.6). self borderColor: self color darker. target _ nil. actionSelector _ #flash. arguments _ EmptyArray. actWhen _ #buttonUp. actWhilePressed _ nil. self setDefaultLabel ! ! !SimpleButtonMorph methodsFor: 'accessing' stamp: 'kfr 4/29/2000 17:03'! actWhilePressed ^ actWhilePressed ! ! !SimpleButtonMorph methodsFor: 'accessing' stamp: 'kfr 4/29/2000 17:03'! actWhilePressed: aSymbolOrString (nil = aSymbolOrString or: ['nil' = aSymbolOrString or: [aSymbolOrString isEmpty]]) ifTrue: [^ actWhilePressed _ nil]. actWhilePressed _ aSymbolOrString asSymbol. ! ! !SimpleButtonMorph methodsFor: 'events' stamp: 'kfr 4/29/2000 17:04'! mouseMove: evt actWhen == #buttonDown ifTrue: [^ self]. (self containsPoint: evt cursorPoint) ifTrue: [oldColor ifNotNil: [self color: (oldColor mixed: 1/2 with: Color white)]. (actWhilePressed ~~ nil) ifTrue: [target perform: actWhilePressed ]. (actWhen == #whilePressed and: [evt anyButtonPressed]) ifTrue: [self doButtonAction. evt hand noteSignificantEvent: evt]] ifFalse: [oldColor ifNotNil: [self color: oldColor]. (actWhilePressed ~~ nil) ifTrue: [target perform: actWhilePressed ]] ! ! !SystemWindow methodsFor: 'initialization' stamp: 'kfr 4/30/2000 15:10'! addCloseBox self addMorph: (closeBox _ SimpleButtonMorph new borderWidth: 1;borderColor: Color black; cornerStyle: #square; label: '' font: Preferences standardButtonFont; color: Color transparent; actionSelector: #delete; target: self; extent: 10@10). closeBox addMorph: (StringMorph new contents: 'x'; font:(StrikeFont familyName: #ComicBold size: 16); position: 2 @-5; color: Color black)! ! !SystemWindow methodsFor: 'initialization' stamp: 'kfr 4/29/2000 17:06'! addCollapseBox self addMorph: (collapseBox _ SimpleButtonMorph new borderWidth: 0; label: '' font: Preferences standardButtonFont; cornerStyle: #square; color: Color transparent ;actWhilePressed: #subMorphAction; actionSelector: #collapseOrExpand; target: self; extent: 10@10). collapseBox addMorph:(RectangleMorph new extent: 6 @ 6; position: 0 @ 0; borderWidth: 1; borderColor: Color black; color: Color transparent). collapseBox addMorph:( RectangleMorph new extent: 7 @ 7;position: 3 @ 3; borderWidth: 1; borderColor: Color black; color: Color transparent). ! ! !SystemWindow methodsFor: 'initialization' stamp: 'kfr 4/29/2000 16:54'! initialize super initialize. allowReframeHandles := true. labelString ifNil: [labelString _ 'Untitled Window']. isCollapsed _ false. activeOnlyOnTop _ true. paneMorphs _ Array new. paneRects _ Array new. borderColor _ #raised. borderWidth _ 1. color _ Color black. stripes _ Array with: (RectangleMorph newBounds: bounds) "see extent:" with: (RectangleMorph newBounds: bounds). self addMorph: (stripes first borderWidth: 1). self addMorph: (stripes second borderWidth: 2). self addMorph: (label _ StringMorph new contents: labelString; font: Preferences windowTitleFont emphasis: 1). self setLabelWidgetAllowance. self addCloseBox. self addMenuControl. self addCollapseBox. Preferences noviceMode ifTrue: [closeBox ifNotNil: [closeBox setBalloonText: 'close window']. menuBox ifNotNil: [menuBox setBalloonText: 'window menu']. collapseBox ifNotNil: [collapseBox setBalloonText: 'collapse/expand window']]. self on: #mouseEnter send: #spawnReframeHandle: to: self. self on: #mouseLeave send: #spawnReframeHandle: to: self. label on: #mouseDown send: #relabelEvent: to: self. self extent: 300@200. mustNotClose _ false. updatablePanes _ Array new.! ! !SystemWindow methodsFor: 'geometry' stamp: 'kfr 4/29/2000 17:06'! extent: newExtent | inner labelRect | isCollapsed ifTrue: [super extent: newExtent x @ (self labelHeight + 2)] ifFalse: [super extent: newExtent]. inner _ self innerBounds. labelRect _ self labelRect. stripes first bounds: (labelRect insetBy: 1). stripes second bounds: (labelRect insetBy: 3). self setStripeColorsFrom: self paneColorToUse. closeBox ifNotNil: [closeBox align: closeBox topLeft with: inner topLeft + (4@3)]. menuBox ifNotNil: [menuBox align: menuBox topLeft with: (inner topLeft + (19@1))]. collapseBox align: collapseBox topRight with: inner topRight - (4@3 negated). collapseBox submorphs first color: self paneColor. collapseBox submorphs second color: self paneColor. label fitContents; setWidth: (label width min: bounds width - self labelWidgetAllowance). label align: label bounds topCenter with: inner topCenter. isCollapsed ifFalse: [self setBoundsOfPaneMorphs]. ! ! !SystemWindow methodsFor: 'resize/collapse' stamp: 'kfr 4/29/2000 16:45'! subMorphAction (collapseBox containsPoint: self cursorPoint) ifTrue:[ collapseBox submorphs first color: Color gray. collapseBox submorphs second color: Color gray. collapseBox color: Color transparent] ifFalse:[ collapseBox submorphs first color: self paneColor. collapseBox submorphs second color: self paneColor] ! ! MorphicModel subclass: #SystemWindow instanceVariableNames: 'labelString stripes label closeBox collapseBox activeOnlyOnTop paneMorphs paneRects collapsedFrame fullFrame isCollapsed menuBox mustNotClose labelWidgetAllowance updatablePanes allowReframeHandles ' classVariableNames: 'TopWindow ' poolDictionaries: '' category: 'Morphic-Windows'!

1 0

Time to make a Freshmeat kind of Squeak site ?
by Karl Ramberg 03 May '00

03 May '00

I'v been following the recent discusions here on the list and there seem to be several parralell projects and unknown projects and different software out there that does a great job in someones image but they newer get out because 1 Nobody know they exitst 2 Nobody now where to look 3 Nobody know how to use them What I sugeste is a Freshmeat ( freshmeat.org ) kind of site on a swiki and some simple rules to how to categorize/describe what the stuff does and then try to get as many as possible to put theyr struff there. Yes I know of the [GOODIE] and [PACKAGE]and stuff but it was a mess last time I peeked, not very usefull. Maybe it's enough to clean up there and categorize and encourage people to submit there. Another thing from Freshmeat is the comments people put in that tells their opinion on the program, easy on the swiki. Karl

3 3

Re: Carrying preferences around
by Roel Wuyts 03 May '00

03 May '00

One of our researchers at the lab (Tom Tourwe) does the following: He changed the methods Preferences class>>#initialValuesA" en Preferences class>>#standardInitialValues, and made a changeset of this change. Filing in this changeset updates his preferences. While this is a crude system, it helps our development, since each week everybody starts out with a fresh image containing the latest changes from everybody. This could help while a better system is under way. PS: There seems to be a problem with this approach in the sense that some preferences are lost after file-in. This indicates that other places update the preferences. Tom's approach did not take that into account. We should think of a real singleton user factory that holds these dispersed settings, and that we can just serialize to disk. >> From: "Lex Spoon" <lex(a)cc.gatech.edu> >> >> Hrm, I would rather preferences be attached to an image, though. For >> instance, I have development images, server images, and an occasional >> screwball image where I'm playing with something really wierd. Each >> image is its own little world, with its own preference settings. > Such a >> system is incompatible with global preferences stored in files. > > The OpenStep Defaults system, for example, allows hierarchical > specification of defaults, very similar to preferences. There is a > global level which is overridden by a per-application level which is > overridden by a per-process level taken from the command-line. The > first two levels are stored in a per-use defaults database. > > A Squeak per-image level would have to be inserted between the > document and process levels as a "document" level. > > Marcel > > -- Roel Wuyts Programming Technology Lab rwuyts(a)vub.ac.be Vrije Universiteit Brussel http://prog.vub.ac.be/~rwuyts Webmaster of European Smalltalk User Group: www.esug.org

1 0

← Newer
1
...
60
61
62
63
64
65
66
67
68
Older →

Jump to page: