Squeak-dev October 2001

squeak-dev@lists.squeakfoundation.org

191 participants
600 discussions

[ENH][SF] VMMaker platforms tree now imported to SourceForge.net
by Tim Rowledge 31 Oct '01

31 Oct '01

I've just imported the platforms tree as suited to VMMaker (see http://sumeru.stanford.edu/tim/pooters/SqFiles/deltas and look for the latest VMMaker* file(s) ) to SourceForge (see http://sourceforge.net/projects/squeak) in the hope it might actually work as a way to let people try it out. Ghu only knows if I managed to do it right; I'm sure a gazillion people will tell me all about it if not. Amongst other things, I have no idea how binary files will have been treated. Another possible problem is that in the process of editing the CVSROOT/modules file to ad a 'platforms' module, SF appeared to lock up on me. Has this left it in a dead state? Am I about to be the first squeaker to be arrested by the FBI for terrorist hacking because of this? Can anybody else even see the files? (should be under 'squeak/platforms' at the moment) tim -- Tim Rowledge, tim(a)sumeru.stanford.edu, http://sumeru.stanford.edu/tim Strange OpCodes: BOZO: Use Multics operating system

7 16

Re: [Celeste] msgIDs and sharing categories
by Lex Spoon 31 Oct '01

31 Oct '01

Celeste progress is gridlocked right now. In no particular order: Message ID's are pretty reliable. The RFC says they "should" be present. Has anyone MD5 is a risky approach, because inserting one random space character will invalidate the MD5 completely. MD5's by design have no notion of closeness. Celeste can certainly access message-id's; you just have to read the message in and then look for the header. Something like: (mailDB getMessage: id) fieldNamed: 'message-id' The above is currently expensive, but, it could be much faster if the index file format were updated, so that message-id's could be saved. With an improved index file format, I don't think you'd care that Celeste is using its own message id's. If the index file is updated, then it also becomes much easier to implement a "leave messages on server" so that it won't download messages multiple times. There have been proposals for updating the index file format, but they are waiting on Filtering Celeste to settle down. Filtering Celeste is fairly well settled, but it can't go into the image until LargeLists does. If LargeLists doesn't, then FilteringCeleste is impractical and needs some redesign. There is dead silence regarding LargeLists, so I'm just a waiting. Probably, people could actually go ahead and work on the index file format if they want to, since it's fairly independent from filtering celeste. But, this trend cannot go on too long -- eventually we need to get all our patches together, or Celeste will disintegrate in a mass of good intentions. It's not inconceivable that modules will arrive before the next Celeste patch gets in the main image. At any rate, whenever Squeak has a good module system, we'll probably at least *try* to offload Celeste (and LargeLists, if necessary) into a separate system for a while. Phew. -Lex

1 0

RE: Lots of concurrency
by G.J.Tielemans＠dinkel.utwente.nl 31 Oct '01

31 Oct '01

Children and students live in this era: - They can send each other emails, it is possible to send eamils to a list - They can send each other SMS, it is possible to send an sms to a group - They can chat in a chatbox, private or in a group > -----Original Message----- > From: Ken Kahn [mailto:kenkahn@toontalk.com] > Sent: dinsdag 30 oktober 2001 20:33 > To: squeak-dev(a)lists.squeakfoundation.org > Subject: Re: Lots of concurrency > > > Richard A. O'Keefe wrote: > > > > > The question I think is interesting is whether telling students to > > think in terms of objects TALKING to each other makes it harder for > > them to think of concurrent implementations. Would some other > > metaphor (perhaps sending couriers with messages, or thinking about > > a factory with things concurrently moving from machine to machine > > at the same time) make it easier for them to think of and understand > > concurrency? > > > > Before settling upon a carrier pidgeon metaphor for > communcation in ToonTalk > I considered lots of other alternatives including a postal > system with mail > carriers and mail boxes, rivers where you can place floating > messages to > those downstream, wires, fax machines, email, and telephones. I never > considered face-to-face talking since I took for granted that > objects are > spread out spatially. I think the answer to Richard's > question is that the > right choice of metaphor can significantly facilitate thinking and > understanding concurrent programs. > > Wires may be how messages are really passed within and > between computers but > wires are not an ideal metaphor. The messages are invisible. > Wires cannot be > sent along wires. Directionality isn't apparent. Wireless > communication is > even worse. > > Talking or telephones aren't good because both parties need > to be available > at the same time. Many-to-1 communication by telephone is > confusing. While > phone numbers can be communicated over the phone that would > lead to a model > where numbers denote objects or communication channels. So > can they be made > up and dialed at random? Not good. > > Even the postal system doesn't provide an ideal metaphor. If > you want to > send the receiving end of a communication channel to someone > do you have to > pick up your mail box and send it by mail to someone? And > then does the > postal system know that the mailbox has moved or do you need > to think about > forwarding addresses? > > A bird that when given something flies to her nest and leaves > it there and > returns works out very well. You can give a bird a box that > might contain > other birds and nests - not too strange. The only strange > thing is that a > bird always finds her nest no matter where it has been moved > to. But that > doesn't seem to cause any confusion. > > > All I know about Ken Kahn's ToonTalk is what I've read in > this thread, > > but it sounds as though uses a "physical" rather than "verbal" > > metaphor, so I think his observations are particularly > interesting here. > > > > You all are welcome to try out ToonTalk. There is a free > trial version and > you are all welcome to try to beta version as well. It only > runs on PC with > Windows (or Macs emulating a PC with Windows - don't know about Linux > emulators). Details at www.toontalk.com > > Best, > > -ken kahn > > > >

3 2

Re: [Modules] DynaBook versus DynaProject
by Richard A. O'Keefe 31 Oct '01

31 Oct '01

Mark van Gulik <ghoul6(a)home.com> wrote: I love to nitpick, so I'd like to point out that you picked two (that I know of) bad examples out of four. The Bible is a collection of unrelated letters, books, songs, etc. (if it were "assembled" today it would probably include tunes from TV jingles, soup can labels, and random phone book pages). The "biblia" (plural) for which the Bible is named are not unrelated. Having been written before the invention of chapter and verse numbering, they couldn't include proper XPointers, but later books are riddled with quotations from and allusions to earlier ones. (One can also say the same for the Talmud and the Church Fathers, of course.) One way in which "ta biblia" make a good example for modularisation is precisely this strong one-way linkage. You _can_ carve the collection up (at chronological "joints"), but the pieces cannot all stand alone. For example, if you chop the collection into the Torah-Nabiim-Kethuvim (spelling?) on one side and the Christian writings on the other, the Old Testament holds together, but the New Testament becomes largely unintelligble. Within the Tanach, the Prophets depend on the Torah. In terms we've seen in this mailing list, I suppose you could say that the New Testament is a DeltaModule. Lord of the Rings is also a series of books, if I recall correctly (I don't know whether J.R.R.T. intended this, or if it was a publishing "artifact"). The first copy I ever had was a single volume. The division of "works" into "books" or "volumes" goes back to classical times: there was only so much you could fit on one scroll. In the case of TLOTR, the physical divisions do no violence to the logical structure of the book, but a hierarchical structure (work, book, chapter, paragraph, sentence) does not a modular structure mark. Try the first book on its own: Here's a story, how does it end? Try the last book on its own: Who _are_ these characters? You might as well call an ANSI Pascal program "modular" because it's structured.

1 0

[ANN][SqueakNews]October issue is out!
by Editor - Squeak News 31 Oct '01

31 Oct '01

Hello, fellow Squeakers, It is my pleasure to announce that the October issue is finally ready to get out of the door. We should complete sending the CDs to our subscribers by Monday. It took quite a while but this issue has the record number of active essays in it. Many of these are collected from teachers. Some are used as tutorials for students and some are genuine work of students using Squeak. This issue will be a great tool to introduce Squeak to schools. Alan Kay's "Stronger Goals for Education" coupled with articles by 3 teachers who use Squeak actively in schools to teach Squeak to children, and an interview with Kim Rose state the case for Squeak in schools as never before. These are supported by Alan's older articles and a host of examples developed by students ranging from elementary to high school. Please contact me if you are interested in using this issue of Squeak News as a tool to introduce Squeak to schools. I guess I don't need to reiterate that Squeak News is a monthly e-zine is a multi-media application which is 100% pure Squeak right from the opening screen. (See the main screen of the October issue at: http://www.squeaknews.com/current/sncoveroctsml.jpg ) It runs on Windows, Mac and Linux and besides being a lot of fun, it includes dynamic content ranging from active essays and demos by Squeak gurus to masterpieces by school kids. Here is the summary for the October issue: Volume 1, Issue 4. 15 October 2001 Powerful Ideas in the Classroom This month's emphasis is classroom and how Squeak is used by students and teachers in schools. We have examples created by students, articles written by teachers, and comments of students about their Squeak experience. There is a great interview we made with Kim Rose, executive director of the new home of Squeak Central: Viewpoints Research Institute, co-editor of the book "Squeak, Open Personal Computing and Media", and a long standing member of Squeak Central. We are very fortunate to be able to publish Alan Kay's "Stronger Goals for Education" which is a must read for anybody who has an interest in education. Alan was kind enough to permit us to publish many of his past articles and active essays as well and and we will publish them in this and the next issue. We also feature an article by Michael Rueger titled 'Plugging in Squeak' which is a great insight to Squeak's usage as a plug-in. Our beginners series continues with the article titled: "From objects to classes". It introduces classes and how they are used, and other Squeak basics such as blocks. As usual the October issue includes the monthly digest of the Squeak mailing list and Bijan Parsia's Squeak Tweaks column. This month Bijan revisits Quickie, a HTML editor for Squeak which is quickly reaching to the level of replacing other HTML editors! John McIntosh has again generously allowed us to use his OOPSLA reports and we took the privilege of combining them with Michael Rueger's pictures to have you relive the OOPSLA experience. From this month on we are ditching our monthly quiz in favor of Squeak's Most Wanted list which we will maintain on an SWiki. You can reach the table of contents of the current issue by following the link: http://www.squeaknews.com/current/index.html and view the past issues of Squeak News by following: http://www.squeaknews.com/archive/index.html Happy Squeaking! Tansel Ersavas Editor, Squeak News http://www.squeaknews.com

3 2

Computer Music Implementation Class
by Mark Guzdial 31 Oct '01

31 Oct '01

At the Squeak BOF at OOPSLA, I was asked to re-send the URL for the Computer Music Implementation class that I've been teaching with Squeak and Active Essays. The class Swiki is at http://swiki.cc.gatech.edu:8080/compMusic The Active Essays are linked off the front page. You might also take a look at some of the students' turnins and project presentations. There are some really interesting projects there: - A couple interesting instruments, e.g., a nice trombone slide, a tuba. - Some neat synthesis algorithms: Waveshaping (couple), granular synthesis. - Some nice effects implementations, such as a flanger effect. Mark -------------------------- Mark Guzdial : Georgia Tech : College of Computing : Atlanta, GA 30332-0280 Associate Professor - Learning Sciences & Technologies. Collaborative Software Lab - http://coweb.cc.gatech.edu/csl/ (404) 894-5618 : Fax (404) 894-0673 : guzdial(a)cc.gatech.edu http://www.cc.gatech.edu/gvu/people/Faculty/Mark.Guzdial.html

