The other day, I set out to put some work into my guild's forum site. Immediately, though, I noticed that the main forum page was taking significantly longer to load on my development server than on my production server. Specifically, it was taking about five minutes to load, as opposed to about a second. That seemed... problematic. Looking at the profiler, I saw that the page was somehow making 38 million calls to Document.UniversalID, which was slightly out of proportion to the two forums listed on the page.

After a lot of trial and error to find out what the problem was, I narrowed down the culprit to the list of recently-updated topics. Each of my model elements is represented by a class implementing the List interface, which is really like a wrapper around the view. Thanks to some debugging output on the console (with apologies to my bloating log), I found that the list, which is only supposed to get the first 15 entries in the view, was repeatedly processing EVERY entry. Spitting out a stack trace revealed that the XSP list processor was converting the entire List object to a string via its own method of using the List's iterator. Huh.

I investigated a bit, looking at the resultant Java code from my custom control, renaming the property that passes in the List, and looking at the name.xsp-config file that accompanies it, all to no avail. Next, I tried re-creating the custom control, but that has the same behavior. For the time being, I've found that copying the code from the control in place of where I use it in the XPage works for some reason.

I have no idea why doing that worked, since in theory it should be performing the same thing, and I have no idea why all my other uses of similar controls--such as the one for recent posts--aren't freaking out. This will certainly bear some investigation, since until I figure it out I'll have to live in fear of the runtime stringify-ing my giant Lists at will.

I was listening to the latest episode of Build and Analyze on the way home from work yesterday and, as I am wont to do, I started yelling at my iPhone when they started talking about Google Reader and the difficulty of syncing. Admittedly, at the end, they got to the fact that, even if you could do it technically, it'd be tough to make money off of providing an RSS sync server. That part is fair enough, but I still can't let the technical difficulties stand, and I've been thinking more about how it would be done in Domino.

In the basic form, the problem in question is pretty much exactly what NSF and the Notes/Domino relationship is designed to do: seamless replication, deletion stubs, unread marks, and so forth. In fact, RSS syncing is a better fit for the model than mail, since mail required adding all kinds of extra (but useful) functionality, while RSS syncing would just be data and an agent to fetch the feeds periodically.

The way I figure it, there would only be a couple technical hurdles, both related to scaling: storing large volumes of data and fetching new feed content periodically.

Storing large volumes of data might not be too bad. There are a couple ways you could do it. One would be to store the user's list of subscribed feeds and "read" stubs in one database per user, and then store the feeds and feed content in another database (or databases), and do all data access via agents or web services that would pull the data from each distinct location. Another way could be to store the feeds and entries in each user's database, keeping the feed content as attached HTML documents and letting DAOS handle efficient storage. The latter route would let you take advantage of the Domino Data Service and read marks (which DAS conveniently supports).

Updating the feeds would be rough for a single server, but the job could be farmed out to many clusters in a server. You could write agents that would determine the server they're on and, based on, say, its name, pick a slice of the feeds to check, so if you have 10 servers, each would update 10% of the feeds.

I'm sure there'd be other roadblocks during actual implementation (it IS Domino, after all), but I think that'd be basically all you'd need on the server side. The client would be a little tougher, since you couldn't just use NSF and selective replication, but that wouldn't be terribly difficult to handle.

It's too bad it's likely not profitable, between licensing, hosting, and bandwidth costs - it'd be a fun project to try out.

A while back, I described the problem I'm having in my guild-forums XPages app, which is that it very easily gets its environment out of whack, to the point where changing any design or data note from outside the XPages environment caused an "X is incompatible with X" ClastCastException. This improved gradually over time. At some point in 8.5.2, I started being able to modify data documents again without the problem. When 8.5.3 came out, it improved again: I can now replicate over changes from my development server to the production one without having to bounce HTTP on the production one. At that point, it became much less of a hassle - sure, I still had to re-save a Java class file every time I modified an XPage, but that was only in development, so I could deal.

However, I think I've found the proper solution. In 8.5.3 (I think), IBM added a custom editor for the xsp.properties file (which you can reach using the "Package Explorer" Eclipse view, in whatever.nsf/WebContent/WEB-INF). Many of the options are duplicates of what you see in the normal "Application Properties" editor (since they edit the same properties), but there are some nifty additional goodies. I won't go into them all, but I urge you to take a look. The important one here is in the "Timeouts" section of the "General" tab: "Refresh entire application when design changes" . That sounded perfect and, lo and behold, it seems to do what I want: once I turned that on, re-saving an XPage stopped causing the ClassCastException. I haven't given it a proper testing, so it might not be everything I'm looking for, but I'm thrilled at the initial results. Having to re-save a Java file for each change wasn't a HUGE problem, but it was a niggling one, and not having to jump through the hoop helps make the whole environment seem less rickety.

Yesterday, I started working on a small sidebar widget app using an XPage, after finding out that XPages can now (as of 8.5.3) be used in the Notes sidebar in the same way that Forms could before. It's quite a simple page, very Twitter-like: one text input field and then a list of posts. However, even though it's very simple, I ran into two annoying bugs quickly.

The first of them is a Schrödinbug. I set up some code in the "onClientLoad" event to start a setInterval to do a partialRefreshGet on the list of posts, so it would keep itself up to date:

<xp:eventHandler event="onClientLoad" submit="false"> <xp:this.script><![CDATA[ setInterval(function() { XSP.partialRefreshGet("#{id:messagesPanel}") }, 2000) ]]></xp:this.script> </xp:eventHandler>

Simple, right? But I started getting errors in the status bar, along the lines of somethingorother not being defined. That's... odd. So I swapped over to Safari, figuring it'd be some XPiNC-specific thing, but I started getting similar (albeit more specific) errors there, along the lines of:

TypeError: 'undefined' is not an object (evaluating '_166.formId')

Huh. So I tracked it down and the problem was in the code called by partialRefreshGet, where "_166" is the name given to the second (optional) parameter. It makes sense when seeing it - though the code is supposed to fail over when it doesn't work, I can see why the browser would throw up its hands when you try to get a property of an un-provided parameter. But this has worked before! Just looking around on the web, you run into plenty of examples that leave out the second parameter. But sure enough, when I changed the line to XSP.partialRefreshGet("#{id:messagesPanel}", {}), it started working great. I can only think that this is either some 8.5.3-specific bug or some weird thing I managed to do in my code... but there's not even really enough code to mess up.

The second bug is still annoying, and it's somewhere in between a bug and a security feature of Firefox/Gecko. Basically, while you can call the .click() method on a button in client-side JavaScript in Gecko, it doesn't behave exactly like clicking on the button. Specifically, I have a CSS-hidden button that executes the actual action of creating the message document from the text you type in, and I want it to happen when you hit enter. The button itself works great - if I have it show up, I can type, click, and it executes the action and partial-refreshes the list of posts. However, if I hit enter, which uses the .click() method, I get an error about not being able to refresh that part of the page, but then it forces a full-page refresh and still works. So it's KIND OF clicking it, but not quite.

I'm not sure what to do about this one. I looked up a couple things online about trying to emulate the actual click event, but with no better results. I could do a REST service on the page, but I don't want to have to roll out the Extension Library to everyone in the company. Maybe I'll look into sending along the data in a XSP.partialRefreshPost call and eliminate the button entirely. We shall see.

