CeBIT is just a massive show - unless you've been here, you literally hav eno idea how big - 22 halls, each as big as the typical trade show venue - software, hardware, consumer goods - it's mind bending. I just spent an hour browsing, and only made it through 5 halls....
It's been a trip so far. Th eflight to London was ok - but here's a tip - don't fly into Heathrow and drive to Gatwick. I'm not familiar enough with the roads to have done that tired :). I finally arrived at my hotel, only to find that I'd been transported back in time - not only was there no internet connectivity, there were no phones in the room! I had to rent an alarm clock...
I had forgotten my itinerary, and making a calling card phone from a public phone boot was an odyssey of begging for operator access. That done, I finally went to bed. Got to Gatwick this morning, hoping to find some WiFi access.... not a chance, not even in Starbucks (there are 2 signals, neither of which gives me net access). So I'm slogging completely off the grid now... who knows when I'll be back on. On to Germany, and hopefully to network access...
Opentalk, Load Balancing, and Multi-Image SUnits
Leonard Lutomski: Cincom
Tuesday 8:30:00 am to 9:15:00 am
Abstract: This talk will demonstrate the new Cincom Opentalk Load Balancing Facility and its designed ability to support several different load balancing architectures. The multi-image SUnit framework used to develop and test the load balancing facility will also be addressed.
Bio: Leonard Lutomski has been developing in Smalltalk since 1985 and manages the Cincom VisualWorks Protocol Team.
See you in Seattle!
0xDECAFBAD points to a net based fundraising effort to save Angel (the TV show). This is an attempt to harness the web in the same way that many political campaigns have been doing this year. Will this spread?
BitWorking comments on how hard character encoding is to get right:
Character encoding is hard. Really. If I could point to one thing that causes feeds to be invalid more than anything else, it would be character encoding.
This is the primary reason that BottomFeeder does not reject bad feeds - it just ignores bad characters to the largest extent that it can and moves on. The reality is, feeds move in and out of well-formedness on a regular basis. There are so many people posting so much content from so large a set of tools, that it's simply unrealistic to expect ongoing perfection. I periodically get errors in some of the CST feeds - I'm not entirely certain how, because it's nearly always a comment that came from - somewhere. Easy enough to fix when I notice, but I don't always notice. I'm pretty sure other content producers have the same problem - most of them aren't hosting the server themselves, and most of them have minimal control over whatever it is that the server does. It's not that anyone wants to create malformed content - it just happens. In the meantime, content consumers still want to read the content, even if it has a few bad characters in it for a period of time. Stating that client side applications should just reject that content out of hand is simply anti-social, IMHO. Sure, notify the user that there's a problem with the content, and let them contact the provider if they feel like it. In the meantime, you shouldn't punish grandma because of an error she has no control over...
Patrick Logan reminds me of one of the reasons that Smalltalk is such a pleasure to develop in - the ability to read code. That sounds silly - after all, you can read code in any language. Smalltalk makes things easier in a few ways though
- It's "Open Source" in the access sense - you have access to all code all the time. This is true even when debugging - none of the system classes or libraries are closed off. Need to change/modify/work around something? Go ahead
- Smalltalk is very easy to read. Keyword messages in particular make it much easier to follow the intent of the code
Here's an example of the latter - the following code in BottomFeeder is used to parse an XML document from an http source into a Feed domain object:
Smalltalk code: Constructor parseAndProcess: anUrl into: aFeed. C style syntax code: Constructor.parseAndProcess (url, aFeed);
The Smalltalk code just seems clearer and easier to read - the keyword style of messaging explains each argument, making it more like a textual sentence than like code to be decoded. That just makes it easier to read code and figure it out. A common complaint I hear from manifest typing advocates is that the lack of type information hinders that reading process - to which I respond: How? Seriously - the fact that some object has a given type tells me far, far less than an intention revealing selector does - and keyword style messaging makes it easier to do that. I've been put into the middle of large bodies of C code, and in the middle of large bodies of Smalltalk code. The type information just doesn't help in the understanding process all that much. I find it far, far easier to figure out a large body of unfamiliar Smalltalk code than a similar body of C, C , Java (etc). The fact that there's typically a whole lot less Smalltalk code to look at also helps a lot.