1 0

[updates] 20 more for 3.2alpha
by Dan Ingalls 30 Oct '01

30 Oct '01

Folks - Herewith the summaries of 20 recent updates just posted to the servers. - Dan ----------------------- 4442LogError2-tk -- Ted Kaehler -- 19 October 2001 Limit the length of each object printed in an error report." 4443stringChangedFix-nk -- Ned Konz -- 12 October 2001 This fixes an error in my prior attempt. In Change Set 4393StringMorphOptimize-nk I missed a changed message in the case where the StringMorph doesn't change size. This fixes that problem and still avoids sending multiple changed messages for a single change." 4444thumbnailScaling-nk -- Ned Konz -- 18 October 2001 The thumbnails used in the PartsBins (including in the Objects tool) were being stretched vertically when they were too long. You can see this in the StringMorph and Clock thumbnails. This fixes the scaling so that the aspect ratio is maintained. The thumbnail cache is cleared, so it may take a few seconds to see some of the objects tool pages the first time." 4445newMorph-sw -- Scott Wallace -- 20 October 2001 Restores the 'from alphabetical list...' option to the non-classic variant of the 'new morph' menu. Provides info such that the ScorePlayer appears in the Objects tool." 4446FuzzyFlood-di -- Dan Ingalls -- 18 October 2001 Introduces a new variant of the bitmap flood algorithm, designed to allow for minor variations of color. This allows one to get reasonable behavior in, eg, images reconstructed from .jpeg files, and images taken from real life. Two preferences, #areaFillsAreTolerant and #areaFillsAreVeryTolerant, govern the use of this feature. They affect both flooding in the SketchEditor and grabbing the screen by flood area. I'm proud of this feature because it does not affect the flooding algorithm or its performance at all. It simply makes the maze follower 'color blind' to minor variations in color by introducing a tweaked color map. Another feature is also added -- the ability to add a border to a sketchMroph. This is located in the 'painting...' option of the SketchMorph menu. It should be generalized in color but, hey, I've done the hard work ;-). 4447numberType-sw -- Scott Wallace -- 23 October 2001 Fleshes out some details of the Number vocabulary by providing accurate result types and argument types for the method-interrfaces. The benefits will show up immediately in a viewer open on a number -- try, for example, evaluating '123 beViewed'" 4448sampleImageForm-sw -- Scott Wallace -- 24 October 2001 Allows Morph classes to provide their own sample images for the benefit of thumbnail creation; this allow us to avoid launching sample instances only for the benefit of obtaining thumbnails when the very launching of those sample instances may have consequences we do not want, such as initializing the network or asking the user an unexpected question. The capabilitiy is illustrated by the AudioChatGUI morph (where we wish to avoid actually initiating a session just to create an icon!), as well as by two StarSqueak examples, where the appearance of the initial condition of a simulation is much less appropriate than the appearance after the simulation has run for some time." 4449zipToolInObjectsTool-nk -- Ned Konz -- 24 October 2001 This adds the Archive Viewer to the Tools and Useful pages of the Object Tool as 'Zip Tool'. It fixes the problem of opening an extra window that my prior change set had. This fix was contributed by Darryl Smith. It also fixes the pane color in the AlternativeWindowLook, which was annoyingly translucent." 4450EtoyRenameFix-svp -- Stephen Pair -- 14 October 2001 Fixes the bug introduced in update 4361UniTile10 that brought about a halt when attempting to rename a script associated with a PasteUpMorph. BugFixing Party @ Camp Smalltalk OOPSLA 2001. The Oct 11, 2001 bug report was: 1. Drag a holder out of the 'Supplies' bin 2. Open a viewer for the holder 3. Create a new script (by dragging out the 'emptyScript' tile) 4. Change the name of the script to anything 5. Hit return, and get a walkback. MessageNotUnderstood: cards" 4451dropGz-sw -- Scott Wallace -- 26 October 2001 When you drop a .gz file onto the icon of a running Squeak application, the contents of it will now be displayed in decompressed form. A more general and modular mechanism is needed to open the right kind of tool upon a drop, but in the interim this provides the convenience that we can quickly see the decompressed contents of .gz files that come flying into our lives via email through direct dNd from the OS, or indeed right from a dNd-friendly email client such as Eudora" 4452EndianStartup-tk -- Ted Kaehler -- 24 October 2001 When an export image segment is brought into an image, it is like an image starting up. Certain startUp messages need to be run. These are byte and word reversals for nonPointer data that comes from a machine of the opposite endianness. We pass over all objects in the segment, and: The first time an instance of class X is encountered, (msg := X startUpFrom: anImageSegment) is sent. If msg is nil, the usual case, it means that instances of X do not need special work. X is included in the IdentitySet, noStartUpNeeded. If msg is not nil, store it in the dictionary startUps which contains (class -> messageSend). When a later instance of X is encountered, if X is in noStartUpNeeded, do nothing. If X is in startUps, send the message to the instance. Typically this is a message like #swapShortObjects. Every class that implements #startUp, should see if it needs a parallel implementation of #startUpFrom:. (Data in 4-byte chunks is handled automatically by the image loader and by fileStream for ImageSegment loading. startUpFrom: is only for exceptions.) Added startUp for class SoundBuffer." 4453ReducePalette-di -- Dan Ingalls -- 17 October 2001 Adds an option to 'reduce color palette' in the SketchMorph 'paint...' menu. In this manner images may be compressed much better, and will work much better with the paint can and select-by-flood tools. Note: this has not been engineered. There are much fancier solutions in the literature. 4454JPEGimageDepth-di -- Dan Ingalls -- 28 October 2001 The method nextImageDitheredToDepth: in JPEGReadWriter is changed to produce a form of the depth supplied (was always 32). At the sytem level, this means that JPEGs produce forms with a depth equal to that of the Squeak display." 4455PastEndPutFix-di -- Dan Ingalls -- 28 October 2001 An optimization to WriteStream nextPutAll: had defeated the operation of LimitedWriteStream. This changeSet suppresses the optimization only when the pastEndPut: logic should be triggered. It also factors pastEndPut: in LimitedWriteStream so that it calls super. Thus besides restoring the function of LimitedWriteStreams, the change consolidates three independent methods for handling stream overflow. 4456AllObjectsSelect-di -- Dan Ingalls -- 28 October 2001 Adds Smalltalk allObjectsSelect: as companion to allObjectsDo:. 4457LowSpaceFix-hg-di -- Dan Ingalls (from Henrik Gedenryd) -- 28 October 2001 Fixes a but that interfered with Low Space notification. 4458JPEGMoreCompact -- Andreas Raab -- 28 October 2001 Reduce the memory requirements for the jpeg reader." 4459CurveTextClipFix-di -- Dan Ingalls -- 28 October 2001 Repairs a bug introduced as part of the conversion of TextMorphs to being bordered. Requires bounds to be sent as a message, esp in innerBounds. 4460keepTextFont-nk -- Ned Konz -- 20 October 2001 When you use TextMorphs from the EToy system and you change their contents using the characters or numericValue slots, font changes get lost. This change set makes it so that if a TextMorph is all the same font, that font is used when the contents change, thereby keeping the appearance of the TextMorph uniform. To use this, just select the entire Text of a TextMorph and apply font color, emphasis, or size changes as desired to the whole text. Then you can change the characters or numericValue slots from EToys without changing the TextMorph's appearance." 4461scriptingCategory-sw -- Scott Wallace -- 29 October 2001 -- Manually Patched sw 10/29/2001 06:07 The 'miscellaneous' category was getting unwieldy in the etoy viewer, so the items that related to starting and stopping scripts are now moved off to their own 'scripting' category. (Caution: order is essential in this fileout)"