Though I've grown to more or less enjoy writing Domino applications, I always feel like this is in spite of the tools, namely Designer. In a lot of ways, Designer has improved significantly over the last couple versions: as long as you ignore the speed, the Eclipse-ified Java and LotusScript editors are miles ahead of the antiquated previous ones, and it's handy to be able to switch to the Java perspective. However, so much else makes it a drag:

• It's a Windows app. I use a Mac, so there's simply a big hurdle to using Designer. Parallels has smoothed the process greatly - running Notes in Coherence mode means I can use my preferred OS while still getting work done. However, it means that it's a big to-do whenever I want to make any tiny change in the code. For day-to-day work stuff, I can get a lot done in the Mac Notes client, since I put in a lot of work to make managing client web sites doable without going to the design side, and the Mac client still lets you edit views and agents. However, there's no XPages editor, so I can't use that for serious work.
• It has a mind of its own. For some reason, a new clean install of Designer I made the other day got it into its head that opening any database should involve recompiling every XPage and Java class. Sometimes, it turns off "Build Automatically" for no reason, and then turning that back on also requires a full recompilation. Sometimes, it just holds up all user actions while it does... something for five minutes. What's it doing? Beats me.
• How many times have I seen this window? A billion?
  Removing a database from the project list has about a 30% chance of causing a crash and quitting Designer has about a 50% chance. Sometimes, saving a form will do it. Sometimes, walking away from the computer to get a drink is enough. And every time it crashes, I have to dismiss the dialogs and check the task manager to get rid of any residual processes, such as an instance of nsd pegging a processor, then relaunch Notes and get back to the environment I had, which is not a speedy process.
• I'm not that crazy about Eclipse. DDE is better than previous Designers, yes, but Eclipse is still a giant beast with the same sense of style and simplicity as the monstrous Java language that spawned it. On the plus side, the blue look that IBM came up with is actually rather attractive compared to the standard Eclipse UI, but little quality-of-life things are a drag. For example, how do you specify how many spaces a tab should take up visually? It's in a couple places in the preferences and you have to set them all, including buried inside a weird sub-preferences dialog for messing with your Java formatter. And how about changing your code syntax coloring? You have to go to a dozen places, one for each syntax type. Compare to a text editor like TextMate, where you can swap between packaged color schemes with a drop-down.

That's enough for the rant. So what is there to do about it? I give this problem a thought from time to time, and I don't think there's really a great option, but there are some places to start. The real key to any alternative scheme is DXL - using that, you can (more or less) view and modify design elements freely. I've toyed with this notion before - maybe a web UI that lets me pick the DB and design element I want so I can tweak the DXL manually for when I want to make a quick change but don't already have Windows or Designer open. It would mostly work, but it would take a lot of work to make it practical.

There's a big sticking point, too: XPages. If you export an XPage to DXL, you can see that the exporter basically punts on it - it's exported as a generic "note" type with Base64-encoded binary fields. I haven't run the field data through a decoder, so maybe one of them is the XML "source" of the page, and maybe it could be made to work, but that just raises more questions. Would that require updating the other binary fields in some way? Would importing it with a DXL importer cause it to generate and compile the Java representations? What about generic Java classes? Would those work with DXL and would they be compiled?

I'm starting to get an idea of what would be ideal and almost practical, though. One could write a WebDAV server (possibly with a complex servlet, a small web server to the side, or other trickery) that represents the design elements as editable files in a folder structure similar to the Java perspective view of the database. Traditional files and image resources could be edited as-is (which I think the built-in WebDAV server does), but design elements could be represented just as DXL and then re-imported when modified. Even if it doesn't support XPages, such a scheme might have a lot of promise and wouldn't be reliant on Designer or any other IDE.

If I ever get brave enough to delve into WebDAV or frustrated enough with Designer, I might just look into it myself.

For the most part, writing agents in LotusScript is the best way to go (at least when it can't be done in formula language), mostly for smoothness of interaction with the built-in libraries (no .recycle()) and because it's less prone to running into memory problems when other agents go wonky than Java agents are. That doesn't mean I have to like it, though.

If there's one thing that drives me nuts about LotusScript more than any other aspect, it's its handling of arrays. This came to the fore with one of my recent projects, which involves spitting out the contents of a view, which in turn entails lots of use of entry.ColumnValues. My first, quick-and-dirty draft ran into an early performance issue, which is that every call of .ColumnValues on a NotesViewEntry seems to be as expensive as the first, meaning that the class doesn't do any internal cacheing and has to re-fetch it every time. Ugh, fine - I'll just assign the value to a new variable at the start:

Dim colValues as Variant colValues = entry.ColumnValues

Not too bad - two extra lines at the start of a loop is a small price to pay for a significant speed improvement. Unfortunately, it doesn't work - it throws a Type Mismatch error at runtime. After doing a TypeName() on entry.ColumnValues, I saw that it considers it "VARIANT( )", an array of Variants, which makes sense. It's a bit weird, since I've stored arrays in Variants before, but whatever - with a quick code adjustment, we're off to the races:

Dim colValues() as Variant colValues = entry.ColumnValues

Great! Now hit Ctrl-S and... compiler error. You're not allowed to assign to the entirety of an array like that. Argh! So I guess I'm going to have to make my new array manually and loop through the original to copy each entry over individually. If you're familiar with another scripting language, that probably sounds like a simple task, but LotusScript's array annoyances continue. Because this is like BASIC, you can't just do "colValues.add(something)" - you have to ReDim the array to the right size. Here's the code I ended up with:

Dim columnValues() As Variant ReDim columnValues(-1 To -1) ForAll columnValue In entry.ColumnValues If UBound(columnValues) = -1 Then ReDim columnValues(0 To 0) Else ReDim Preserve columnValues(0 To UBound(columnValues)+1) End If columnValues(UBound(columnValues)) = columnValue End ForAll

Before you look at that "-1 to -1" crap and deem me insane, hear me out. Though the NotesViewEntry doesn't cache its property value, the ForAll loop does, meaning that, according to the profiler, ColumnValues is only called once for each view entry, which is about as efficient as it gets. All that extra crap about ReDim'ing the array over and over instead of just once is essentially "free" compared to the expense of the product object call, so it ends up being completely worth it.

So there's a new round of talk lately about syncing and the trouble involved, thanks to some changes in Google Reader's behavior and the desire to find a new safe haven for RSS syncing. The best example is, unsurprisingly, from Brent Simmons:

Google Reader and Mac/iOS RSS readers that sync

However, the whole time I was reading this article, my brain kept yelling at me, louder and louder as time passed:

This is Lotus Notes! The system you're describing is Lotus Notes! It does syncing and deletion stubs and read marks! IT'S LOTUS NOTES!

This kind of thing would indeed be really easy in Notes/Domino, particularly if you were actually using the Notes client (though it wouldn't be much to look at). Subsets of data, managing deleted elements, timed refreshes from the source, storing each feed entry as its own entity, and offline access that can have its changed synced back to the master are all things that Notes has done since its conception - the only problem is that it's so ugly and arcane that mass-market appeal is nigh-impossible.

Nonetheless, it got me thinking about the viability of using Domino as a syncing server for this. You wouldn't be able to use NSF files in your RSS clients, which would make the job a bit tougher, but the new "XWork" licensing model would fit into this nicely. Scalability would be a serious concern, but the simple nature of the data would keep view updates quick, and it'd just be a bit of cleverness in the database layout to direct users to the correct place. Toss a couple clustered servers in there and you should have some good load balancing, too. The Domino Data Services API might be enough to handle data access from the client, but, if it's not, a couple simple agents would do it.

I'm sort of tempted to try hashing something out.

Though I don't have a use for it currently, I can't help but get kind of excited about the Domino Data Services in 8.5.3 and the Extension Library. If you're writing a normal Domino application - using either legacy elements or XPages - you probably won't have terribly much use for it.

However, the really cool aspect of it is that it significantly smooths the process of using Domino as a backing data store for another front end written in PHP, Ruby, or anything else. This has always been sort of possible - you could use the Java API or a combination of ?ReadViewEntries, ?CreateDocument, and ?SaveDocument URL commands to access Domino data without actually being in Domino, but it wasn't exactly a smooth process. With the Data Services, now Domino is very similar to, say, CouchDB, but with reader fields and impenetrable licensing terms for non-vendors.

