PDC 2003

blog

The Gallery

It will surprise nobody that I tend to put off packing until the last minute. This tends to minimize effort but leads to exciting surprises. For example, only today we noticed that our old carry-on overnight bag has been disposed of. Actually, it barely held together for many years, but I had forgotten that we had finally thrown it away on our last use of it. If I had remembered, we would have made a special trip during the week to buy more luggage, but with three hours to catch the plane, I just grabbed a duffle bag the kids had. It has no wheels or shoulder strap, so it’s a pain to lug around.

So there’s a long line at the Alaska counter, and my e-ticket doesn’t clear through. Uh oh. So I get into the long line and wait. The buzz in the line is that all flights are cancelled to Southern California due to smoke. After about 45 minutes in the line I wonder if my stop-over in Portland should matter. Some folks are heading home to work out their reschedule from there. I step up to the e-ticket kiosk and try again, and I’m rewarded with a boarding pass. After stripping and unpacking for the security checkpoint, I get to the loading counter at the gate. The agent there is surprised that I have a pass through to LAX, but her computer shows that the flight is still scheduled. I choose to risk a night stranded in Portland for the chance to get to the PDC and board the plane.

The board shows my flight still scheduled to LAX, though it is running a little late. We’re hearing that the fires forced the evacuation of one of the regional air traffic control centers. The Alaska employee next to me on the plane said that all flights south of Portland have been cancelled, but things seem to be better than this. The delay leaves time enough to stop in the Powell’s Book Store for some reading material. I have hope that it will beat the average airport bookshop. It’s still an airport bookshop, but a pretty good one. I pick up Dilbert and the Way of the Weasel. Trudy says that it will be good reading for me, and I’m sure that my brother-in-law would agree.

Now the flight is delays until 8:45, about three hours later than originally scheduled. Apparently the smoke delayed our flight crew in arriving in Portland.

Landed at the airport and taxied over to the hotel. We did see the fires while flying over LA, and it looks pretty bad. There is a smoky haze over the city, though that itself isn’t that unusual.

I had wondered if the air delays were going to affect the PDC, and there was a note on the desk to that effect. It looks like it’s going to be a short night.

I’ve spent a few minutes trying to get fast internet connection in my hotel room, and something isn’t working, but I’m going to bail for the night. Hope to get you all updated tomorrow.

Do you remember the “juice” scene from Tron? The programs find the glowing fluid and get a quick refreshment. I just found an internet port at the PDC (in the Tools & Language Lounge) and I’m feeling refreshed. I seem to have the port in my hotel room working, but only intermittently. I do wonder if the PDC has turned downtown Los Angeles into Nerd City; we may be overtaxing the infrastructure.

The average PDC attendee looks about like me: white, older, male. We are also uniform with respect to dress, though I suspect that the pioneers wearing shorts will have more company tomorrow.

There weren’t any real lines for registration, maybe because I’m here so early. Bill’s Keynote is in an hour or so. There is one downer to the morning already, though. As a Microsoft employee, I cannot pick up the cool registration goody bag, which includes a real backpack sponsored by HP. (Everything is sponsored here. My shuttle was sponsored by Borland.) This is despite the fact that I think I paid the full registration price to attend. Oh well.

This is a geekfest! To be specific, there are 7,000 geeks in attendance, watching TV together. Even when the live presenters are speaking, some of us have block views of the podia, and the TVs are really big. Of course, the front and center section was reserved for the press. Microsoft employees are supposed to cede the best seats; I picked an obstructed view seat.

The presentation started with a super geeky Halo video which was a little too much MSFT rah-rah for the audience’s taste, I think. Bill showed a video that was better received, showing the “mistakes” of the past decade. Highlights were Microsoft making fun of itself for Bob (user testing with chimpanzees) and missing the boat on the internet. I expect that this video may be shown at the company meeting on Thursday.

Bill has become a polished speaker, but perhaps not a totally captivating one. I still reflexively count the number of times he pushes his glasses up his nose. Today’s count (in about 40 minutes of stage time) was one. In 1990, he averaged slightly over 1/minute in a meeting I attended. But his hands still fly around a lot, and that certainly makes the clicker in his hands distracting.

Jeff will be glad to know that Trustworthy computing was his first and highest priority topic. I think that managed code is supposed to make things better.

In talking about the Watson technology, Bill notes that it can be used to report “help that isn’t helpful.” This was a major problem when WinHelp shipped in Windows 3.0 (in 1990); too bad it still is an issue.

