Smalltalk Tidbits, Industry Rants

Phillippe Marschall is talking about the problem of managing Smalltalk servers Once you get past one server (and in many cases, even with one) - there's the "manage the server outside the image" problem. You can roll your own, but then you have to maintain that. Phillippe is showing us how he integrated Squeak and JMX to allow the use of "off the shelf" management tools.

Serge notes that the OLPC can run with Squeak NOS:

Technorati Tags: olpc, squeak, smalltalk

At Smalltalk Solutions last June, Randy Coulman presented his work integrating Cruise Control with Smalltalk - at Key Technologies, their builds are managed with Cruise Control. In fact, he's submitted the Smalltalk integration into the Cruise Control stream.

Technorati Tags: smalltalk, cruise control, sts08

Tudor is telling us about Pier, the open source Content Management System for Smalltalk. It looks pretty cool - and I'm capturing what he's doing onscreen for the video.

Technorati Tags: smalltalk, seaside, pier

Georg is giving the same talk he gave at StS 2008 (slides here) - but I'll have better quality video from ESUG.

Technorati Tags: bach, smalltalk, seaside

We missed the 8:40 bus to the venue this morning, so the video of Eliot's talk starts a couple of minutes in. Also, there seems to be some kind of construction going on above us, so I'm sure the audio will be affected badly by that. Nothing I can do about that :)

Anyway - Eliot's describing the work he's doing on Cog, which will be a modern VM for Croquet/Squeak. I'm not going to go into details on the technology choices here - Niall's report, when it appears, will have all of that. The short answer: Cog will have closures, a JIT - pretty much all of the things that appeared in the VW VM over the last decade.

Eliot's documenting his work on his blog - you can get all the details of the in-progress work there (and Eliot promises to update that with his latest progress soon).

On timing, the fast JIT should appear for one platform in April 2009, with more platforms appearing in 2010 and 2011.

Technorati Tags: smalltalk, Cog, Squeak, Croquet

Travis is boning up on Cairo while we're at ESUG 2008:

While many of the Smalltalk faithful are gathering in Amsterdam at ESUG this week, I'm at the Mozilla office in Mountain View California, attending Cairo Summit 2008.

Technorati Tags: cincom smalltalk, cairo, graphics

Here we are again (myself, Arden Thomas, and Dirk Verleyson), giving our impressions of day 2 of ESUG. We had an odd break in the middle - I had to go break down my camera when they wanted us out of the room, and we finished the recording outside. We talked about Monday evening's innovation competitors, the meta modeling talks, and the Gemstone overview talk. We'll be back tonight with a day three report.

Technorati Tags: esug08, smalltalk

Whoa - Yann Monclair has some fairly stunning stats on the size of the project he works on at JPM:

• Base image size: 2200 classes
• Image size with all app classes loaded: 22,000
• Roughly 60 - 150 classes touched per day
• Every day, around 25 change sets are applied (each with 5-8 classes)

Those stats are why they need to get a new image every day - they really, really need to avoid long splits (forks).

Heh: "We try to keep Envy happy. Normally, Envy is very grumpy". They work to keep changes small, pre-reqs accurate, and to avoid unknown dependencies. So for instance, they really don't like using things like:

#{MyClassOrVariable} ifDefinedDo: [:cls | "specific code here" ].

Now Yann is doing a quick overview of Envy. I'll leave that aside - you can look here (yes, it's old, but it's still accurate for all the basics, as Envy is mostly moribund).

The process of loading up:

• Load the top level map (base Envy functionality)
• Validate the build
• two types of test:
• Code driven (SUnit)
• Data driven (end to end testing

Developers always start with a fresh image each day. They then ensure that all of their code loads into that fresh image. They don't like saving images, because it leads to working on older snapshots. The build process they have takes 40 minutes, but it ensures that all developers are always working on an image that can be reproduced from scratch. A saved image cannot necessarily be reproduced.

Builds will break - just accept this, and don't let the build process become a burden. In this context, a broken build means that one or more SUnit tests failed. Heh - identifying the issue isn't like "House" - failure to identify it in 30 minutes won't kill your project :)Two main culpruts for failures of SUnits: bad code, bad data.

More problematic is when the build doesn't even complete (i.e., a really broken build). Well - this being Smalltalk, we can debug it :)

The most common reason for such problems: prerequisite issues.

The next potential problem: a working build, but your code doesn't load into it. For JPM, this often has to do with the mixture of overrides and Envy. Another issue: do you rely on code that is not released by some other member of the team ? Botched merges can cause problems as well.

"Always find the error before you fix it" - i.e., you're not "House".

Technorati Tags: smalltalk, envy, kapital

Martin McClure is giving a general overview of GemStone/S - the fact that we have an audience that needs a general overview of GemStone is a good sign all by itself (lots of new people).

If you're interested in getting started with GemStone, you can head over to their site and grab their non-commercial download.

Here's an interesting observation: How big can GS go?

• Not limited by memory (it's a database)
• Billions of Objects
• Hundreds of Gigabytes
• Collection sizes in the millions
• indexes
• queries

"GemStone isn't an object database, it's a Smalltalk implementation that includes an object database". Another interesting thing - you can work with different versions of objects (not everything needs to be migrated up at once).

In general, this is a great introduction to GemStone/S - the video will be worth watching.

It's meta day here at ESUG - we have a panel on Meta Modeling now with Lukas Renggli & Mathieu van Echtelt & Adrian Kuhn & Tijs van der Storm. They are trying to cover:

• A Comparison of the systems (the ones they all work on)
• Roundtrip engineering
• Collection Hierarchy?
• Versioning
• Broader Context than "business" relationships
• Portability, Interoperability vs. Productivity, Reuse
• User Interface rendering, naked objects
• Persistence
• Simplicity

Code generation (Lukas doesn't do it in Magritte) is getting a lot of talk here. Again, I'm getting a video of the entire talk, so you should be able to see the whole thing fairly soon.

The topic of code generation - good or bad - seems to be generating a lot of discussion with the audience.

A quick talk from Lukas on Magritte - just a quick overview of the work, which you can find more about here.

One of the main things Lukas is getting across is that Magritte is not related to Seaside - there's a Seaside plugin, but that's the extent of the relationship.

Technorati Tags: smalltalk, magritte

Tijs van der Storm is talking about some of the Meta Modeling work that goes on here at CWI. In some ways, this talk is reminding me of the OMeta talk at StS - I'll let the experts decide how much overlap there really is :)

About 37 minutes in, Niall Ross asked specifically about this stuff compared to OMeta. So - OMeta uses PEGS, they use SDF. And their work is less general than OMeta - they are targeting DSLs, as opposed to general language development.

Adrian Kuhn is talking about meta modeling in Smalltalk - Fame is a project that grew out of Moose.

As ESUG continues, slides and other stuff is being posted here. I'll be throwing the video over to the STIC site, just as I've been pushing StS content there.

Mathieu van Echtelt is giving a talk on what CosmoCows does - they build custom software, all of it web based. They had an interesting business plan in 2000 build a business "doing something with Squeak". For fairly obvious reasons, that didn't really work out that well :)

They now build custom software, most of it web based - for customers like fire departments and ambulance services in the Netherlands.

They are using Cincom Smalltalk for all this - I have most of his demo on video (my camera needed to be unloaded during the break, and that process ran over into his talk. I will have the full audio from my H2).

Technorati Tags: smalltalk

Gilad Bracha is giving the same talk he gave at StS 2008 - the slides are here. I'm taking video again, so I should have that up by the by.

It looks emptier than yesterday because this is the first talk of the day, and more than half the crowd wandered in late :)

I also took this photo - I have no idea how the bubble artifact got into the image :)

At the end of the day yesterday, Arden, Dirk, and I sat down and had a short conversation about the first day of ESUG - it's a short mp3. We plan to do one of these each night, and I should be able to post the audio each morning.

Technorati Tags: smalltalk, esug, esug08

Patrick Logan explains :)

This is interesting - Julian Lombardi is giving a talk on Croquet via skype, with Robert Shepard handling questions that come up (he's using a headset to prevent audio feedback). Here's a picture - we have Julian on one screen, and the presentation (and later, Croquet) on the other:

It'll be interesting to see how the video for this comes out.

I missed Bruce's talk on Smalltalk standards - the heating control talk went long. I'm back in the main room for Bruce's talk on a Smalltalk interface to Syslog. It's a commonly needed thing - realistically speaking, a Smalltalk server might want to produce system logs instead of its own, so that standard tools could "see" them.

Bruce has this working in VW and Gemstone (using his Sport framework for portability). He's demonstrating it for us, so we can see that it works :)

The talks are split up now - I'm watching Alfred Wullschleger talk about the heating control system he uses to run his house - which he wrote himself in VisualWorks.

It's a commercially built control unit, so it's not a homebrew, one-off kind of thing. What was the motivation? The control unit that shipped with the unit wasn't running efficiently, and Alfred likes to tinker with hardware - so using a standard PC with USB interface, he was able to simulate the unit in Smalltalk, and, via DLLCC, hook up to the C API.

• Interface to the hardware via a DLL for the Minilab 1008
• All control done in Smalltalk
• User Interface in Smalltalk
• Sockets for remote operation
• Simulation of all hardware
• Uses an energy flow model
• Runs 10x faster than the real system!

The end result is that he has a headless image on the boiler connected computer, and he can remotely control it via any system on his LAN.

John McIntosh is giving a talk on GStreamer - he's been building support for it into Squeak lately. It's a pretty large open source library that supports various media formats, so it's a useful thing to have.

Sounds like he's added it in using primitives, and is thus far 32 bit only. Interesting problem in terms of supporting it from Smalltalk - GStreamer has its own garbage collection framework. That means you don't want to try to have its objects managed from the Smalltalk side, of course.

Leandro Caniglia is reprising his talk from Smalltalk Solutions (grab the slides here) - here he is getting started:

In the context of this talk, "changes" are not analogous to the Smalltalk change log - he's talking about the changes made by an end user of an application

Arden is giving the roadmap talk here - it's the same one he gave at StS 2008, so I won't go into too many details - you can see the slides here, the StS 2008 talk here, and the web page over here. I will be posting the video and audio stream from here though, as I get to that.

The main point Arden's getting across is that we want to hear from you about where the products should go. If you're interested in VisualWorks, ObjectStudio, or Web Velocity, and you have suggestions, let Arden know.

Technorati Tags: smalltalk, cincom smalltalk

The Squeak Guys generated a fairly good crowd for "Camp Smalltalk" over the weekend:

Over fifty Smalltalk developers have spent the last two days working on a variety of Camp Smalltalk

Technorati Tags: smalltalk, camp smalltalk

ESUG 2008 has started here in Amsterdam - I'm taking video of the sessions, and will be posting that over the next few weeks/months (as I am with the stuff from StS 2008). Anyway - some photos:

Opening Remarks

The crowd at the conference:

Monty's Wife, Bruce Badger, and Monty during a coffee break:

Andres Valloud talking about Assessments:

The last photo above is of Andres Valloud giving his talk on testing and Hashing in Smalltalk. The talk is largely about his work on Assessments, which he's written about extensively on his blog

Here's Rob Rothwell from Smalltalk Solutions 2008 - thus far, I don't have the video - I'm still having issues converting from the Sony HD format I have to something I can work with. In any event, grab the slides here, and the audio here.

The talk was very well received (I missed it, since I was giving a Seaside tutorial) - but people were talkng about it very favorably.

Well, I've finally arrived at the hotel, and discovered bad internet :) Everything seems to be blocked other than port 80, Skype, and - inexplicably - Yahoo Messenger. No mail other than Google via the browser, either. So - I'll be hard to reach this week :)

I must be catching whatever it is that Michael has when he travels. My flight to Philadelphia was delayed, which blew my flight to Amsterdam. So USAirways sent me to Munich. The good news - I got here in time for the 11:15 flight. The bad news - I'm in "peasant class", so they wanted 100 euros to get on that one. I decided to pass, and take the 2:55 flight they have me booked on.

There's wifi here (paid), but all the electric outlets are well away from the chairs. So after an all night flight, I'm sitting on the floor, with nearly 4 hours to kill :)