One nice little side effect of the fact that it uses the HTTP stack is that DSAPI modules work. When I was testing around, I was able to get a list of available forums as Anonymous, resulting in only the two visible ones. When I included my user authentication filter cookie, it started showing me the rest of the forums that the user could access, exactly like you'd want. While you could presumably just pass the username and password in each request using normal HTTP authentication, it's pretty cool that any alternate methods like this work as well.

It's all pretty exciting, and I'm itching to find a use for it.

Since 8.5.3 has been out for about 24 hours now, I naturally rolled it out on both my development and production servers. Fortunately, my irresponsibility was greatly rewarded: the largest problems I've had so far were a change in the way Java classes are accessed in JavaScript (I could no longer just call methods on non-public classes defined in the same file as a public one, so I had to split them out into their own files... which is what you're supposed to do anyway) and a minor CSS change where the top borders of my Dojo tabbed tables are now back to a light grey color, so I need to find the new CSS rule to change them back to brown.

I'm rather happy so far about two minor things in particular: CSS/JavaScript aggregation and OSGi auto-loading.

The CSS/JavaScript aggregation is almost a freebie: once you have Designer 8.5.3, you get a new option in the database properties sheet to turn this on and then 8.5.3 servers will happily obey it. I immediately noticed a decent load-speed increase of about 1/3 and one non-technical guild member said that the odd problem of dog-slowness that they (and not other people) had has been fixed. My favorite aspect of this is that it's a smart feature: due to the way you define Dojo modules in an XPage as <xp:dojoModule/> elements and not just text on a page like normal HTML, Domino knows ahead of time what you're using and can thus feel free to optimize it in transparent ways like this. It feels good seeing the same code go from one form to a more efficient one just by virtue of done the "right thing" when writing it to begin with.

The OSGi plugin auto-loading was mentioned briefly on Dec's Dom Blog back in June and I hadn't seen much reference to it since, so I was afraid it wouldn't necessarily make it in. Fortunately, it has: I created a new Update Site with the template from an 8.5.3 server, imported the latest Extension Library, ran the "Sign All" agent, and set the notes.ini parameter to look there. And lo and behold: it properly loads up the extensions from the NSF, so I was finally able to delete the filesystem versions that were previously necessary. This makes managing the Extension Library much smoother and it's one less potential gotcha when I upgrade my dev server first and then want to deploy it - since the Update Site has a replica on both servers, the upgrade is handled with the normal replication process and I don't have to remember to copy any files over from server to server. And the fact that it's an NSF theoretically gives you all kinds of other, more complicated deployment options, like server-based Reader field control or partial replication to control which servers see which plugins if you're so inclined. Very cool - I approve, IBM.

I don't have access to the beta versions of new Notes/Domino versions, so I haven't been able to tinker around with all the cool new things that are slated to appear in 8.5.3, but that hasn't stopped me from getting pretty excited about some of them. The big-ticket items are clear: the new Domino Data Services and relational database access (through the Extension Library, which may as well be standard) could make practical very different ways of using Domino either as a standalone data source with a different front end or as a standalone front end with a relational data source. While both types of setups are theoretically possible now, they're such a hassle that they're not worth it - but, with 8.5.3, they're almost top-tier choices for system architectures.

However, since I don't plan to rewrite my entire architecture, those new features probably won't affect my day-to-day life for a while yet. What has me most excited in a practical sense is much more lowly: being able to do a full-text search with sorted results. I've found that one of the big bottlenecks in my guild-forums app is the sheer size of the views, particularly the Posts one. I used to stuff pretty much all of the summary data into the view, but then I found that removing the non-sorted columns sped up responsiveness dramatically. That whetted my appetite for clearing out unneeded sorted columns - since each sorted column contains a full view index, having even a handful can increase the total index size dramatically. Since it appears that FTSearch's performance is almost (but not quite) as good as getting all entries by a key, I'll be able to remove the rarely-used sorted columns, speeding up all the common operations in exchange for a very minor hit in the rare case. Plus, it'll just feel good to put Domino's searching capabilities to proper use.