You're going to end up reading a lot more code than you write - the easier it is to do that, the more productive you are likely to be.
We have a preferred conference rate with a reserved block of rooms at the Crowne Plaza, but the block is reserved only until April 2nd, after which time the preferred rate may not be available. Also, if we don't have the block filled by then the conference may have to pay a penalty. So it would probably be good to make room reservations before then if you haven't already done so
See you there!
I just made an update to the category search function. I implemented a cache that sped things up, but there was no proper date sorting. Why was that? Well, this is one of those times when my implementation "leaked". Here's what I did:
- Each posting has a category
- I created a cache dictionary, where the keys are the categories, and the values are a collection of files that have (one or more) postings of that category in them (there's one file per day in this implementation)
- The collection was a Set, to prevent the same file from showing up twice in a particular category list
Using a Set, and then not dealing with that was the problem. When posts are placed on the screen, I just iterate over all the posts asked for and render them. That's fine when they are already in reverse chronological order - but a Set is unordered. I hadn't considered that, which is why - until five minutes ago - category searches were bringing back posts in a random order. It's fixed now :)
I'm off to Europe tomorrow morning - I'll catch the tail end of CeBIT, then meet with a bunch of our customers in Germany. From there, I'll be heading to the UK to speak at Ot2004. I'll be blogging the conference - I'm hoping there will be WiFi access. This is a long trip - 10 days. It's likely that BottomFeeder updates will slow to a trickle while I'm on the road - especially if connectivity is an issue.
Making relational data first class
Avi Bryant: Beta4.com
Tuesday 8:30:00 am to 9:15:00 am
Abstract: ROE, the Relational Object Expression library, models relational queries as first class Smalltalk expressions. This has several advantages over using SQL strings directly:
ROE can either be used directly as a better interface to relational databases, or as a new foundation for object/relational mapping tools. Currently, it has only been tested using PostgreSQL and Squeak Smalltalk, but it could easily be ported to other platforms.
- queries can be built using familiar Smalltalk syntax and without worrying about binding and escaping data
- queries can be easily composed, so that a complex query can be built up over several methods, none of which know any details about the others
- queries look like ordered collections of tuples, but with most operations other than #do: defined lazily; for example, #copyFrom:to: won't pull in any data, but will simply produce a new query with an extra clause
- queries maintain a rich set of metadata, so that, for example, the columns of any query can be automatically grouped by table and its rows mapped properly into objects.
And don't miss Avi's keynote on Seaside!. See you in Seattle!
Scoble makes a point that I've made here many, many times - making software is expensive, and it has to be paid for somehow. There are a few people who don't seem to get this - apparently, the notion of having to pay bills never occurs to them. Now, this really doesn't have anything specifically to do with Closed Source vs. Open Source; you can make money off of either model. Here's what Scoble had to say:
Eben Moglen asks an interesting question: "If I can provide to everyone all goods of intellectual value or beauty, for the same price that I can provide the first copy of those works to anyone, why is it ever moral to exclude anyone from anything?"
Because humans are incented to do more when there's motivation, that's why.
Society learns this over and over and over and over. Communism vs. Capitalism. In every instance, humans do better when the people who do more for society are rewarded.
Which is exactly correct. People will only do so much (and in most cases, that's not a lot) out of altruism. Take BottomFeeder, for instance. Sure, it's freely available, and it'll stay that way. However, my ability to work on it is directly related to it's connection to my job (promoting Smalltalk in general, and Cincom Smalltalk in particular). Don't believe me? Go look at SourceForge - how many of those projects are active? Look at the big, successful open source projects - they all have funding (either direct or indirect). Non-trivial software projects are just too hard to be sustained by any model that doesn't involve some kind of funding.