3 2

Re: Debian and SqueakL revisited again...(was Re: Debian source package)
by goran.hultgren＠bluefish.se 30 Oct '01

30 Oct '01

Hi! Andrew - you don't have to continue this thread if you are "sick of it" - I know that you have been answering questions like these for a long time on the list. :-) Still, I haven't really gotten "LGPL/GPL vs Squeak-L" fully clear in my head and perhaps there are others out there with a similar confusion. Thanks for your answers sofar! "Andrew C. Greenberg" <werdna(a)mucow.com> wrote: > > But the module system will also probably fulfill the role of something > > like CPAN where people publish their apps and libraries not necessarily > > meaning that they are defined as being "base Squeak" modules. And you > > should be able to publish something that you have built without being > > forced to use Squeak-L2, right? :-) > > Except in particular cases, you have never been "forced to use" any > license when deriving work from other people's IP, particularly in > Squeak. I see no reason, however, to encourage use of anything other > than Squeak-L, modular or otherwise. Ultimately, mixed licensing will > likely corrupt a monolithic image and make it either unusable or > unpublishable and that is, IMHO, a bad thing. Ok, maybe we are misunderstanding each other. I am not talking about what currently is in the image - which probably will turn into a defined subset of all the published modules in the module system. We will probably say "These modules are in the officially blessed Squeak base" or something similar, and yes, those modules should all be under one license. and yes, the modules system will open up the possibilities of more "controlled forks" but most of us are aware of the sociological (?) forces at work in an OpenSource community - forks are pretty unlikely, and if they occur successfully then they are probably allright. My guess is that the "kernel" of Squeak will stay focused since it is "hard technical work" and the fluff around it will turn into modules that are exchangeable. I think the Linux kernel with all the distributions around it is a good analogy in fact. I am not sure if you have followed the discussions on the module system but it will among other things give us the ability to name/find & download a "module" using only a path-notation like for example "People/gh/SuperTetris". So it will be a general publishing system as well as a system for maintaining what currently is in the base image. And thus I would like people to be able to publish, let's say - a shareware little game as outlined above. Or, to be more concrete - I think that Flow by Craig Latta is under the LGPL. That module should also be publishable in the modules system. Do note that the modules system is a virtual hierarchy (yes, it's in some ways a graph, but...) where subtrees can be located on different ftp/http servers all over the Internet. To start with we will probably set up one single root (but some form of mirrors of that root would be good for bandwidth/failure reasons) - an ftp directory on for example squeakfoundation.org. Then in that directory tree I could ask to place a "forwarder" at "People/gh" directing all subnames of that over to my ftp server in Sweden. (BTW - I have written up a few thoughts about the repository side of the modules system especially concerning how we handle the update streams, more on that in another thread...) > > Could I publish my GPLed code as a Squeak project in for example .pr > > form and include source with that and then bundle that with a Squeak VM > > and a small "boot" image that loads the .pr when started? (More or less > > mimicking how the Java guys do it). > > Not according to RMS. Certainly, any image derived from a mixed use > will be unpublishable. Such is the nature of FSF's notion of "free" > software. > > > If we could consider loading a .pr or ImageSegment as being "linking" > > then it should be ok, right? > > It should be. RMS thinks not. Oh wait, oops, I think I might have expressed this the wrong way. "Linking" is ok for LGPL but not for GPL. Sorry, I was implying that linking is ok for GPL - which of course is quite the opposite. I checked out the FAQ and found: http://www.gnu.org/licenses/gpl-faq.html#TOCInterpreterIncompat And that seems to imply that I could use a GPL license for my app with perhaps an added clause saying that running it in Squeak is fine. Or something like that. I also found: http://www.debian.org/doc/manuals/debian-java-faq/ch2.html (last paragraph) I assume that shipping a premade image is probably stretching it, that was why I was jibbering about a .pr file that the user loads when starting the app. I wondering how many Java developers know this... > > I would just like to have an answer when a GPL-advocate says he can't > > use Squeak for his app since it is not "GPL-compatible"... > > Answer, he is correct, at least as RMS understand the GPL. Squeak-L is > not GPL-compatible, and vice-versa. Reasonable lawyers may differ with > RMS on his interpretations of the agreement. Could you contrast this with my "findings" above? It all seems to come down to how "static" or "dynamic" the bindings are... > >> All we really need is the will to do it. In the meanwhile, the failure > >> to include Squeak in Debian reflects, to me, more of a weakness in > >> Debian policies than in Squeak-L, and is a shame. But in preparation > > > > Could you elaborate on that? I mean, do you think they are "wrong" in > > their argument or that they just are too "picky"? > > Clearly, they are free to add or refuse whatever they want. I think > that they are wrong to refuse to publish Squeak-L for the reasons they > choose. I'm a big fan of open source and of some, but not all, of the So... you mean that refusing Squeak-L because of the export restrictions (which I think was a biggie for Debian) was wrong? My personal view: Is that I understand Debian's puritan view of "free". They are creating a free OS for anybody to use - including people in countries which the US has decided restrict export to. ;-) I also understand the BSD/MIT camp and the RMS view of it all. I think it just depends on your viewpoint and the situation at hand. They all seem to have valid arguments from their respective viewpoints. BSD/MIT is the simplest one - free. Period. More or less like Public domain without giving up the copyright. Easy to understand. But those licenses also might suffer from poor feedback of contributions and they (as RMS would put it) do not assure that "the next guy in the foodchain" get the same rights as you do. So again, my view is that given a speicific situation I would probably choose between these: - GPL (perhaps without the "version 2 or higher" in case RMS looses it even more... :-) If I want my program not to be turned commercial by someone else - LGPL if I feel that GPL does a good job, but I want to enable people to use it commercially - BSD (or something similar) if I don't really care at all what people will do with my program (turn it commercial or whatever) - Non OpenSource (shareware, fully commercial etc.) - Another license that you can't really change, like Squeak-L for base image contributions I have a hard time seeing that Public Domain has any advantage and all the other licenses out there don't seem to really have any distinctive differences. > tenents of the free software movement. However, ideology has overcome > reason, so far as I am concerned. I do not share RMS' definition of > free which, for example in the case of software based on a monolithic > late-bound program image, makes it impossible to have mixed and non- > GPL'd software. > > In my view, such software isn't free in the common, linguistic sense of > the word, but hopelessly and unrationally over-restricted. No amount of > big-brotheresque double-talk to redefine the ordinary use of the word > suffices as argument to the contrary. > > As RMS freely admits, he is no advocate of open source software, if it > isn't "RMS-free." Likewise, I am no advocate of "RMS-free" software if > it isn't, in fact, free. Sounds like you are a BSD/MIT guy then! :-) regards, Göran