Bill handed the presentation over to Hillel for a pretty good gee-whiz demonstration of the shell. Yep, Ian’s stock ticker was there, working on canned input, but actually changing during the presentation. Not a single mention from the presenters, though.

Hillel almost displayed one of my top-ten hopes for the PDC. During his presentation of thumbnailing the scrollbar, he showed a running movie in a flyout off of the scrollbar. Just one small step away from a movie on a button!

Jim promised to up the geek level, and showed some architecture block diagrams that would almost please ScottW. We also saw real code being written before our eyes. Jim was clearly working from a script, but just as clearly was making adjustments on the fly for ways in which the earlier coders had moved off of the script. Fairly impressive. Also, Financial apps were a “bullet-point” in possible clients for central contacts, though no scenario was spoken.

In what must be the highpoint of the PDC for me, though, the live coding demos exposed my favorite weakness in managed code. They were using Indigo interfaces to connect to a web service, and apparently ports are not managed. (Long-time readers will remember that my biggest bug-a-boo about managed code and the CLR is that implementers cannot build objects that manage their resources implicitly.) So Chris is proud that he remembered to wrap his Indigo C# code with the appropriate try/finally blocks. Already my smug rating has hit 7. But wait, there’s more. He neglects to protect a similar example written in VB (which I think they are using just to show off). I don’t even know if VB has the constructs to build safe clients in the case where the service objects are unsafe. As luck would have it, an unrelated bug that VB allowed crashed the application. Since it was a sidebar tile, it brought down Explorer as well, which is pretty spectacular, though Jim later claimed no crashes during the demos. When everything was started up again and the mis-matched parameters were fixed up, it still didn’t work, because the Indigo port was still open. It just doesn’t get any better than this; I’m at a 10+10 on the smug+snide meter. The developer renamed the port that the example was using and recompiled the application to go on with the demo; I don’t know what is actually necessary to close it down in this case.

Amazon showed a pretty good demo of exposing their existing web service to a Longhorn client. Didn’t mention local Wallet or transaction storage scenarios, but they could be their. Their business model is clear; they expose e-commerce.

An interesting note on the keynotes today: The Longhorn “periodic chart” seems to have disappeared.

Man, is this food blah. On the other hand, I don’t think I’ve ever been in such a large cafeteria before, so it’s probably hard to have food that isn’t totally overcooked. I think they were trying for oriental.

Too geeky for words. This presentation relied on showing IL and Assembly language to make its points.

Don Box appears to be a rock star at the PDC. Folks attended this session just because he was presenting. This turns out to be a pretty general presentation, though the meat of it (that services require a protocol-based rather than an object-based interface) is something to think about.

I walked back to the hotel, which took just about ½ hour. It turns out that this is the hotel that was the setting for the horse/motorcycle elevator chase in True Lies. They have a plaque commemorating the movie on the door to my elevator. I wonder if this was where Arnold rode the horse or where the terrorist was on the motorcycle. Anyway, I need to go out and find some dinner.

The network in my room is barely working; I can’t reliably use Frontpage to update this log or check my MSFT email. These things may have to wait until tomorrow. The connectivity was also strained at the PDC, but I have hope that things will be clear early in the morning.

Today’s general sessions are truly geeky but still pretty abstract. Eric’s speech hints at a number of areas that I want to know more about, without actually telling me enough. For example, he discusses that C++ will implement templates in the CLR that compare a little with C#’s generics. Also, apparently something called STL.Net is under development.

Super geek note. In a demo of a distributed, mobile app, they application does a real-time check for the devices location and accesses a weather service that shows that it is 73 degrees and fair in 90017. My reaction – “Wow, sounds like a nice day out there.” We’re sitting in a dark hall with 7,000 geeks hanging on every powerpoint slide; the sunshine seems as far away as Seattle.

It also occurs to me that there is only one valid preparation for trying to digest a presentation given under these circumstances: an education at a school with large, impersonal lecture halls. I won’t be in a hall with less than 200 “students” all day, and the material is new and demanding. I’m finally thankful for my state-school degree.

Yukon is going to be a big release for SQL. I’m sorry Karan, but that’s what I’ve got.

Actually, lunch for the laptop. These batteries barely last through the morning session, so I find an outlet and check on email from home and work. Some folks have seen this blog, which is cool. I don’t really need to head all the way over to the meal hall, since I can pick up a banana and pretzels from the snack tables.

On my way to the first afternoon breakout, I run by Eric Doerr. He seems to think that my taste in presentations is nerdy even for this crowd. We agree that the best presentation would be accompanied with alcohol, and pledge to find it.

This talk was to be given by Herb Sutter, and Brandon does a pretty good job of presenting somebody else’s deck. I’ve been doing a bit of that lately, and it is very difficult.

So once again a C++ rep seems to be saying all of the right things. They want to bring the best of .Net (well, all of .Net anyway) to C++, and the best of C++ to .Net. True generics, better memory management, and (tum-duh-duh-tah!) deterministic finalization will be supported. When? Around Beta 1 I think. Now, why can’t we find out about these developments back in the shop?

Highlights: We all had a smug chuckle about yesterday's crash and choke. They promise to provide the full range of expected C++ semantics (both manual destruction for "pointers" and scope-level destruction for stack objects). So smart pointers, control over copy semantics, etc. seem to be coming back to the CLR. You can also get compile-time type checking on templates involving CLR objects; generics are still run-time checked.

Coolest Hack: Some constructs are going to require marshaling between the C++ (unmanaged) heap and the CLR GC sandbox. They note, however, that the simple case of declaring a managed object on the stack doesn't really require an unmanaged proxy; the programmer is just asking for scope-lifetime semantics. So the compiler just assures us that Dispose() is called appropriately, and you have real resource safety in managed code. Unfortunately, this approach only works when the objects client is written in C++. The irony is that C++ appears to be the safest language for writing clients in the managed code world.

I skipped part 1, but this is about secure, reliable, and transacted services. Some of you gentle readers won’t consider “transacted” much of an adjective, but then you wouldn’t have enjoyed the presentation.

The recommended design is to have tight transactions within systems and to build compensation systems that recognize the inherent latency in distributed systems, but their system can be used to keep it tight. I haven't seen a good discussion of using Indigo between systems on the same box yet.

This may be unfair, but the audience is a schizophrenic as WinFS. Some folks just want to use the file system, and others are clearly interested in persistent relational objects. I continue to wonder how the WinFS team will reconcile this dual charter. Some customers are asking for SQL server on the box, and others are questioning if all of what is planned can be accomplished by Longhorn ship.

WinFS looks like one step short of what we’d like. For example, it will be missing synchronous notifications.

During the question period of the previous session I think I heard the collective sigh of those of us in camp 2. We are going to have to figure out how to hack up what we need, because what WinFS will be is very interesting.

Since my connectivity back at the hotel is suspect, I’m spending a bit of time here reading email, organizing my notes, and getting them posted. I’ll update the breakout notes during my next connection, but I need to go offline now to head on back for dinner.

It turns out that the PDC is supplying food and beer at the "Meet the Experts" evening session. A double score. I even meet a couple of experts, but they are looking for customers instead of Microsoft employees to work with.

It's just a little weird being a Microsoft attendee here. There are lots of us, but the convention is not for us. We don't have any duties in the booths or presentations,  but nobody quite treats us like attendees either. For example, everybody has a goody bag that includes some informational material, but Microsoft employees did not receive it. I haven't figured out if we're invited to the big party at Universal Studios tonight; I don't seem to have the invitation.

The attitude of the non-Microsoft attendees is interesting. These are not the slash-dot crowd; in fact we can laugh at them together. Though they are fans of Microsoft technologies (and they have made big bets on them), they don't exactly like or even respect us. Many presenters openly acknowledge that we don't always make these vendors happy. (Apparently, what makes them happy is promising that 3 lines of VB will allow them to securely log in an user to her personal information.) Folks applaud during the presentations, but fairly quietly, and there has been nothing like real cheer during the convention. In some cases (e.g. with WinFS), there is open skepticism about the goals and ability to execute all of them.

Getting yesterday's presentations up to date. Microsoft employees can see the powerpoint slides at \\russpj\public\PDC.

Rick shows some gee-whiz stuff with graphics using the GPU as more of a general-purpose processor, but still for the display. He says that the processor is great for detailed modeling of physics, textures, etc under general scene control from the CPU. The pixel-heads seemed to enjoy the live water demo.

Jim Gray (database rock star) showed the mega-astronomy database, SDSS. (see http://skyserver.sdss.org) This is not only cool, but a vehicle for real research and an example for research publishing. MicM should probably take a look. You can download the software and build your own site.

Eric George will be glad to know that social computing is alive and well. Lili Cheng showed some software that tracks your electronic conversations to determine who your contacts are and what groups they fall into. There is some neat display UI that will even tell you who your friends in common are.

But if you want to get applause from the PDC, don’t show social software: show automatic mathematics modeling. John SanGiovanni showed two pieces of software for Tablet PCs in the University classroom: a handwriting shell on top of mathlab that automatically graphs and animates your scrawled equations, and a physical modeling package that will animate your scrawled physical drawing. I think that I’ve seen a website demo of the second one, Magic Paper, with balls, inclined planes, and springs. Money Devs?

Schemas. Patrick may have called his box-diagram slides “mockitecture.” If he didn’t, I want credit for coining the term. Is a WinFS “folder” a good way to collect related things, like a separate, consistent relational database?

He later shows the whole WinFS schema in Visio to show what real architecture looks like.

He mentions extensions, subclasses, new types, but I still don't have a real feel for the decision matrix between those mechanisms.

I wandered through the Expo hall. I've never been a tools-oriented developer, and most of the booths aren't interesting to my. But they have long lines. I check out the UX booth in the Microsoft square to see how we look. The machines have the quote tile running in demo mode, and I check to see that internet connectivity is indeed pretty lame in the hall.

I chat for a bit with some of the MSFT employees manning the booth. They seem to be having a good time, and they are interested in how we see the show. I'm being asked if I'm learning stuff, if I'm getting what I need. I don't have a good answer yet; there's a lot of information, much of it is interesting and inspirational, but I haven't organized it well enough yet. Lots of customers seem to think that MSFT has solved some important problems, but I can't quite see how to inject these technologies into our products. At least one problem is that their catalog is not the same as mine. I'm interested in Indigo on single boxes, and they haven't really addressed that. We have a huge code base in C++/Win32, and even the interop and performance presentations haven't really charted out a migration path.

I think there may be 900 tables or so in this meal hall. Each table can seat 10 folks, so the whole convention could sit down at one time. As it happens, we don't, but I don't think that the staff is trying to re-use tables during the meal time anyway.

A whole lot of motherhood and apple pie. Seems like naive use of the runtime can have bad performance. Some easy things can be very expensive, and measurement will often point out hotspots for this reason.

Interesting GC note: non-trivial destructors cannot be run at GC time, even though they are triggered at this time. They actually cause the object to be retained for another round of garbage collection. Use Dispose().

The best and loudest advice is to actually measure the performance of your app. The CLRProfile tool (available from msdn) can show how you are interacting with the garbage collector, which can also be a source of slowness.

This session spells out some of the way you can build UI to work with WinFS objects. Again, the presentation is pretty file-centric, but at least they show contacts from time to time.

Nat makes a distinction between static item sets (which might be defined explicitly by relationships) and dynamic item sets (which are defined by queries on the properties of items).

Chris is having a bad demo day, but it isn't actually affecting the presentation. The presenters almost always show code, and most presentations show code being written, compiled and run. Even the keynote had (buggy and crashing) fresh code in it. It's an interesting stunt to see, and works in a couple of ways. It gives the presentation and presenter credibility. Real code using real tools doing real things, and those programmers must be quite something. During the keynote there was also a running gag on using legacy editors. Jim Allchin used vi, and emacs and SlickEdit were also exhibited. VisualStudio didn't make an appearance until VB was demoed; all of the other code (written with other editors) was crash-free.

But it isn't important to show off in this way for most presentations, and Chris' didn't suffer, even though it flustered him. He did a pretty good job of showing the development lifetime of an app in real-time. (One trick often used in VisualStudio is to have the goodies stored in the Clipboard Ring, so you can paste in the tricky parts.)

A second benefit of showing real coding is to gauge the customers' reactions. They really like single lines that do a lot, for example, and aren't so fond of editing raw XAML. These are the definitely the folks that .Net was written for. (Actually, at lunch yesterday, I heard a great conversation worried about handling all of the complixity of WinFX [the cool new name for the Longhorn/.Net API].) 

Jan used to work on the Money team, and I see the influence of his work on Money in his presentation. It is really cool to see those concepts we think work really well in our application being evangelized to the Windows ISV world. I wonder if we are keeping the loop closed with Jan, though.

He showed the abilities of the navigation-based app, which basically populates the non-client area of the app frame with working browser controls that work with built-in navigation commands. He also has a pretty good explanation of the different times when the navigation style is really attractive.

It turns out that I was invited to the party after all. I picked up my green wristband caught the bus to Hollywood. The bus was slow, but they provided refreshments for us. I was able to add value for my companions in the front of the bus with my time-tested ability to yell for food and drink to have it passed from the back. The ride was actually fairly quick, once we got on the freeway.

I had a good chance to talk to a bunch of customers tonight. It's a pretty good night to represent Microsoft; we're seen as the hosts (even though attendees where charged close to $2000 for the event), and the food, fun, and alcohol are flowing. A couple of themes may have emerged. Folks don't necessarily like us, trust us, or respect us, but they do like knowing our address. A few people have contrasted this world with the open source world, and these people like the fact that Microsoft is in control here. Perhaps they like knowing whom to blame.

I drank some beer, ate some (not too bad) Mexican food, and went on the tour. The tour is not usually done at night, and many portions were dark and empty. There was no chance of seeing a production in progress, and many of the outdoor sets are vacant and not lit. I have some dark pictures of stuff. The subway tunnel collapsed and the shark attacked us, but I think we missed almost everything else. Also, the ride spit on us. This seems to be a theme in theme rides today, because water is a fairly convenient way to actually touch somebody. The 3-D movie spit on us, the tour spit on us, and the log ride dinosaurs spit on us. Especially for glasses wearers at night, this is inconvenient. (But perhaps that's not a fair whine in every case. The log ride is supposed to get your wet, and everybody was wearing glasses in the 3-D movie). The Shrek movie was billed as 4-D; the extra dimension is actually the spitting, shaking, and puffing that you also get.

I've checked out of my hotel, and checked in my luggage at the PDC. Later today I need to see if I can change to an earlier flight.

The convention is winding down; many people have already left. We don't have a big general session this morning, and the breakout sessions seem to have scattered attendance. Again, I gravitate towards the geekier subjects. I did arrive too late for the introductions, but we seem to have some C# as well as CLR folks on the panel.

Someone asked the deterministic finalization question. Here are the conclusions. The CLR and C# designers don't exhibit an understanding of the solution that C++ affords; they think that the using() construct is a solution. They confuse possible problems in an object-based solution with insurmountable obstacles. They pose strawman options and burn them down. They may have good answers to the questions, but they don't speak my language. At least one problem is that as long as a solution leaves anything in the programmer's hands, they consider it incomplete, and they don't have my taste for which incomplete solutions are preferable.  (I think that Jim Miller and George Bosworth may be good contacts in this discussion, but they probably don't want to hear from me.)

There was a concession that scope-level destruction semantics were interesting, but I think they were talking about tuning the GC with that information.

WinFS, Avalon, ADO, and Outlook folks. It's interesting to wonder if we could learn a bit from Outlook; they are working more closely with Longhorn and are pretty close to our space. Unfortunately, I walked out on this one before I learned very much.

This is actually the kind of panel we could have been on. I wonder where the Indigo folks were, though.

The Indigo folks still aren't answering my questions. I suspect that the problem is that I haven't formulated the questions very distinctly yet. For example, I haven't really seen them factor out client-side and server-side services that they offer. They seem to be most focused on server-side work between web services.

So I'm really just showing off by attending the last session of the conference. The languages guys are fun to watch, anyway. Microsoft is represented by VB, C#, and C++. Also attending are two Cobol vendors, Eiffel, Fortran, and Delphi. On the Microsoft side, the VB and C# folks seem to talk a lot, even though the understand each other's products without wanting to commit to feature parity. The C++ folks don't share the language, and there is almost open hostility in some areas (Anders call templates "macro-like").

Some good questions about the difficulty of incorporating XML schemas into a type system (that perhaps the WinFS folks should be listening to), the fact that nobody except VB has really incorporated RDBS support, and who wants a standard anyway were asked. The answers ranged all over the board.

Most of the languages are "just" incorporating .NET features into their languages. The panelists generally agreed that their first priority was to their established constituencies (though the C# philosophy must be somewhat more aggressively evangelical). So perhaps the most interesting discussions were about the features that would be left out of languages and the agreement that feature parity was not necessarily a goal for anybody.

Standby. I'm hoping to catch a 7:20 flight instead of my scheduled 9:00 flight, but the terminal is full of geeks trying to get home.

My head's a little full right now, but I think I can make some sense of the PDC. First, the major reason that I felt a little disconnected at times is the shear amount of information being thrown at us, but it is also true that the PDC was aimed at an audience that I'm not really a part of.

PDC Home Page
Copyright © 2003 Russell S. Paul-Jones. All Rights Reserved.
Contact: admin@paul-jones.org