So to get back to the question Scoble answered - if you provide a good freely, it's value will end up approaching zero. If there's no compensation, no one will feel any compulsion to support or create that good. Ultimately, we all have to eat :)
I've just updated the blog - category searches are now much, much faster. I'm working on a fix for title/keyword searches that will be as dramatic, but don't have that done yet. The category search was fairly simple - I just added a cache. The cache is dead simple - a dictionary lookup by category name, with the value being a collection of all files that have at least one post from that category in them. Now, when you do a category search, the system looks in that cache instead of across all blog posts every made.
Now Gartner says that there aren't enough skilled Java developers to go around:
Research firm Gartner says only 32% of the 2.5 million Java developers in the world have genuine knowledge, which means there is a serious lack of high-level development skills. "This has resulted in a tremendous backlog of projects," says Aad Van Schetsen, Compuware sales director for application development and integration solutions in the Europe, Middle East, Africa region. "Some European companies need to complete up to 500 projects in the next two years."
But go ahead and ask their analysts which languages to use, and they'll say Java and C# - they are mainstream, and thus "safe". Never mind the lack of skilled developers (meaning: you'll have to train them, just as you would with a non-mainstream solution). Never mind that you'll be competing with tons of other companies for these people. Never mind Gartner's own analysts pointing to high failure rates. Never mind any of that, because being popular is all that matters. Pay no attention to the man behind the curtain!
Replacing Oracle with GemStone/S: The Agony and the Ecstacy
Monday 4:45:00 pm to 5:30:00 pm
Abstract: Replacing Oracle with GemStone/S was a painfully process. Despite the pain, it is paying off handsomely. This presentation will cover the business, cultural and technical aspects of making such a transition.
Bio: I've spent the last several years(1996-1999) building and enhancing web systems for the travel industry in VisualWave and GemStone/S.
After a couple of years working as a Software Engineer at GemStone, I'm back in app development. My current project is replacing Oracle with GemStone/S as the Transaction DB for a banking application. There's nothing more satisfying that ripping out O-R mapping code. ;-)
See you in Seattle!
Peter Lount will discuss FastCGI, Smalltalk, and Apache in his talk at StS 2004. Come see how it all fits together:
FastCGI for Smalltalk: Integrating Smalltalk Into An Apache based Web Site
Peter Lount: Active Information Corp.
Monday 4:45:00 pm to 5:30:00 pm
FastCGI for Smalltalk: Integrating Smalltalk Into An Apache based Web Site
A presentation on the FastCGI system created by Tomas Vanak and Peter William Lount. FastCGI for Smalltalk enables a Smalltalk virtual machine to be able to integrate seamlessly into an Apache Web Server based web site along site with PHP, Perl, JSP or whatever else the site is running. FastCGI for Smalltalk is written entirely in Smalltalk and has versions for Smalltalk/X (the original version), Squeak and Dolphin Smalltalk. Porting is quite easy. FastCGI for Smalltalk takes advantage of and implements the FastCGI Protocol which is a protocol that can be added to Apache. It's implemented as a C based Apache Module. Before selecting FastCGI, Peter looked at a number of other methods of integrating Smalltalk and Apache and choose FastCGI as the quickest path to achieve this. The FastCGI for Smalltalk effectively enables Smalltalk based web applications that run on the same server or on application servers. FastCGI for Smalltalk was written in less than a month by Tomas Vanak at the request Peter William Lount. The request was published on http://mod.smalltalk.org/ and after six months Tomas picked up the torch. Peter and Tomas worked together designing, debugging and extending the main body of Tomas's code. A FastCGI "hub router" prototype was written in six hours that takes FastCGI requests and "routes" the requests to multiple Smalltalk images located on the same or other servers. Peter William Lount has written an extensive web engine known as the AIMS Web Engine that is being used in a number of web sites under development (including Smalltalk.org). Recently a HTTP server has been added to the FastCGI for Smalltalk capabilities. This was possible since the Smalltalk based FastCGI Server has a lot in common with an HTTP server. FastCGI for Smalltalk is open source.
What is FastCGI?