4 4

Re: Lots of concurrency
by Ken Kahn 30 Oct '01

30 Oct '01

Richard A. O'Keefe wrote: > > The question I think is interesting is whether telling students to > think in terms of objects TALKING to each other makes it harder for > them to think of concurrent implementations. Would some other > metaphor (perhaps sending couriers with messages, or thinking about > a factory with things concurrently moving from machine to machine > at the same time) make it easier for them to think of and understand > concurrency? > Before settling upon a carrier pidgeon metaphor for communcation in ToonTalk I considered lots of other alternatives including a postal system with mail carriers and mail boxes, rivers where you can place floating messages to those downstream, wires, fax machines, email, and telephones. I never considered face-to-face talking since I took for granted that objects are spread out spatially. I think the answer to Richard's question is that the right choice of metaphor can significantly facilitate thinking and understanding concurrent programs. Wires may be how messages are really passed within and between computers but wires are not an ideal metaphor. The messages are invisible. Wires cannot be sent along wires. Directionality isn't apparent. Wireless communication is even worse. Talking or telephones aren't good because both parties need to be available at the same time. Many-to-1 communication by telephone is confusing. While phone numbers can be communicated over the phone that would lead to a model where numbers denote objects or communication channels. So can they be made up and dialed at random? Not good. Even the postal system doesn't provide an ideal metaphor. If you want to send the receiving end of a communication channel to someone do you have to pick up your mail box and send it by mail to someone? And then does the postal system know that the mailbox has moved or do you need to think about forwarding addresses? A bird that when given something flies to her nest and leaves it there and returns works out very well. You can give a bird a box that might contain other birds and nests - not too strange. The only strange thing is that a bird always finds her nest no matter where it has been moved to. But that doesn't seem to cause any confusion. > All I know about Ken Kahn's ToonTalk is what I've read in this thread, > but it sounds as though uses a "physical" rather than "verbal" > metaphor, so I think his observations are particularly interesting here. > You all are welcome to try out ToonTalk. There is a free trial version and you are all welcome to try to beta version as well. It only runs on PC with Windows (or Macs emulating a PC with Windows - don't know about Linux emulators). Details at www.toontalk.com Best, -ken kahn

2 1

Re: Lots of concurrency
by Ken Kahn 30 Oct '01

30 Oct '01

Andrew C. Greenberg wrote: > > The bottom line is this: sequential programming can be verified (proved) > correct by straight-line Hoare or Dijkstra-style reasoning. Absent > simplifying invariants, concurrency without straight-line execution > requires proof that EVERY ONE of the combinatorial explosion of possible > executions are correct, or properly sequenced, PLUS an additional proof > of no deadlock. Whether or not we attempt to formally validate our > code, the cerebral effort needs to be made informally. > > It is bad enough to try to agonize over the cases of state-change in a > simple loop (and for beginners, hard enough to comprehend the notion of > a loop invariant). Imagine doing this for every line of code! > I think you've just provided another argument against adding threads to a sequential model of computation. Consider in contrast the actor model of computation pioneered by Carl Hewitt. Since all state is encapsulated in actors (or concurrent objects if you prefer) and all communication is by asynchrounous message passing you don't need to consider all possible interleavings of subprogram executions. The only non-determinism is due to the arrival ordering of messages. Each actor just pops a message off its incoming queue and processes it. Reasoning about invariants that an actor maintains isn't harder than reasoning about invariants in sequential code. Earlier in this thread there was a discussion of messages like "increment balance by" rather than "set balance" and "get balance". An important point underlying this is that "increment balance by" is a commutative operation so arrival ordering does not affect the resulting state when the actor has finished processing incoming messages. One can't always stick with commutative operations but it often simplifies things. Another recent message mentioned Linda and tuple spaces. I don't like the lack of modularity and scalability of tuple spaces but the major appeal they have is that they are insensitive to the order of reads and writes (since reads that occur too early suspend until the matching write happens). Again concurrency needn't make it hard to reason about your program if you have a decent model of concurrent computation. Best, -ken kahn ( www.toontalk.com )

1 0

← Newer
1
...
4
5
6
7
8
9
10
...
60
Older →

Jump to page: