Category: Software

It’s been excessively frustrating and tiring. To cap off the season I was hoping to release the overdue 1.6.8 today, but there are still difficulties in getting the new defaults system working correctly.

The debugger is failing in the middle of an initializer method (self becomes 0×1, and the argument becomes invalid), which typically indicates an invalid pointer, but I’m just using -[NSCharacterScanner scanInt:], which shouldn’t be so problematic. (But it is, after all, Cocoa.) And it turns out that isn’t the correct tactic to use anyway as I want to extract info from the internal version number (MMmmb.bs.rr), not the public (MM.mm.bb[s[rr]]), and Cocoa doesn’t have an API for handling fixed-length records. (Possibly the only functionality from COBOL I actually miss! I mean, there are BCD libraries for C, right?)

So I’ve spent much of the day tussling through docs and links for various RegEx frameworks for Cocoa, and trying to add RegexKit to the Blocks plugin framework; first 0.5beta and then 0.4beta via compilation. Now I’m attempting to use the prebuilt binary, which is problematic because frameworks are normally used by applications, not plugins, so it doesn’t exactly fit right (or run). But as the clock’s been ticking, it looks like it’s time to punt and attempt another tactic by hard-coding a parser for the internal version scheme or using defaults for the optional fields in the public version scheme.

This is just to fix the bugs in the version class I added in an earlier version, which I have to get out of the way to continue fleshing out the revamped defaults system I created with unit tests and component code in order to release! Argh!

It hasn’t been all sweat and tears, though. I went through a period of idea generation (both software and marketing/entrepreneurism) early during the process and also, last night as I was organizing several years of accumulated papers, bills and what-not, found several new UI ideas I had come up with (and forgotten) and the missing FreeHand install CDs.

Now to get these final PITA problems licked so I can continue to move Mori forward!

While trying to solve a user’s problem with an mGTD script, I came across a subtle issue that demonstrates some issues that arise when violating a programming philosophy, tackling bugs in other people’s code, and general uncertainty whenever coding in AppleScript.

Working with AppleScript is generally considered iffy, because a lot seems ambiguous and so much is dependent on how the dialect is interpreted and how scriptable apps handle some of the application events which scripting is dependent on. I’ve written scripts before, some I’m pretty awed by (that it works, actually, but also what it does), but I’m still hesitant to tackle some scripting issues. In addition, being a GTD greenhorn, and an mGTD noob made trying to respond to this issue authoritatively very questionable.

Thankfully, BMEGuy, mGTD’s author and all-around community nice guy, tackled the question with a quick solution. But the updated script was still problematic, and so I felt I really needed to participate in coming up with a solution.

Again, being an mGTD noob and all, it took me at least half an hour to figure out how the plugin worked, and the script on top of that. Then, after I was able to get the script to run, it worked for me. Hmm.

But that’s because I was testing with an entry with a date due of today. Once I switched it to later in the week, the entry was still showing up for today. Isn’t that odd? It seemed I had inadvertently left in the date line from the original script. When I removed it, I witnessed the same problem.

It turns out there’s a bug in MOX 10.4.11’s iCal 2.0.5 (I’m guessing it’s present in earlier versions as well) where it doesn’t properly update the calendar display for new events made by the script. You won’t see it in the monthly view. However, you might notice a little oddness in the weekly view.

You can see the event if you add ’show theEvent’ after the script makes a new display alarm for the event (between the 2nd and 3rd ‘end tell’ up from the bottom). This will display it’s properties in the info drawer, but you won’t see the event anywhere on the calendar (in either week or month view) until iCal is restarted.

Running the script in monthly view doesn’t show any artifact in the calendar, but the data is shown in the info drawer.

You could also run the script in the weekly view and then switch to the monthly view, in which case you get this:

So now that the question of the event’s presence in the calendar was settled in my mind, I had to figure out why my faulty script displayed the event, but not the proper one; and how to coax iCal to display it.

Being unfamiliar with mGTD still, I tried to figure out the difference between the attribute name “dateDue” and due date. due date is one of the standard properties for entries in a Mori document. attribute name “dateDue” is a user column added in the example mGTD notebook. You can view them all the user columns by selecting the menu item Edit > Edit Notebook Columns…

Okay, good so far, but why would one cause iCal to display properly and not the other? After moving the due date line about for a while, I checked Script Editor’s Event Log, and saw

The event reply for the due date had a missing value! Mori wasn’t returning a value for the due date property because it wasn’t set (and wouldn’t be in the example notebook). Now I had to find a way to use one of those missing values to make theEvent visible without setting it to the wrong date. And the problem with that is most of the properties used in Mori’s entries aren’t appropriate for an iCal event.

I eventually thought about re-ordering the messages to iCal instead of being so fixated on a change in the messages to Mori or playing with the properties being set in creating the event. What I came up with was a plan to use the messed up missing value date as before to make the event visible first, and then set the date correctly. The code turned out like this:

tell application "Mori"
	tell current entry
		set theDate to (get attribute name "dateDue")
		set faultyDate to due date
		set theName to name
		set theNote to note
	end tell
end tell
tell application "iCal"
	tell calendar "Scramble" -- the user should specify the name of the target calendar here
		set theEvent to make new event at end with properties {description:theNote, summary:theName, start date:faultyDate, allday event:true}
		tell theEvent
			make new display alarm at end with properties {trigger date:theDate}
		end tell
		-- show theEvent
		set theEvent's start date to theDate
	end tell
end tell

And to my surprise, it worked! So as I began gathering the materials together for my reply to the issue, I noticed something in the event’s info drawer that had escaped my attention before:

iCal, that’s just crazy talk! But at least it would explain why it would display traces of an event, if anything at all; and why it wasn’t noticeable earlier: iCal would correct the event data when reading it in when it started (”iCal database, that’s just crazy talk!”). But somebody forgot to add a sanity check when creating a new event from the properties passed to it by our script. (This is an example of why the Once and Only Once principle should be heeded. If there’s only one place where events are synthesized from pre-recorded values, whether those values are from a stored file, a script or the UI, then all those code paths will benefit from any sanity checks added to event creation.)

Knowing this, here’s another means of working around this bug, by sending iCal info that won’t confuse it:

tell application "Mori"
	tell current entry
		set theDate to (get attribute name "dateDue")
		-- set faultyDate to due date
		set theName to name
		set theNote to note
	end tell
end tell
tell application "iCal"
	tell calendar "Scramble" -- the user should specify the name of the target calendar here
		set theEvent to make new event at end with properties {description:theNote, summary:theName, start date:theDate, end date:(theDate + 1), allday event:true}
		tell theEvent
			make new display alarm at end with properties {trigger date:theDate}
		end tell
		-- show theEvent
		-- set theEvent's start date to theDate
	end tell
end tell

Thinking about these two solutions it’s clear that picking the latter one, with well-formed properties, is the safest choice to make. Here’s additional proof: the first solution, the one which plays with the start date to make the event appear, will indeed make the event appear. But if there’s less than 24 hours until the event begins, it will appear on the wrong date and still require iCal to be restarted to appear in the proper location!

It just goes to show you, while you might be able to get away with just the barest minimum, and someone else might normally clean up after you, it’s best if you did the job correctly from the start in case your safety net disappears from under you.

When I purchased Mori, one of the first things I did was generate some documents about its codebase. For this, the main tool I used was Xcode’s Class Model tool to generate charts of the various classes involved. I spent several days laying out the classes on the charts, then printing and then folding and taping the pages together. (This is a process which I am replacing with specific related classes that occupy only one or two pages, so I can keep them in a notebook; or sending a PDF to Kinko’s the next time I need to print the whole chart.) They helped me get an understanding of how code was laid out, and their relationships.

After a while, the tape that held the charts up on the wall would lose their adhesiveness and down they’d come…again and again, eventually falling into disuse. Having gotten a digital camera, I spent a day reassembling the crumpled remains of the charts about a week or so ago, and snapped a few shots which I present here.

There are 33 classes, and 25 protocols defined for the document back-end plugin, and quite a few supplementary category methods extending Cocoa’s foundation classes.

The user interface plugin has 85 classes, 18 protocols, and its own quantity of supplementary category methods.

At least, that’s what Mori’s classes looked like when the shots were taken. Today it looks a tad different.

Oneill, the branch for Mori 1.7 is under active development again and will bring substantial changes to the UI and back-end architecture. You shouldn’t notice any hiccups in the file format, but you should see better functionality and performance.

At least that’s what the testing support should help me do. Thankfully, I make heavy use of the Subversion SCM system. Just in case.