"FastCGI is a language independent, scalable, open extension to CGI that provides high performance and persistence without the limitations of server specific APIs. ... FastCGI applications use (TCP or Unix) sockets to communicate with the web server. This scalable architecture allows applications to run on the same platform as the web server or on many machines scattered across an enterprise network. ... FastCGI applications are fast because they're persistent. There is no per-request startup and initialization overhead. This makes possible the development of applications which would otherwise be impractical within the CGI paradigm (i.e. a huge Perl script, or an application which requires a connection to one or more databases)." http://mod.smalltalk.org/
This presentation will be presented by Peter William Lount.
Bio: Peter William Lount is the president of Active Information Corporation and has been using Smalltalk for over twenty years since the early 1980's. Active Info has worked with companies such as JPMorgan, Fannie Mae and Deluxe Check Printing applying his Smalltalk knowledge in the banking and financial services industry as well as the construction, real estate, medical, web services and other industries. Peter wrote and owns the Metameric Bridge software that was used to design and build the Vancouver Sky Train "via-duct" (bridge) and other bridges around the world. Through Active Info Peter offers consulting, mentoring and coaching services. In addition Peter is the senior editor of Smalltalk.org, a Smalltalk advocacy web site. For more info please see http://www.activeinfo.ca, http://www.peter.lount.com, and http://www.MetaMere.com.
See you in Seattle!
A Very simple deployment tool. It wraps the complexity of RTP (by avoiding most of its features) to allow for a fairly simple way of creating a sealed image and a Windows executable. There are 3 steps:
- Seal the image. This assumes that you have an image ready to be sealed. You startup Deployment.DeploymentToolUI, fill in the image name you want (NOT the same one you last saved the working one with :) ), the startup class, and the startup method. Specify the namespace using dotted notation. Then hit "Seal Image". That will save the image without doing any stripping.
- After that's done, bring up a working image with the deployment tool loaded. Start it up, and compress the image. This will create a smaller image that can be combined with the VM for a single Windows exe
- After that's done, the last button will create the executable.
That's it. If you know how to use the RTP, you can create a smaller, more efficient runtime with it - but this works fairly well, and is very simple. Enjoy, and report any issues to me
I received this question in email this morning:
With the help of David Pennell, I've gotten Seaside working. When I compare the behaviour of Seaside under VW, it differs from that under Squeak---aka, the VW port has some bugs. Now, the first thing I'd normally do in a file-based system is to do a compare of the sources. But in Smalltalk, how do I do this?
Well, comparing sources in a Smalltalk system isn't hard, assuming that all such sources are in the version control system - for VisualWorks, that would mean Store. So assuming the base Squeak code (which would not work in VW) were versioned off, one could run a comparison using the Store tools:
- Connect to Store
- Load the version you want to compare
- Select a package or bundle in that version and - on the right mouse menu - do a comparison with the original (presumably Squeak in this case) version
Now, there are going to be some issues with this kind of thing - Squeak and VW are not plug compatible - meaning, you might not be able to get the base Squeak version loaded into an image (and thus versioned into Store) in the first place.
Now, if you have the code in chunk format from Squeak (i.e., a file out) - you can run a comparison independent of Store. (this is how I compile diffs between versions of VW here).
- Open a ChangeList tool
- Read Sources from the chunk file exported from Squeak
- On the Remove menu, Same Code as System and Same Source as System
- On the Forget menu, select Forget All Marked
Now you have the differences, which you can get a reporrt on by running Compare with System (assuming you have Seaside loaded). Now, there's another issue you'll run into here - VW has namespaces, and Squeak doesn't. If the Seaside code in VW is in a differnt namespace than Smalltalk, then pretty much everything is going to come up as a difference. That's a harder issue to deal with...
I saw a complaint about how hard it is to deploy a Smalltalk (VisualWorks) application earlier today (in email). Now, engineering is working on this problem - the goal is to produce a runtime environment for VW that makes deployment straightforward. In the meantime, however, it's still harder than it needs to be. For WebToolkit apps, Alan has made it pretty simple - there's a simple set of options under the "web" menu once you load Web Toolkit. What about GUI apps? Well, I've been working on a simple tool for that today. I'll push it to the public Store once I'm done testing it - it should be as close to "press a button and deploy" as you can get. There will be two steps:
- Seal the image (saves a sealed image, with a starting class/method specified)
- Create a Windows executable, ready to deploy
The first step (borrowing heavily from Alan's approach) wraps the Runtime Packager, taking the simplest way through. If you actually understand RTP, you'll want to do it the hard way. The second step simply wraps the image compression and executable creation steps (Windows only) that are not as easy to find as they should be. I'll post on this again once I push the tool out
d2r has an interesting post up. Apparently, a post made last year by this blogger has attracted legal attention - the post in question disparaged the educational credentials of this well known personality. This is fascinating. What we have here is a transient post from months ago (which a google search doesn't immediately turn up) - but the serving of papers is likely to generate a small avalanche of "isn't this absurd" posts from all over. You would think that famous people would be somewhat clued in to the dangers of publicizing things that could backfire, but I guess not. Negative marketing lives on....
I've had an older set of stereo equipment in my office for a few years now. I had been thinking that there was a problem with it; one channel was always playing much softer than the other. Then today I was looking at the controls - I was seeing if I could receive AM stations on it (turns out I need an antenna). That was when I noticed something interesting. There's a mono/stereo button on the front panel of the system, and it's been set to mono (who knows for how long). After pushing that to "stereo", suddenly I had the other channel. So much for any thought that I had a discerning ear for music :)
It looks like some marketing departments can't be left without adult supervision. While this is a rather extreme "marketing gone wrong" event, it just goes to show that marketing messages are too important to be left solely to marketing....
I upgraded to Eudora 6 awhile back, and I've been mostly happy with it. The spam filtering seemed mostly good as well, until recently - when I started noticing that scads of good mail was being junked. I spent the last week or so fishing things out of the junk folder, until it finally dawned on me that maybe there were settings for this. Lo and behold, there is.
It turns out that Eudora assigns a "spam ranking" to each incoming mail - a number between 0 and 100. There's an option to set the minimum ranking before having something get junked, and that was unset. I'm hoping that setting that will result in fewer lost messages.
One of the interesting things to me is how many people will say things like "Yes, Smalltalk is more productive, but...". The next thing you'll get is some statement about how hard it is to find Smalltalkers, or about how the syntax is wrong (what, developers can't learn a new language?). You'll get the same complaints about any niche language - Lisp, Ruby, Python, Scheme - they all get tarred with the same brush.
Heck, the major analyst groups are particularly bad about this - Gartner will admit that Smalltalk is more productive (i.e., will deliver results faster at a lower cost) - and then tell you to use Java or C#. It's always high school in the hallowed halls of Gartner, and all that matters is being close to the "popular" crowd. When you look at the data from SPR (I'm getting the actual tables from SPR; they charge for them now) - you can see hard data backing this up. The tragedy is, this is old news.
Georg heeg will discuss outsourcing and whether or not software development is financially feasible in the developed world - in this session at StS 2004. Register today so you can participate in this timely discussion
Is Software Development in Developed Countries still affordable?
Georg Heeg: George Heeg eK
Monday 4:00:00 pm to 4:45:00 pm
Abstract: This is political/technical talk about alternatives to outsourcing.
Bio: 50 years old, founder of Georg Heeg eK, Germany oldest Smalltalk enterprise
See you in Seattle!
Charles Miller defends Wiki style markup for wikis:
Sure, you end up with something that's significantly less powerful than HTML. This is a feature. A wiki page isn't a place for complicated markup, it's for writing stuff down. The more power you put in the markup language, the more people are going to be wanking around with the precise arrangement of angle-brackets that will make their paragraphs step from left-to-right in pixel-perfect harmony in lieu of saying something.
I support the same style of markup in the Comment poster in BottomFeeder, and in the client-side posting tool for this blog. I'm going to add it to the web comment form as well, for the reasons outlined by Charles. Most people - techies included - don't want to use html markup....
If Seaside and continuations are of interest to you, then you'll want to read the following items:
- Peter Lount's questions (no better links available; he's runing an older SmallBlog implementation)
- Avi's response
- Learning Seaside's take
It's great to see a lively technical discussion on this stuff - and the general interest that is being stirred up by it
You can adjust the font scales globally in BottomFeeder in settings - the pane under "user interface". If you want to adjust the scale of the HTML pane only, the third toolbar button from the right (Glasses with a bi-directional red arrow) allows you to adjust the font scale just for the browser pane.
Danny Ayers was apparently noodling around in Smalltalk and came across the "NonBoolean receiver - proceed for truth" error. I remember when I first hit that one, many years ago - it baffled me. Even now, I usually do a double take on it :)
One of the interesting things you can see in software (probably anywhere) is the overly developed sense of importance many developers attach to themselves and their projects. Certainly we Smalltalkers are not immune to this; we often talk about Smalltalk as if it's the second coming or something :) There's a larger thing at work here though - it's touched on in this rant from the Bile Blog - and while the commentary there is a little rough, there's a good point hidden in all that anger - we often choose the complex over the simple for reasons that have nothing to so with the actual problem. Here's an example, related to me by a friend.
This guy has recently taken a new job - same kind of work he's always done, but with a different consulting outfit. During the initial process of looking at available projects, he's talking to someone about a job that involves getting data from a database, allowing users to interact with it, and possibly updating the database - you know, the standard CRUD thing. He asks what approach they are going to take, and is told that it'll use and EJB server, a big database, and a Java client - a three tier system. This is when my friend made the mistake of asking the following:
Question: How many users will this system have?
Answer: 10 at first, maybe 20 eventually
Question: Why not just use an Access front end to the database?
He says that the lead developer on that project distrusts anything he says now. Stop and think about that, and about the general problem - most of the systems we build are not big. Most of them need to scale to small numbers of users, and most of them are simple CRUD systems. What do many, many developers immediately do? Over complicate them. Suddenly, it's not such a surprise that the failure rates for software development jobs are so high - we tend to over estimate the importance of projects and over complicate them
Take this blog, for instance - I've had a few people ask me what database I use, and they are often horrified when I tell them that I don't use a database. Why should I? I save the data in serialized object files, with the date of the postings encoded in the file name (one file per day). It's easy to find posts that way (the GUID is also an encoded date), and the file system serves things up pretty darn fast. I can back all the content (and infrastructure) files for all the CST blogs in less than 3 MB using a shell script. All my data is still in object form; I didn't have to worry about any impedance mismatch between Smalltalk and an RDBMS (or about db architecture issues, about which I know very little). Sometimes, simple problems only need a simple solution
I've seen the same sorts of things in many shops over the years. I recall one large development project in particular, on which I and a few other Smalltalkers consulted back in the late 90's. At its peak, there were 170 or so Smalltalkers, and nearly double that working in Cobol on a mainframe. Over lunch, a few of us determined that the entire thing could have been done with 2 dozen or fewer people, and likely in a matter of months - but that wouldn't have helped the management there build their empires. Well after I left that project, I learned that large parts of it had been outsourced. No surprise there - years of game playing and complexity adding had to be paid for at some point.
What's my point? Well, next time someone starts waxing rhapsodic over the need to use a whole set of complex technologies to solve some problem, sit back and ask yourself just how complex the problem really is, and shed any prejudices you might have about the issues of client/server development - it's really not the case that all problems require a 3 tier solution...
I just found out that there's a new blog search site around - BlogRunner. It seems similar to Feedster, but with one important difference - Feedster offers support for ad-hoc (search) feeds (as does BlogDigger, and BlogRunner doesn't. Limits its usefullness, to my mind. Not as bad as the hokey email registration scheme that pubsub uses, but heck - you would think a service that indexes blogs would grok the importance of syndication...
The OpenSkills Skillsbase Project
Bruce Badger: OpenSkills
Monday 4:00:00 pm to 4:45:00 pm
Abstract: The OpenSkills SkillsBase is implemented using the Swazoo HTTP server running in GemStone, all hiding behind a Squid reverse proxy. Bruce will discuss the goals of OpenSkills, and how you can help out
See you in Seattle!